Brainstorm A.I 2022

Andrew Ng, Founder and CEO, Landing A.I. 
Interviewer: Jeremy Kahn, FORTUNE

Brainstorm A.I. 2022 - Big Data Vs. Good Data

Conferences

Andrew Ng, Founder and CEO, Landing A.I. 
Interviewer: Jeremy Kahn, FORTUNE


Brainstorm A.I. 2022 - How Much Data Do We Need?

Brainstorm A.I. 2022 - How Much Data Do We Nee

Pieter Abeel, Co-founder, Covariant
Amit Bendov, Co-founder and CEO, Gong
H. James Wilson, Global Managing Director of Thought Leadership & Technology Research, Accenture and Author, Radically Human: How New Technology Is Transforming Business and Shaping Our Future 
Moderator: Rana el Kaliouby, Smart Eye

Brainstorm A.I. 2022 - Robots Around The Water Cooler

Pieter Abeel, Covariant
Amit Bendov, Gong
H. James Wilson, How New Technology Is Transforming Business and Shaping Our Future 
Rana el Kaliouby, Smart Eye

Robots Around The Water Cooler


Pieter Abeel, Covariant
Amit Bendov, Gong
H. James Wilson, Accenture and Author, Radically Human: How New Technology Is Transforming Business and Shaping Our Future 
Moderator: Rana el Kaliouby, Smart Eye

Pieter Abeel, Covariant
Amit Bendov, Gong
H. James Wilson, How New Technology Is Transforming Business and Shaping Our Future 
Moderator: Rana el Kaliouby, Smart Eye

Brainstorm A.I. 2022 -  Robots Around The Water Cooler


Fei-Fei Li, Co-director, Stanford Institute for Human-Centered AI; Professor of Computer Science, Stanford University
Interviewer: Jeremy Kahn, Senior Writer, FORTUNE; Co-chair, Fortune Brainstorm A.I.

Brainstorm A.I. 2022 - Building The Future We Want: A.I’s Human Factor

Fei-Fei Li, Co-director, Stanford Institute for Human-Centered AI; Professor of Computer Science, Stanford University
Jeremy Kahn, Senior Writer, FORTUNE

Building The Future We Want: A.I’s Human Factor

Fei-Fei Li, Co-director, Stanford Institute for Human-Centered AI; Professor of Computer Science, Stanford University
Jeremy Kahn, Senior Writer, FORTUNE; Co-chair, Fortune Brainstorm A.I.

Hi, everybody. Welcome to Fortune Brainstorm A I 2022. Um I'm Mihara, I'm editor at large at Fortune and co-chair of the A I Summit. Um It is great to be with all of you here in San Francisco in my own backyard. Um especially during this really exciting period in the development of A I, what could be better than intelligent human beings such as yourselves, gathering to share insights and expertise on the simulation of human intelligence, right? Artificial intelligence is transforming business around the world as we all know, often working quietly behind the scenes, but make no mistake about its growing impact, which some say could be as great as electricity or maybe even more. Who knows over the course of the next few days, we're going to separate AI hype from reality. While AI adoption is on the rise, there have been some headwinds as we know talent shortages, the inability of companies to share data efficiently and even a lack of data. Above all the industry is realizing that for A I to reach its full potential, it will need to solve growing ethical and safety concerns. I'm sure that'll come up repeatedly over these conversations. Even with these issues AI is already having some really impressive impact and delivering on some of its promises from speed speeding up new drug discoveries to increasing the efficiency of cloud computing. There are still so many outstanding questions. What does the expanding adoption of A I mean for business and for employees, how are firms changing, how they build and share values? Should we have an A I bill of rights? What does governance look like? We're going to discuss all of this and more. We're going to talk to some of A I's top leaders including Andrew Ang Kevin, Scott Reed Hoffman, as well as get an inside look at the A I Strategies of Apple, Microsoft and Google. And because seeing is sometimes believing we'll also enjoy demos from open A I and Stability A I. So get ready for those. Uh These conversations are going to be led by members of Fortunes editorial team and by my fellow brainstorm A I coaches who I'd like to take just a quick moment to recognize Fortune, senior writer, Jeremy Kahn, fortune, senior editor, Vern Coo and deputy CEO of smart eye doctor Rana El Kabi. And now is a great time to also thank our fantastic event partners for their work in bringing us all together here today. Our presenting partner, Accenture and our partner C three A I and Intel. We greatly appreciate your work in making these conversations come to life. And now before we tick off just a few light housekeeping reminders, what said here on stage and then breakout sessions is going to be on the record. Um What's said over the dinner table on a break on a walk between two people is off the record also at some of our private lunches unless mutually agreed otherwise. Ok. And with that, all those rules, let's get started as the pace of A I innovation and breakthroughs accelerates. So has the conviction that its application should be safe, ethical and inclusive in our theme setting conversation. An Ai Luminary makes the case for a Human Centered approach, a future where people and machines are collaborators, not competitive. We'll ask Dr Fei Fei lie about her work with ambient intelligence in health care and robots and explore the crossover of visual and artificial intelligence that enable computers to see what us intelligent humans see as well as what they don't see. Please welcome to the stage, Dr Fay Fay, co director of the Stanford Institute for Human Centered and A I and professor of Computer Science at Stanford University for a conversation with Fortune brings from A I CO chair, Jeremy Kahn. Great. And thank you all for being here. Um We are going to try to take some questions from the audience so please have your questions ready. There's a couple of people with mics, they'll come around and if you want to ask a question, you can raise your hand, they'll get to you. Um So I want to start out by asking a lot of people are saying A I is now reaching its industrialization phase. The first question I wanted to ask you is just, do you agree with that? Do you think that's sort of where we are in the development of this technology right now? Yeah. Well, first of all, Jeremy and uh everybody, thank you for inviting me here. I absolutely do agree with it. I think A I is a very nice field of all sciences and technology is one of the younger field. We are a little more than half a century old. Uh But um if you look at the history of A I and look at where A I is now, I think this technology has absolutely reached an inflection point where um it's being used, um you know, worldwide in a daily basis in many applications and its potential is just still gonna be on the rise and quite explosive. I guess if that's the case, if we are sort of in the industrialization phase of this technology in the past, industrialization has often kind of had a connotation with, with some dehumanization, you know, things machines have replaced people or people have been made to act in a more mechanized or machine like way. How can we avoid this kind of dehumanization with industrialized ai how can we make sure that people remain at the center of what we're trying to do with this technology. Yeah, thank you Jeremy for asking this. This has been a question that my colleagues and I at Stanford have been asking for the past three plus years. It's the core question to this new wave of technological transformation. So first of all, let's just recognize this throughout the entire history of human civilization. Tools are always double edged sword. Every time we invent and innovate a new tool from fire to steam to electricity to biotech. We recognize that there is the pros and cons of the tools. But at the very heart of tool, innovation and tool usage are the humans. Humans are the inventors. Humans are the other users. Humans are at the receiving end as well as at the innovation end of this this technology. So at Stafford's Human Center A I Institute, we feel very privileged and lucky to be in a place where uh we are inventors of this technology. Stafford has a proud history of being at the front. Um A frontier of uh many of the A I technologies that we're we're seeing today. But we are also we get to work with multidisciplinary thinkers and researchers and students to ask this very fundamental question of what it means to be human centered. And just in short, I think it's really important that we recognize um the the impact of A I, the Human Center impact of A I is at individual level at community level as well as at societal level. So in every part of our design, from the problem that we're trying to use A I to solve, to the data we curate to the algorithm we write to the system we build and to the to the to the applications we deploy that entire pipeline, we think hard about its individual impact, community impact and societal impact, right? And I know H A I is very multidisciplinary in the way it's been set up. Do you think there's a sort of lesson there for how companies should also think about building the teams that are creating A I technology? Absolutely, I I think our whole society right now at large is having this conversation of stakeholder economy in addition to shareholder economy. And I think that's part of the same conversation because uh um in this process of creating and deploying A I, it will impact all of our lives. And it is really important that we include the, the stakeholders of, of the, of this technology in the design process. My colleagues at Stanford, some of my esteemed, you know, ethicists and political philosophers always reminds me who is a technologist that let's design ethics in the get go. Not, not think about it as the aftermath. There are a lot of human issues that we can be mindful of in the design and deployment of A I. Great. I want to make sure we have time for audience questions. So I will go and see if there is. Does anyone have a question for? If not, I will carry on with my questions, but I just want to make sure we have time to audience questions. They want to have a, have a question already for a uh I'm not seeing any right now, but I will, I'll put this one there. Oh, yes. Um I need to wait, wait, hang on, hang on one second, wait till Mike gets here. There we go. There we go. All right. And if you could identify who you are also, Dan, we're in accenture. How do you design ethics into software? That's a great question. We need to break this down a little bit. There are many aspects depending on the applications as well. So there are some major known issues, for example, bias of the system privacy remaining applications as well as other other factors including, you know, human rights and other issues. And then for example, bias a lot of us are, are starting to be quite aware of this issue, right? Where did the data come from? Where, what's the representation of this data and what is the downstream impact? Um I work a lot in health care. It's very obvious that if our data comes from certain populations or, or, or socioeconomic class, it will have a pretty profound downstream um downstream impact. So at Stanford, for example, uh I think I believe Stanford University um and A I is the first university that pioneered an ethics and so society, society review process for our A I uh internal A I research projects because we don't have adequate um um um process uh yet today to review that and then we ask our research to, to think how to describe their um what their technology might impact at a human society. At the beginning, we really thought this is going to be a little bit of a burden for our students and faculty. It turned out this process was so welcome. We got so much feedback from faculty and students themselves who are innovating A I and say we thank you for this process because it gets us to think about how to design the fairness, design, privacy awareness and design human well being human dignity into our uh technology. Great other questions from the audience. If we can get a, a mic down here just down in the frontier and if you could identify where you're from. Hi, I'm from better brain. Um My question is at what point does A I start to take away jobs rather than augment teams? Yeah. Great question A I and labor market disruption and transformation is a major topic. You, you'll have more speakers on that too. At A I, we actually have founded a digital economy lab with experts across computer science all the way to economics, to study this. It's quite a nuanced question because the very definition of jobs is actually a collection of tasks. It's, it's most jobs is hardly one single task. So when we're talking about the disruption and transformation of jobs, we need to break it down to the impact of the technology on different tasks. And um you know, we don't have time to explore deeper into this right now that the huge large language models are definitely changing the way that text based and, and uh uh tasks are, are, are, are done. We, we're seeing software engineering assistance and all that. So the, the the 30,000 ft um comment I can make is, is absolutely gonna gonna transform jobs. How is, is it pure simple replacement versus transformation? And who are the people um who are going to be impacted the most? And how do we um um how do we guide our society through that transformation that is less painful or cruel? Are all important questions that we have to study forecast and, and, and really frankly work with policymakers? Great. Uh Another question from the audience, if we have one, they're just here at this table here. Uh The woman in the leather jacket, I think Rebecca Finley uh from the partnership on A I. Good to see you again. Um When we think about human A I collaboration and integrating A I tools into the workplace, one of the key challenges is around explainable and all the different dimensions of explainable depending upon the context. How would you rate sort of the current state of the art with regard to the ability of A I in the workplace setting or even in, in the research context? Great question, Rebecca. Um So for those of you who are not um um totally fully aware, explain is kind of a general word to um that encompasses just the understanding of the what the algorithm do. And in fact itself has different dimensions. Sometimes it's the mechanistic and algorithm uh explain explain of exactly how we get this inference results. Sometimes it's a causal explain that doesn't necessarily explain mechanism. For example, my favorite example is Tylenol to this day, we don't know exactly how that works, but we have causal ability of this drug and FDA approval. So that's another aspect of explain. Sometimes explain is just a trustworthy um trustworthy communication with the the human users that doesn't necessarily give you the mechanistic mechanistic details, but it induces enough trust. So um all three fronts A I technology is still nascent. We're missing a lot of ability on all three fronts and the applications of this is evolving and developing really fast. And this begs um industries uh academia as well as society at large for, for, for working this, for example, Stanford A I last week just rolled out the first benchmark of large language models. We're seeing a lot of that, you know, I'm sure chat GP P is the talk of the conference today. And how do we um how do we know how these uh large company, big tech uh large language model work against? What benchmark, what, where are the pitfalls and biases and Stanford as a open neutral academic platform that doesn't have another agenda is actually at the helm of uh trying to explain this by the way that the whole system is called helm. So no pun in terms. Um So you mentioned large language models, a lot of excitement about these. There's a lot of talk about generative ai you know, you have a great overview of the whole field. What what sort of most excites you about what's happening now? Are there particular innovations that you look at and say, oh, this is something that I think is really significant? Honestly, I just feel so privileged and and still feel surreal to be born into this time of history and be in the middle of this technology because it's, it's like, you know, every few days now or sometimes weeks, months, you see this incredible breakthroughs. That's just so exciting as a technology nerd, I wake up and just feeling, feeling so fortunate, for example, large foundation models are truly exciting me because it it will be it will be an empowering technology to many, many sectors, not just text based work, but also image based structure database and, and all that. So that's one area generative model is also, you know, especially as a vision is incredibly exciting. I'm just thinking about uh not only uh goofy arts we can do but thinking about its usage in, in medicine, in education, in sustainability. There, there is far more human centered applications in in this area and then some applications that are very exciting. Like you said, I work in health. But I think there is more, how do we empower our teachers, how do we empower our doctors and nurses? How do we empower actually um uh help uh governance and to help, to help people? These are all great areas. But you know, there we should not be completely um delusional about the negative impact and, and, and the human aspect of this. This is why it's a mixture of excitement and sense of responsibility. Excellent. Well, thank you so much, Doctor Lee. We're out of time, but thank you very much for coming and sharing those thoughts to open up this conference. It's fantastic.

Brainstorm A.I. 2022 - Salvador Dali Meets Wall-E

across our conversations, we've been discussing how AI is evolving and how it's being integrated into the fabric of all types of organizations. Now we'd like to share a demo on AI is remarkable creative abilities brought to you by one of the world's leading AI research labs. Open AI imagine creating original digital images simply by describing with words what you would like to see in a bespoke demo. Let's meet Valley. A nod to both Wally, the 2008 animated movie about an autonomous robot and Salvador Dali, the surrealist painter. Please welcome to the brainstorm Ai stage. Mark Chen, research scientist at Open I AI together with dolly. Hi everyone. Thank you for the kind introduction. And I'm really excited to be here at fortune brainstorm AI And today I'm here to talk to you about why what and how of dolly too. And um I want to sincerely apologize if you're here to talk about chat GPT but we'll leave that for another conference. So before I begin, I want to talk a little bit about open AI and our core mission. So our core mission is twofold. First we want to build safe and beneficial artificial general intelligence and by A G. I. We mean systems that can match or outperform humans at most economically valuable tasks. It's equally important though to ensure that A. G. I benefits all of humanity and not just a small slice. So earlier this year we announced dolly to and we released it in a platform called labs and as Fortune pointed out this represented a big step forward in text image generation And the big breakthrough here was approaching photo realism. By which I mean that at a quick glance, it's difficult to tell whether some of these images are computer generated or human generated. So at its core, dolly provides an interface where a user can provide a prompt, which is a sentence, a couple sentences of text and dolly can render a high resolution image that matches this prompt. So if I said something like fortune brainstorm ai conference in the style of Picasso, you might see an image like shown on the screens over here. So let's peer under the hood a little bit and talk about how dolly too was trained. So you can imagine dolly to at the beginning of training like a small child who has very little concept of what text or images are all about. And you start showing this child many flash cards, one at a time. Each flashcard has an image and a caption on it. And you can imagine that after seeing a million hundreds of millions of these flash cards that the child starts to form some associations like when it sees the word panda, it might say, oh hey in the image there's a fuzzy animal that's black and white. And so it actually learns through seeing so many images and text pairs, this own internal language which were representing by this strip of colors uh, for representing both vision and language. Now in order to generate a new image, What the model does is it takes the prompt, let's say fall colors next to a lake and it turns it into its internal language or representation. And then it uses what's called a diffusion model to render it into a highly realistic image. Now, if you've been following our work since Dolly one, this was using a technology more similar to gpt three, where we're generating the image row by row and then column by column within the each individual row. In contrast what powers dolly too is a diffusion model and you can see a visualization of how these are working in the slide. So you can see that the image starts out as very grainy, noisy textures. And over the course of generation it resolves into something that's crisp and sharp. So as I said at its core value to has texted image capabilities, which means that given a prompt we can create an image that's true to its concept, attributes and styles. And just to illustrate this point, let's say that I have a prompt like the following say it's a sea otter with sunglasses and a suit, classic Hong kong film actor, action movie close up with dramatic lighting. So even though sea otters wearing sunglasses and suits don't appear in our data set at all. Our models able to generalize and basically compose all these elements together to form a realistic image. And it's also not just able to combine these objects together. It also respects the mood lighting that you want and the overall ambiance that you want. Another example here is if I ask for vibrant and smiling, bowling balls rolling down a bowling alley in the style of digital art, it's creative enough to imagine that the finger holes could be eyes and the thumb hole could be a mouth and it really synthesize it in the aesthetic of like this kind of neon bowling alley vibe. So text image isn't the only capability of dolly. There's also in painting in which you can make realistic edits to existing photographs. And to give you an example, say I start with this image of a yellow couch in a modern living room and I don't really like this couch. I want to change it to something like chic grade chair so I can just mask out the middle part of the image and ask for a C grade chair and I can get one in its place and notice that this respects all the lighting and shadows that are implied by the original image, The lights coming in from the right side and you can see that the shadows are generated appropriately. One other interesting thing is that you can change these edits in succession so I don't have to stop here. I can ask for a modern fireplace on the wall and so you can imagine this kind of very iterative editing procedure with images. Next we have out painting which is powered by the same technology as our in painting. So what we can do here is we can take an original image and draw around the borders of it to synthesize a larger, arbitrary aspect ratio image. So let's take a famous painting, let's say the girl with a pearl earring and artist august camp use our tool which you can kind of see visualized here. And drew around this painting to visualize what could have been in the background and you can see that plausibly this girl was in a very messy kitchen and living room. And finally one thing we can do is variations, which means that we can take an input image and synthesize other images that look like this image but with slightly different features to give you an example, I can start with a dog dressed in formal attire and if I run it through our Variations engine, I can get these other images and you can notice that you know, they're all dogs dressed in formal attire but they have different ties are facing slightly different ways. And we've seen artists respond really well to this, they can visualize their work from different perspectives or take some art that they feel like isn't quite right and run it through the variations engine to find out and and come up with new inspirations. So finally, I want to talk about a couple of practical applications of dali. So when we launched ali, one interesting class of users who came to us were urban planners and in particular this one organization Better Streets ai they were occupied with. How do we show people what our streets could look like if we designed them around pedestrian experiences rather than vehicles and you can see that here, they've taken some city streets and rendered pedestrian and walkable spaces. And this illustrates the power of showing people, not just telling people next when you were creating dolly One tagline that we always thought about was we wanted to create an image generator for your imagination or image, search for your imagination and Microsoft this fairly literally. Um if you're not able to find the images that you want in being searched, you're able to go and synthesize your own images. Of course the potential for video games is also very clear, and in this case we had a video game designer create a character sprite sheet and also all the background and playable elements using dali and actually create a playable video game out of it. One kind of niche but fairly heartwarming application is in generating graphics for Children's storybooks. So Children's story books, of course they have this like very crayon drawn aesthetic. Um and it's very vibrant colors and we've noticed that many people have been bringing their story books to life. Uh here is one particular example where M I. T. Professor, he had a Children's book that he wrote a while ago and made images for them. And finally I want to showcase Kela, which is an end to end uh fashion ideation and design platform. And this is an application where users can go and use dolly to synthesize different clothing that they feel like could suit them and get it made and shipped to them in the real world. So this illustrates one really strong point of using Dali. It allows you to personalize images so that they suit individual tastes. And this is also a big reason that I was so invested in creating dolly for myself. I'm not a very artistic person and I feel like it's so empowering to give myself the ability to create images that are personalized for me. And to conclude, we went through what dolly is all about, how it works, uh the different modes in which you can use that you can use it for text image for in painting, for out painting for variations. We went through some applications for how you can deploy in the real world and I still think the most compelling applications for this technology are still to come. Thank you very much

James Manyika, Senior Vice President, Technology and Society, Google
Interviewer: Alan Murray, Chief Executive Officer, FORTUNE

Brainstorm A.I. 2022 - Can A.I. Help Solve The World’s Biggest Problems

Can A.I. Help Solve The World’s Biggest Problems?

Brainstorm A.I. 2022 - Can A.I. Help Solve The World’s Biggest Problems?

last september, our next guest announced Google's $25 million initiative to help organizations use AI in support of the U. N. Sustainable development goals, reporting straight to the Ceo James Manyika is tasked with assessing the impact of Google's tech on society. He also advises the White House on A. I policy in his role as vice chair of the US National Ai advisory Committee. We'll sit down with this season tech executive and techno optimist to explore how AI can be leveraged to address the world's most critical problems. Please welcome James Manyika, senior vice president of technology and society at Google for a conversation with fortune. Media ceo Allen Murray. Thank you James. Great to be with you. Good to be with you the last time we did this you had a different job. That's true. That was what five years ago it was pre pandemic. Well, uh congratulations, fascinating position you're in. Google has such huge ambitions in a I mean in the last couple of months I've heard talk about predicting natural disasters, dealing with climate change, healthcare, uh broad remit and healthcare, uh writing code et cetera. What is google's Ai mission? Well, first of all, I mean I think you have to go to the to the founding of the company. I mean at the end of the day it's a computer science information company. And I think if you look across all the things that Google is trying to do. I think they fall into a couple categories. # one a huge focus on actually advancing the field of AI so we can talk about the extraordinary number of breakthrough research and results. In fact a lot of the excitement we're all having now about generative models can be go back down to a particular set of research papers that came out of google, the classic attentions, all you need that opened up the space around transformer architectures that have led to these large language models. So the advancing the field itself is one of those goals. I think there's then there's also the practical goal about how do we actually bring these technologies to everybody. So through products, tools, infrastructure services, yeah, basically, you know bring this to the world and in the case of google is doing that in several ways, one is through google's products. So hopefully you've all noticed that search go better uh you know, google photos is much better etcetera, but also new products, but think about also the other companies if you like that google's birthing that would not have been possible. But for Ai so think Waymo think more fake, which is now a new drug discovery companies only made possible by the breakthroughs in AI So that's one said, I should also mention though Alan that I think google are also quite committed to this idea of how do we help others make use of these technologies, whether it's in on societal issues, whether it's climate change, flood forecasting science and discovery is another part of I think what google is trying to do and with ai embedded that deeply in the entire company. What's your job like how does that work? You report directly to the ceo you have responsible for a I had a responsibility for ai what does that mean? Well, so you know this is a dream job for me, alan I did my PhD in and robotics a very long time ago and to see the field come before you became a consultant, you're being very careful not to mention your former employer because even though Mackenzie, so if you look at some of the workouts doing at Mckinsey Global Institute looking a lot of the impacts of these technologies and air on business and the economy, I'm sure we'll get some of the business implications. But I think so my role is to connect the ideas and the research across all of google and alphabet and work with the team's leading that work to advance if you like. Our most ambitious research and innovations which include a lot of ai a lot of quantum computing and all these big computing infrastructure initiatives to make sure that they're being useful to people to society and that we're doing it responsibly and tie all the pieces together all the pieces together. So so you said when we were talking earlier, you said the last year and a half have been phenomenal year and a half for a I explain why well air is already making progress, but if you look at what's happened in the last couple of years I mentioned the transformer paper which has now led to this foundational uh, models that we're now talking about faith was talking about them. Look at what's happened with, for example, innovations such as uh, you know, uh, if Like diffusion models that have led to all these generative capabilities, I think the reason the last year has been interesting that I think we've, we've come to realize a few things. Alan one is these, these technologies have now become multi model so you can use them from text to images, texts too. Uh, code text, all these rights of application areas and similar. We've also discovered that in fact when you build, when you scale them up, you start to get properties and performance that goes far beyond what was possible. Before I'll give you a very good example. Uh, if you take translation, everyone loves Google translation, it works pretty well. And if you notice it says, you know historically cover covers about 230 ISH languages, it does probably 40 or 50 of them extraordinarily well before this last year and a half, that was kind of the story. Now we know that with these very large models, you can actually take languages that were historically called low resource languages where we didn't have the equivalent datasets to train them and you can translate those two. So this year, for example, because of these Large models, we added another 25 languages spoken by 300 million people like Bengali or Uganda, which people thought you couldn't really do those. And now we've just announced in fact we're gonna aim for 1000 languages. Why not? Yeah. When we were talking earlier about that, you said the science implications of this are particularly big. Can you give some examples? Yeah, I think the science arena fascinating things. You've seen what's happened in biology, protein folding. Our team at Deepmind basically went from, you know, this is a 50 year grand challenge where, you know, computational biologists didn't know how you can go from amino acid sequences to understand the structural proteins and every thought this was in theory Possible. So we went from zero between in the 18 months between, you know, the summer of 2020 from showing that you can do this with a I techniques to earlier this year and June predicting the protein structures of all 200 million proteins known to science. And we actually open That data set. So now you now have like, I think close to 600,000 biologists who are accessing these protein structure data set. So that science that's biology, you can go to physics. Uh if you if you care about fusion for example, it turns out one of the biggest challenges your infusion, high energy physics is how do you do plasma containment? Uh if you go to the Topamax in europe and other places are building. So it turns out you can actually use ai to stabilize uh you know, the plasma structures in physics, you go all the way to quantum computing. You probably saw last week we announced an incredible result in quantum computing, which shows that you can actually even create gravitational wormholes. So you can go in all these fields in science and I didn't mention mathematics. So we're starting to be able to show that AI can actually help you help us make extraordinary breakthroughs in this foundational fields that are going to be incredibly, what's the business model behind all that work in science? Uh Well, if you take an area like drug discovery in biology is straightforward, you hopefully we'll discover something extraordinary therapies and drugs we've now launched. I mean for google though. What? Yeah, I mean you you you talk about this almost like google is a university. Oh no it's not, it's a business. So the business model for us is I said come back to how we're doing it, we have a cloud infrastructure that's now ai enabled google cloud which is making extraordinary I think uniquely distinctive uses of AI. We also have products, core products, search and other things that are now, I think that's better, but also creating new products. So you're gonna see a whole range of new products that are only possible through a I come from, we have such knowledge in this audience and my knowledge is limited. So I really want you all to engage if you have questions for James or comments or something that you're particularly interested in hearing about, raise your hand and we'll get a uh we'll get a microphone to you right away so you can ask but but James, you're as you described your job, you're responsible for monitoring the interaction and the effects on society of all these massive changes in technology faithfully was talking about that a little bit earlier. How do you do that? I mean it's well to be perfectly my job is not so much to monitor but to work with our teams to make sure we are building the most useful technologies and do it responsibly. So to predict it in some ways to see it and to see the problems that may come in advance. Yes, but also make sure being we're building the most useful things now to be perfectly clear to your point, these technologies have come with extraordinary range of risks and challenges, right? We know that sometimes their own work is expected right? The leader to issues of bias and toxicity in their outputs and quite frankly non factuality. So that's a problem. There's also other issues that kick in such as for example the possible use and misuse of these technologies which you can imagine all the kinds of you know miss uses this could be put to. Then you've also got the kinds of things that we alluded to this, where we have to think about the impact in the labor market and the jobs. So there's a whole range of things we have to be thoughtful about. And that's why you probably noticed that in the case of google, google has been I think extraordinarily thoughtful about not rushing to throw these things out into the world without actually doing proper research and fully understanding what they're capable of. And these days, what at the top of your personal list of concerns? I mean when we were talking four years ago there was still a lot of concern about, oh this is going to destroy jobs, take my job away. I'm not gonna be able to work. Well we're 3.7% unemployment. Everybody who wants work can find work somewhere that doesn't seem to be the issue. But what are the issues that are top of mind for you? Well, I'll tell you the ones that I think a lot about. I think these issues to do when when they technology don't work as intended. So the, you know, the toxicity bias and so forth. And I think we're making progress on those things. I'm happy to talk about that. But I also worry about what I think of as the use misuse questions when is it gonna be appropriate to use these for what and quite frankly how we regulate that. I actually think, you know, this may surprise people, I think you'll find that, you know, I actually, you know, many of us embrace regulation because we're gonna have to be thoughtful about when is it appropriate to use these technologies? How do we put them out into the world so that we're not only, you know, not to throttle them. I don't think we should do that, but make sure we're using them in most useful ways. The most appropriate way with sufficient oversight and thought into them and that confidence that there will be somebody to rein in the inappropriate use is absolutely, there's a question uh back there in the back Yes. Mark Anderson Ceo of pattern computer. First of all, I want to say how cool, I think is what you guys are doing. You know, we we are involved in the same world in a tiny way, but google is a bright, shining light and I'm really impressed and I appreciate all you do. So that's a heartfelt thing. I will mention that we now have demonstrated explain ability. So, um I'd love to talk to you about that. We've actually done it. Berkeley Labs has kind of certified it for us. So I think we're good and it makes a huge difference. The I was inspired by the title which I assume you chose for this conversation and it had something to with the world's largest problems and we worked on that pattern. Um we're creating five new novel drugs for treating the top five cancers every month. So um we care a lot about this kind of stuff. There's a question I have a really personal question, I have to say that first. So when you is a little bit like the last question, but I hope more piercing when you personally look for what to work on, I think that there's this world of Ai discoveries can be broken into incremental and Pat Collison thinks it's too incremental or major. We call them pattern discoveries. Do you see that difference and how do you navigate that? Well, I do see the difference that I appreciate the question. I just think that the the kinds of questions in science and biology and in health that you're describing related drug discovery are such huge important issues for society. So think about those, think also about what's happening with climate change and sustainability. One of the things that didn't explain for example is it turns out that in fact you can do a whole range of things in climate sustainability and and the people affected by those, whether it's around flood forecasting and those kinds of solutions that are incredibly useful and important. So I think the big questions are really, really, really important and I think there's an interesting intersection here between businesses like google and others pursuing those opportunities, response and making sure everybody benefits from those solutions, but at the same time being responsible. You know, often this question of AI and responsibilities kind of framed as a either or in my mind, being responsible about AI involves both, capitalizing on all the amazing things that it can do for people and for society while being responsible. It's gotta be both. We have to do both. That's fascinating. Thank you James so much. Forget helping. Uh we didn't get to the business to the business application issues, but we will throughout the conference, you've really given us a great framework to think about what's happening in AI. Thanks for taking the time. Great to see you again.

Vinod Bidarkoppa, Senior Vice President, Walmart; Chief Technology Officer, Sam’s Club 
Nitzan Mekel-Bobrov, Chief A.I. Officer, eBay
Rathi Murthy, Chief Technology Officer and President, Product and Technology, Expedia Group
Moderator: Rana el Kaliouby, Deputy CEO, Smart Eye; Co-chair, Fortune Brainstorm A.I.

Brainstorm A.I. 2022 - The Age Of Intent: Knowing What Customers Need Before They Do

Vinod Bidarkoppa, Sam’s Club 
Nitzan Mekel-Bobrov, eBay
Rathi Murthy, Expedia Group
Moderator: Rana el Kaliouby, Fortune Brainstorm A.I.

Knowing What Customers Need Before They Do


Brian Schimpf, Chief Executive Officer, Anduril Industries 
Alexandr Wang, Founder and CEO, Scale AI 
Moderator: Michal Lev-Ram, FORTUNE

Brainstorm A.I. 2022 - Reinventing National Security: A.I. And Defense

Reinventing National Security: A.I. And Defense

Yael Garten, Director of AI/ML Data Science and Engineering, Apple
Interviewer: Verne Kopytoff, Senior Editor, FORTUNE; Co-chair, Fortune Brainstorm A.I.

Brainstorm A.I. 2022 - How To Turn Good Products Into Great Ones With A.I. 

How To Turn Good Products Into Great Ones With A.I.

Brainstorm A.I. 2022 - How To Turn Good Products Into Great Ones With A.I.

consumers expect elegant design and products that are intuitively easy to use right beneath the smooth surface. The smooth interface, artificial intelligence already powers much of its operating system from Siri to face I. D. But how is this tech giant leveraging Ai and machine learning into emerging products and services? Will ask the senior Apple executives to share her insights into Ai advances at the world's most valuable company. She is also on the board of a 169 year old retail company and can address how a legacy brand embraces the technology and is able to revitalize its processes and products. Anybody know what that retailer is? Please welcome Yael Garten, director of Ai ml data science and engineering at Apple. And member of the board of directors at Levi Strauss and co for an interview with Fortune Senior editor VErne kopytoff. Hello everyone and thank you for coming. You thank you thank you. Uh to start things off, could you set the stage and list a few? I mean Apple has so many products. Could you list a few that um that use ai your favorites for example. I love them all. Um But yeah, if you think about our hardware, software and services ai and Ml Ds really extensively across them. To name a few of these, a couple of them in the information seeking space uh theory and search research for music and and movies and a lot of great content news um that those are some examples um you already mentioned face I. D. Earlier. Um If you think about the accessibility area, a lot of what we focus on is making technology accessible to everyone and making the world accessible. So one of my favorite features that we recently announced is a door detection which enables people that have low vision or blind to use their iphone camera to detect the door that they're about to near. And it reads two texts that you see on the door, how to open the door handle. So that's really a great example of making the world accessible and using technology for that. And it's powered by machine learning. Um and then maybe a third example in our in our health uh areas of investment um is um what we call walking steadiness. Um So this is an example where your iphone detecting how you walk and your stability and and uh we are able to use machine learning in this case to identify trends in your walking stability um and and identify you when there are changes to those trends and how you might take actions to improve that to help reduce uh risk of falling and other other kinds of things. So really expensively across hardware, software services, those are just a few. And Apple is so big and has billions of users. What are the unique challenges of dealing with data for such a huge operation. Yeah, well I don't know if this is a challenge. I kind of view it as an opportunity and one of the things that we um that we take very, very seriously is is the privacy of our user data. I mean, so that's something that I would just highlight as um really we put a lot of focus and thought and energy into how can we create a great machine learning based products that are also privacy preserving. Um and so that involves right, taking more time, more thoughtfulness in the design of features. How can we do more and more on the device, really leveraging the integrated hardware, software staff that we have thinking about using the hardware using the framework building applications um that enable us to preserve privacy and develop a great machine learning based products. Um so that's one of the areas where I'd highlight as we we spend a lot of time and and focus on that as well as making that something that third party developers can, can leverage as well. And when you're working on your projects? I mean, you've worked on so many, what are the key lessons and sort of pitfalls you've you've come across? Yeah. Um well maybe a company of any size, not just one the size of Apple. Yeah, well if you're if you're talking about how you implement machine learning and maybe more more broadly, I would say there's this life cycle of uh when you want to deploy a machine learning based solution, um how do you take into consideration all of the phases? And this is, again, really speaking pretty broadly, how do you take into account all of the phases of that development life cycle and plan it out? I think I mentioned earlier this lifecycle of machine learning development right. Which starts from what's the problem? We're trying to solve what's the product experience, what's the workflow? Um and this can be really in any company. Right. Apple at Levi's um and really any company. So if you think about what's the product experience and then you move on to what the data that we need to collect um to be able to create that machine learning based experience and then how are we going to train the model and how do we ensure that the data is representative? And how do we then evaluate the model before we deploy it? And how do you then ship it in terms of a really cohesive user experience and how do you later continue to evaluate the performance of that model as it's out there in the real world and as the world is changing and people are changing macro trends are changing and continuing to have that responsibility of thinking through all of those steps. I would say that in general, um underestimating the um the amount of time and investment it takes is probably one of the pitfalls that I would say, you know, entity, small and large industry academics, you know, etcetera underestimate. And so just thinking about that in advance and planning for that is something that's pretty important um, as well, putting in place the structure in terms of, you know, team and organization to have um, adoption of machine learning based technologies as something that is part of the conversation and something that companies and entities are really planning for. So that's probably kind of two of the common pitfalls and what's the level of investment, what's the workflow and then what's the team and structure that we need to have in place for that adoption and what's the, what's the secret to knowing that you're even building the correct models? What steps do you take to make sure you are? That's a great question. Um, you know, I would say it really just comes back to what's the use case, So really what is um and again really broadly, you know, what's the business use case you're trying to solve? What's the product experience you're trying to solve? Um so if I take, for example, if I think about how we develop things that Apple, we really try to think about what is that end user experience that we're trying to deliver and really obsessing over that at the beginning and ideation phase all the way through the end stage. And I think that that's really important because then, you know, she learning for example, which was talking about a lot today is is an an ai part of that, but that's not it. So, having uh, that question really guide us then brings the right people to the table. Right? So if its product managers, designers, engineers have different types of hardware, software, machine learning data engineers coming together uh bringing in, you know, members from from marketing and from the legal team to really think through what are the ethical implications of what we're doing? Um What are the intended consequences and what are the unintended consequences that might come out? So bringing together all of those components as you're developing the feature, as you're developing the solution is gonna be some of the key aspects to ensuring that it does what you want it to do and it doesn't do what you don't want it to do or that you've carefully thought through all those steps as you're going into production. Um and then again really putting in place um mechanisms and rigorous evaluation programs to evaluate after deployment. Um is it working as intended? If not, what are the changes to make? I think a big part of that is really having metrics that help you understand how the models and how the solutions are performing in the real world and being able to adopt and adapt to those metrics as kind of the North Star of are we on the right track? Is it going where we wanted it to, we go into questions in a in a couple minutes, so please have some ready. Uh Now you're on the board of 100 and 70 year old company levi's what are the challenges of getting such a legacy company going with Ai and what are some of the projects they've been able to pull up? Yeah, great. Um yeah, Levi is a really iconic brand and it's one that I would say it's actually been investing in technology and innovation for for many, many years and including in recent years in data and AI. Um and so there's it's been really exciting to see because a piece of it has been, you know, digitizing um data that wasn't digital before. Right? So moving everything from manual to digitized in a way that it can be a basis for building applications on top of it some of it interestingly has been taking data that they've had and making interesting use cases of it. So for example, think about archives of genes going back to the 1850s and how can you take um images of those genes and then start to think about ways and they're doing this in ways of um uh which are ai assisted to design new products and new fabrics based on those archives and based on analysis of those. Um So that's really interesting ways of leveraging existing data and then also great examples of bringing in external data. So for example, um an example I really liked was bringing in weather data um and using building models off of that and showing how um actually when you have differences, for example, slight differences in weather. Um the team found that in Milan versus Rome, the period of time which skirts were worn was different or the color of denim that was purchased was darker due to the winter coming. And so that starts to enable the company to shift strategically in ways that enable different distribution of goods, uh different perch of goods that are coming to the different stores and you start to think about that at a global um you know, basis and it really starts to change how goods are distributed, how you operate distribution centers, how you plan your pricing, um how you order in advance and so it starts to bring in really agile ways of um running the company um in ways that you know, prior were done differently. And so as more and more of the company starts to embrace these different, great you know, examples of great ways that machine learning or uh you know, data data science have been integrated. More of the company starts to um to leverage that in addition to just really great ideas that many companies are doing and many of you that we spoke about earlier are doing bootcamps. Right? So how do you train more and more of the workforce to understand what data science is and what a i is and how you can leverage that so that it's really something that is coming from more and more pockets of the company versus for example, just one centralized team that's developing the solutions and so those are some of the examples of how they're approaching. Um Great, great. Let's go to questions, we have a microphone in the room um just raise your hand please and stand up and identify yourself. There we go. Hi, I'm kai from iron clad uh and I was wondering are there things that you have thought or advances that you've seen in ai that you thought would be coming for many, many years to come? I mean how do you design an organization that works around kind of the unpredictability of how advanced what tasks are going to be possible versus what tasks are um maybe disappointing in how long it's taken them to be accomplished. Yeah. Um you know, I think in terms of how you, how you organize, you have to have a portfolio, a mixed portfolio of investments. Right? So we know that the field of machine learning, artificial intelligence data science are rapidly advancing and in some ways that we can't yet anticipate. Right? And so you have to have an ability to have investments that give you, you know, a steady straight steady stream of wins, um whether it wins your business when your product wins to the team's agility. Right. A lot of um aI can be used now to really help teams be more operationally effective. Um So really having a steady stream of investments that gives you um you know gains while also making some investments in the team to seed uh you know early research or investigations that might yield fruit later but may also change as the trajectory of technology is changing and we've talked about it today, it's changing so rapidly. Um and so really having a mixed portfolio is is probably what I'd recommend. Do you have another question? Anyone, anyone anyone uh Now there's a lot of hype that Ai is a solution to every problem. What do you say about that I mean and what should executives ask themselves to decide whether ai is the solution? Yeah, so I agree with you that there's a lot of hype and I also do think that um while we're at brainstorm ai and there are a lot of great uh innovations and and and enablement that I bring. I don't think that Ai and machine learning is the solution to every problem. So I do really think it comes back to the core of asking what is the business problem, I'm trying to solve? What is the product problem trying to solve and being really clear on that to then be able to map out, you know, what is, what is the hammer that fits my all right um or sorry what is the nail that I need and what anyway, you understand what I'm trying to say but but I don't think it's the solution to everything. So I think it's really important to focus on what is it that you're trying to deliver and identify is very simpler solution or it's actually the right time and space for you to go and invest in a machine learning based approach or an artificial intelligence based approach. Um I think when you've decided that it is the time to invest um and that it is the right solution for you, I do think it comes back to you touched on this earlier, it really comes back to these kind of a couple of areas of investment that it's important to to do. I think one is really thinking through that workflow end to end of, how is if you're going to put a model into production, what is that workflow, who are the end users that are going to be interacting with it and really stretch that out and to end so that you thought about all the consideration going from what data you'll collect, how will this be interacted with. Right. Is it a dashboard? Is that a ui is it a product, is it in the field? You know, what else is going to interfere without being successful? What is going to enable it being successful and really thinking through that whole workflow into. And I think we see a lot of examples where in clinical health care, it's just uh the there are a lot of great ideas on the technology side, but getting it into the clinical workflow um isn't happening and it could be because the doctors and nurses don't know how to interact with it is because it could be because I haven't thought through the financial construct of who's going to pay for it. So I think that thinking through that workflow is really important of how is somebody going to interact with the model doesn't need to be explainable. Um Is it explainable? Right, so a lot of those aspects um the second is really say around the data. Right, so how are you going to the data cleanse the data, train the day to make sure that the data is representative and kind of protect that data and really thinking through that workflow. Um And probably the third area is around um the team, right who is going to be implementing the models and how does that really get structured in terms of your organizational structure for um joint success and joint accountability. Um So thinking through what you have, is this under a general manager working closely with a data leader, where you have shared responsibility for implementing that model in production um and where the outcomes are you know feeding back to your business. So really thinking through that construct is something that helps set it for success. Yeah, thank you so much for coming. Thank you. Thank you all

Teddy Bekele, Chief Technology Officer, Land O'Lakes, Inc.
Fiona Tan, Chief Technology Officer, Wayfair LLC
Moderator:  Verne Kopytoff, FORTUNE

Brainstorm A.I. 2022 - How A.I. Is Spreading Throughout The Supply Chain

How A.I. Is Spreading Throughout The Supply Chain

Sasan Goodarzi, Chief Executive Officer, Intuit
Ashok Srivastava, Senior Vice President and Chief Data Officer, Intuit
Moderator: Jeremy Kahn, FORTUNE

Brainstorm A.I. 2022 - View From The Top

Sanjeev Vohra, Global Lead - Applied Intelligence, Accenture
Javier Polit, Chief Information and Digital Officer, Mondelez International
Moderator: Alan Murray, FORTUNE

Brainstorm A.I. 2022 - Leading With A.I.

Sanjeev Vohra, Accenture
Javier Polit, Mondelez International
Moderator: Alan Murray, FORTUNE

Brainstorm A.I. 2022 - Leading With A.I.


Kevin Scott, Chief Technology Officer and Executive Vice President, Technology and Research, Microsoft 
Interviewer: Jeremy Kahn, FORTUNE

Brainstorm A.I. 2022 - A.I. And The Promise of Large Language Models

A.I. And The Promise of Large Language Models


Brainstorm A.I. 2022 - A.I. And The Promise of Large Language Models


Welcome back everybody. Welcome to day two of Fortune brainstorm AI. We have got a great program ahead. We're going to explore the amazing advances made by large language models, and then we're going to get a firsthand account of the Ai focused digital transformation at a snack food giant. We're going to close out our day with a keynote conversation with Reid Hoffman at the end. So many of the people who have joined the conversations here at our gathering offer great examples of the kind of purposeful leadership that we want and need more of in the world. And that's exactly what led Fortune to create. Our newest membership community called Fortune connect. It's designed to help prepare those who aspire to become leaders of the future and we're really, really proud of this program. It's seen tremendous growth since its inception just 2/2 years ago. The goal is to create truly purpose driven leaders and to help them understand the dynamics of stakeholder capitalism. What it means, how it works when society is demanding so much more from companies such as yourself. The program is designed for mid career professionals who are rising fast on the executive track. I'm sure you have folks like that in your organizations and I would encourage you to consider putting them in this program. Um please check it out at fortune connect dot com when you can. Okay, before we start, I'd like to thank our partners for their work in bringing us all together here are presenting partner partner accenture and our partners C three ai and intel, we greatly appreciate your work in making all of these conversations come to life now let's get to it. They're called foundational models giant Ai systems trained on vast amount of text sometimes combined with images or even video. The results are systems that have achieved breakthroughs in a range of natural language tasks, from writing to summarization, to question answering as well as generating new visual content. Many see the models ushering in a revolution as these systems are deployed to accomplish knowledge work that was previously previously reserved for us. Lowly humans. But does big tech have a blind spot when it comes to the limitations of large language models? How do we assess these risks? We are really really fortunate to be able sit down with a veteran engineer at the top of the industry to share his insight into the promise of foundational models and his views on responsible ai. So please welcome kevin scott, Chief technology officer and executive Vice president of technology and research at Microsoft for a conversation with fortunes, jeremy Kahn. Hello everybody. Thanks for joining us this morning. Um Kevin, I want to start out by asking you why are these large language models and large foundation models? Such a big deal. Why do you think they're so important? Well, I think a bunch of people are getting clued into why they're a big deal. If you just look at what's happening on twitter with chat GPT over the past handful of days. Um so I think the thing that's really interesting about them is uh this very idea that they are foundations that they are platforms on which you can build a whole bunch of other things that they have just become powerful enough, where you're no longer in this mode that you have to do machine learning, where you have a narrow slice of data, your model, your training pipeline, and you're building something that is narrowly focused to a particular feature or product. Like we are now able to invest in building these these really big things and then thousands, tens of thousands, hundreds of thousands of people can use them to build all sorts of incredible things uh which is great because one of the things that it does, it really opens up the aperture of who can actually use Ai now. Uh so you uh will need new sorts of specialties in the future, but you don't need to have a PhD in computer science anymore to build an AI application, which I think is really, really exciting. Yeah, fantastic. Um I know that Microsoft uh you own GIT hub and with GIT hub you created something called co pilot, Can you tell us a little about what co pilot is, how it works and how you see that as a model for other products and features that you're hoping to roll out going forward. Yeah, so co pilot is the thing that we built in uh collaboration with open AI. so um we had this epiphany pretty early in the development of GPT three, that these large language models were actually able to write code. Um And so we uh invested additional you know energy and focus and built this product called Get Up Co pilot which is a coding assistance, so it allows a developer to express in natural language, a piece of code that they would like help writing and it admits it. Um and it it is you have been one of those things where the productivity gains are high and maybe more importantly the amount of love that developers who use it express for it and like how quickly it's become an essential part of their workflow has been just sort of surprising and wonderful to see that the you know, the thing that I think is interesting is that uh it is like get up co pilot for coding assistance is just 11 sort of co pilot that will exist in the world. Like I think in the limit you will have lots and lots of these things helping people with a pretty wide range of tasks, whether you are a video editor trying to pull together a piece of content or you're a journalist doing research or writing an article if you are a doctor trying to deal with your caseload, Like almost everything, I think that you can imagine where you're doing cognitive work, you will be able to see things like co pilot assisting with that work for everything, correct. Um How do you respond to customers who are really interested in this? Think this sounds you know like something that might help them in their business, but they're basically kind of afraid of using these models because they didn't train them themselves. They don't really understand the biases. I mean how do you sort of respond to customers who have those concerns? Yeah, like it's a thing that we hear all the time and I think the way that we get past this is we just sort of see how you you know, you build these dependencies like if you look inside of an organization, like any of you who run engineering teams, like this idea of taking a dependency on something that someone else built is just a hard thing to do in general. Um And yet we we do it all the time. Like people don't think about uh when they need to write a piece of software, do I have to write my own compiler or do I have to build my own operating system from scratch? Uh And so what what I think will happen. What we'll see over the coming years is that some part of the functionality of these big foundation models will start to look more like an operating system, a thing that, you know, it's just so complex where your organization probably won't be able to make the investment in building the most powerful possible thing um, and where like a small number of folks will be able to just because they are building a platform where the cost of building the platform gets amortized over a whole bunch of different things. Uh, and then we will have to figure out what the business model for this stuff looks like because you know, I think Bill Gates said once or at least it's been apocrypha li attributed to him that the true measure of success of a platform is when most of the economics and the value created on the platform, go to the creators, not to the person who built the platform, wow. Um, now I think we'd be remiss if we didn't ask you a little bit about this issue with co pilot, which is that as you know, has now been sued by a developer who says, well, my code was used in the training of this, um, but I'm not getting any compensation and I don't think, you know, is the proper use of the license of that. Is there anything you can say about that lawsuit? Well, obviously like, I can't say much about it. Um, but you know, the thing that I will say is like a bunch of this stuff will get figured out over the, over the coming years. Um, yeah, we they're, they're a bunch of different philosophical ways of thinking about how you build these models, uh, and I think we'll be okay as long as, you know, the world is building things that create value for lots and lots of people versus building things where you're sort of concentrating things into fewer and fewer hands, like that's where we will get into like real, real trouble. Right? Obviously, there are other concerns about safety around these models. You know, what are they gonna, will they produce biased content? How how have you at Microsoft? I thought about trying to mitigate those. Yeah, we we started many years ago before uh like transformers and deep neural networks were uh were even things thinking about how do you build a responsible AI practice? And it started as a thing in Microsoft research. We have now built a real responsible Ai discipline where we have an office of responsible ai that sits under brad smith and our legal team and we have a uh ethics committee that is composed of not just computer scientists, but social scientists and uh you know, folks who are experts in different disciplines that are very relevant to deciding, you know, what's safe, what's responsible, what's ethical when you are thinking about applications, these models, we have published a thing called the Responsible AI Standard, that's in its second version, it's like available for folks to have a look at. And like we have rigorous processes inside of the company, the try to evaluate uh what we're launching against that standard to make sure that at the very least we're being consistent with our own uh stated guidelines. And we're more than happy to have uh public debate. That's the reason why we published the standard about like where the lines ought to be and like we've drawn, you know, uh a bunch of very clear lines about like what a sensitive uses and like in in the cases of some of these sensitive uses, uh you know how you should deploy machine learning or AI systems, if at all. Right. Um Are there questions in the audience? I want to I want to let you all get a chance to to ask kevin anything that's on your mind. So please think of think of questions. Um there's one there go first. Please stand up and state where you're from. Hey, I'm Jill from capital G. Thank you so much. Um Obviously with co pilot, you guys had a really clear example of using the innovation from open AI sort of tightly coupled with the distribution of GIT hub. How do you think about other examples of that where open AI and Microsoft might work together. And I think you mentioned sort of other co pilot examples maybe, but a little bit more about how you expect that relationship to continue forward over the next several years. Yeah, I mean we we've got a really awesome partnership with open AI that's multifaceted. So like the biggest thing that we do together is we build ai computing infrastructure for training big models, so GPT three? GPT 35 which just got released chat. GPT like all of these things, like we're not only trained on Microsoft built infrastructure but uh the imprinting parts of what they're doing, so when you make an api call like that is going into like highly optimized infrastructure to support some like pretty technically complicated things that a lot of work went into. Um but we also collaborate with them on, you know, hey, here's an interesting application for these models. We have a Azure version of the Open AI A P I that runs inside of our cloud that is enterprise hardened and has all of the things that some customers both expected need from, from an enterprise version of something like this. Um And you know, we we even do a bunch of work together on safety and responsibility. Um So we've we have an awesome partnership with them. I just can't say enough good things about, about the people at open Ai they're just extraordinary folks. Other questions, there's a question down here in front, we can get a mic to this gentleman, please stand up and broad Meadows from the AI Foundation. I'm curious based on these new foundation models has the profiler skill sets that you hire for changed in the last couple years. Yeah. Um one of the big transitions that we've had and this has been going on for, you know, four or five years now, is that um what are machine learning engineers do uh in their job day to day has changed pretty dramatically. Um So yeah, I referenced it earlier, like it used to be the case that a machine learning applications team, inside of a product or feature team at the company, like had their own data, like had their own model, like they decided what the model architecture looked like, they would train it, they would uh like do the evaluation on the model that they trained, they would deploy it into some testing infrastructure and see how it performs and then they would sort of run their own like feedback cycle on top of it. Um More and more of what we're doing right now is like we have a very small number of foundation models that we use inside of the company that people use either directly like zero shot or one shot where there really is no fine tuning for a particular application or that you do a few patterns of like fine tuning to get it to perform in a particular way. Sometimes it's cost, sometimes it's uh yeah, you needed to have some performance characteristics in terms of quality that the base model doesn't have uh with, with no fine tuning. Um And so that like that's just a different mindset than the one that we had just a few years ago and like it's an interesting transition for folks in the organization to go through. Um and like oftentimes, it's a interesting thing and the teams who have little machine learning history, uh like can just go faster because they don't have a bunch of things, they have to go unlearned. Uh and they're just sort of open to a new approach rather than, you know, digging in and saying like, oh no, but my way is better. Um so it's a big transition happening. Um I think for folks who don't have their own specialized machine learning teams, like all of the foundation model stuff is almost nothing but goodness because it basically means that you have a more approachable set of expertise that you need to like get going. Um and that like, we, we still need that deep, deep expertise on like how to do all of the optimization, but like we can again, you know, focus that on foundations and then like that work gets, you know, advertised across the gazillion applications, kevin in 2020 you wrote a book called reprogramming the american dream where you kind of argued that, you know, Silicon Valley is this great place, It works really well, but we, we need to create this opportunity everywhere, not just in Silicon Valley. I'm curious if there's a, I have a role to play in that vision, I think it absolutely does. Um the the reason that I wrote that book is there, there's a vision of the future where um where these foundation models don't become platforms and where you see further concentration um into a smaller and smaller number of companies and you know, hands of experts that are mostly in like urban innovation centers in the United States and a couple of other places globally. And I think, you know the interesting thing if you concentrate on making these things available as platforms to lots of people and like you try to uh it's not even that they're free, but like if you get lots of people using it, you can drive the unit economics of what they cost to access even down. Um and it just creates a bunch of opportunity. Like if I look at the part of what I did in the book is I went back home to rural central Virginia where I grew up and I talked to a bunch of people who had entrepreneurial mindset and given these tools like without computer science degrees, like these folks will be able to see the opportunities that they give them and like they will absolutely be able to incorporate them into the things that they're doing and their businesses and to create new businesses that they hadn't imagined were possible before. And to me that feels really exciting, right? So I mean the idea is you could use machine learning in your business even though you have no background actually in it, you don't know how to code, you just instructed, you know, here's what I'd like to have this function do this app to and it is built for you. Yeah, I mean a lot of and and again you all are sort of sort of seeing it play out in real time. Like we we have uh we've been on this journey for many years now just because we we sort of had confidence that this is the way things were going for a while, but with chat gpt like what you're seeing is you can via prompt engineering with like expressing what you want to have accomplished a natural language, like you can get these models to do very very powerful things and so like that's just a much more approachable way of being able to harness the power of digital technology and like I've been talking about this for several years and like this is actually cribbed from Peter nor big, who gave a talk about it like seven or eight years ago where you know the thing that machine learning and ai does is it teaches, it transforms the task of Harnessing the power of your machine from programming to teaching. Uh and programming is an activity that's like hard to learn. Like you have to go to school, you have to practice like it, you know, believe Malcolm Gladwell or not like you probably need to put in your 10,000 hours to get really, really good at it. Whereas a toddler can teach someone how to do something like it is a thing that's inherent to us as humans. And so I just think it would be a more much more accessible way for people to get real power out of their technology than than this paradigm that we've had since Ada Lovelace, you know, 170 something years ago. And right now there's still quite a lot of art to kind of prompt engineering. And there's people saying, oh well we should start teaching kids how to prompt engineer. But I've heard other people say, no, I don't think we're gonna need to do that in five years. What's what's your view of? Oh I I think you are going to have to. I mean I was chatting with one of the teachers at my kid's school who saw chat GPT and they're like, oh my God, like now I'm not gonna be able to get my kids to write essays anymore because they can, you know, cheat, use chat Gpt to do the work for them. And I was like that is it's a pedagogical mistake to think that the essay like the artifact that you get at the end is the important thing about what it is you're trying to teach a student. Uh And so in a sense like nothing really is changing here other than you have this tool and like the student themselves like has to become the teacher to the model. So like the first thing they ask it is gonna be crap. Like it'll have factual mistakes that will the models. Like the hallucination rates like everybody should just sort of be tracking or like going down like you know chat gPT hallucinates way less than GPT three. And it's just getting better and better over time. But like still like the thing will make mistakes and so your job is like be the editor for this thing, be the teacher like coax it into getting you the output that you really need. Like that's the important thing. That's the special thing about us. Like the thing is just a tool so won't won't know much about history but we'll know a lot about prompt engineering. Like um No kevin thank you so much. We're just about out of time but thank you so much for joining us. This is a fantastic discussion.

Stefano Corazza, Head of Studio, Roblox
Richard Kerris, Vice President, Omniverse, NVIDIA
Natalie Monbiot, Head of Strategy, Hour One
Moderator: Rana el Kaliouby, Smart Eye

Brainstorm A.I. 2022 - A.I.’s Role In Shaping The Metaverse

Stefano Corazza, Roblox
Richard Kerris, NVIDIA
Natalie Monbiot, Hour One
Moderator: Rana el Kaliouby, Smart Eye

Emad Mostaque, Chief Executive Officer, Stability.Ai
Interviewer: Jeremy Kahn, FORTUNE

Brainstorm A.I. 2022 - Generative A.I.: Silicon Valley’s Newest Gold Rush

Generative A.I.: Silicon Valley’s Newest Gold Rush

Welcome back. I hope you all have a chance to stretch your legs and grab a coffee. And now let's continue with our main stage conversations. There's probably no hotter topic in ai these days than generative ai. It's an umbrella term for software that instead of just analyzing existing data can create original blog posts, Wikipedia entries, artistic images, video code and more up next a demo of the remarkable tech from stability ai followed by a conversation with the firm's founder, Ahmad MMA Stock. He'll lay out the economic and creative upside that will follow from making this open source free to use code available to anyone who wants it. Please welcome to the brainstorm. Ai stage. Amada stock founder and ceo of stability ai we'll get to the demo in just a minute. But first of all I want to ask you, why did you do this and especially why did you do this as an open source project? I mean potentially you know, this could make a lot of money and film entertainment industry, why do it open source for fun and profit? I think I saw this technology and this technology classical AI was built for one thing there that can do anything and I realized it would be like servers and databases which are almost all open source and it's all about scale and service customization. So I thought rather than wait five years to get there, let's just go there at once and allow it to spread everywhere, which it has. So like for example, stable diffusion has just overtaken Bitcoin and ethereum on get hub stars in terms of developer popularity, The fastest growing open source software ever. With the biggest community building. The craziest things, wow. Um now one of the things when stability sort of 1.0 came out, there was a little bit controversial is that unlike some of your competitors like dolly, you didn't apply any filters to either the training data or the output to make sure that people couldn't generate pornography or other sort of not safe for work images. Now with the two point oh version that was just debuted last week, I believe. You know, you've actually rolled that back a bit. You have applied some filters that make it much harder to generate pornography. Why the reversal on that. So the first part was actually a research initiative led by the LMU. This team at the University of Munich with collaboration from ourselves runway Ml and some others thus providing the main computer infrastructure and Robin Rombach, our lead generative code are doing it and they said, let's start it as a base and make it just an image of the internet and see what we can do. And that's what came out. But we put a safety filter on it and an ethical use license as well to limit the amount of output for our own release of stability. We decided and listen to the feedback that we would filter out not safe for work elements. So more people could use it. We decided that we would also make it a more open data set. So we can really understand how the model worked. We spent millions of dollars kind of doing that as well, which is now results in the model that anyone can use in a commercial and corporate setting. Great, well let's let's let's see the model. And this is a special demonstration. This is the first time I think you've shown this particular version of the model. Can you, can you explain what you're gonna be showing us? Sure. So basically, um we released 2.12 days ago, released the original one august 23rd release 2.1 tomorrow. Um this is the place of kind of innovation here to put one is very special because when we released 1.1 point four, actually in august 23rd, it was 5.6 seconds per generation. We released that produced that 2.9 seconds for an image. Um just a few days ago, as of next week, it will be 30 images in a second. So we designed it to be fast and now we've got to real time. This unfortunate isn't quite real time, but we've got the one second generation and again it can generate anything. So you take words, you put it in and for example, is on stage between yes, you can do a lovely labrador, right labradoodle. We like doodles. Labradoodle and then in the second it generates a picture of multiple, lovely labra doodles with a hat. And then boom, it generates it with a hat. I think it's just loading on the internet and then we can do by van Gogh and it becomes by van Gogh. Um so this is actually a very interesting thing because 100,000 gigabytes of image is compressed into a 1.6 gigabyte file that works on your Macbook. And as of last week it works natively. So you'll be able to generate these images in a second in a stained glass window. So you can generate just about anything because it understands two billion different concepts. And this is just unimaginable. Right? And like that. As of next week, it will be 30 frames a 2nd, 2.1 adds back a lot of the artistic elements because some lovely stained glass windows. Isn't that nice? But then it's also very flexible. So for example, what? We can do it the wrong way, We can do, let's do nine images here. We took four images of jeremy and we decided to create a jeremy can model this is just with four images. Who knows if this will work? Uh a picture of jeremy Kahn as an astronaut. You wanna go to space and then we can do HD matte painting, why not live with the new technology. Who knows if it will work or not? We'll find out in a second. The suspense is killing me how we doing there. Maybe we should make a professional actually it wasn't professional enough. It's interesting because it understands all these kind of modifiers. If you do a professional not professional, you want to negative prompts. You can say not ugly and it becomes not ugly. Um You gotta do not ugly for me, definitely jeremy like any of these like how is this one here? Yes, not bad, not bad. There we go. Just four images. We've got jeremy there and now we obviously do by Van Gogh. Right, so I think you know, super powerful technology that can create any image unimaginable. Just that's not that's not me. That's not another jeremy's emerge. Yeah, so matt professional, let's do photo with some 35 millimeters studio lighting to make it good. So I mean I think this is the thing, this technology is still nascent, right? This is unimaginable. Just a few months ago, let alone in the last year, it's just evolving at such a rapid rate. Still not used. Well, it's closer but lovely. This technology is designed to be kind of zero shot where you just put something in you get out. It's part of a whole design process. So we have people seen people take this and create entire movies from it because they take frame by frame like into the Spiderverse and they transform it into an anime movie. We've seen people take this and create architectural designs at scale and you know, again it's part of the process. It's amazingly powerful tiny file. We put it out to the world. And now people are developing around it, developing ecosystem that they can then take and build really useful applications. That's what you need actually useful things. This is why the number one app store, app one app store of the year is one bow which uses our technology. Number one on the app store right now is lens, which creates these types of customized profiles based on this version of the technology as well. So I think it's gonna be very interesting to see how this develops going forward. Absolutely. And there, I mean there are skeptics out here. I would say this is great. It's a lot of pretty pictures. But what's, you know, what's the real business impact of this? So like if you look at movies for example, this will cut down movie generation time by half. So we have an example of creative director was shooting a movie in the desert about serial killer with some very famous actors and actresses, serial killer had a book of all the victims that he had and they're gonna have to hire 25 actresses to fill that book, put them in the seventies makeup and you know, fly them out. Sag rates make up all this kind of thing. $600,000. He trained a custom model overnight on seventies pictures. It was done, save $600,000. That's just one example when you look at video games, movies, other things that transforms them any type of design process of working with automotive manufacturers to create brand new car designs. Because you can do these kind of dynamically, you know, you can just type in like an awesome super car in SAn Francisco and it will generate those but you can customize it with your own assets. This thing when you take the standard model and customize it, it becomes super duper powerful. That's an awesome super car in SAn Francisco. And are you thinking of using this beyond just images? Are you thinking about using the same underlying technology for generating other sorts of things? Yeah. So we have the most popular language model in the world. DPJ Neo and others from our Luther ai kind of community and we actually do models of every single type. So we have audio models that can generate any type of music. We've got three D models coming video models. Ultimately the whole thing is that there's no barrier to creation when you have this next generation of models from a big data age. We took big data extrapolated for a specific models. These very general models to understand principles like understands the word supercar SAn Francisco. We talked about jeremy maybe it needs more than four images. We have jeremy, right? And this is the next stage in AI and the next stage in real usability of this technology and for you guys the stability, what's your kind of business model going to be going forward, Every single creative industry in the world is gonna have to react to this because like I said, as of next week, this is 30 images a second, which is basically video brand new experience where you can create anything dynamically. We're going to go to them, take their exabytes, petabytes of data and convert them into models because again, 100,000 gigabytes to 1.6 gigabytes. Something quite crazy. Yeah, absolutely. Um I said I wasn't gonna take questions in the audience but I think I will because we have a little bit of time. Who has questions for Ahmad? Anyone have questions right down here in front, wait for the mic to get to you. Please please stand and stay here you are. What would you say to you can you can you are my name is Ian you tilly attention life. What would you say to creatives because this can either augment a creative development of what they do or they can kind of feel like I know that you can take my job. So on the optimistic purse, how are you encouraging creative and creative teams to leverage this technology to do what they do better. So I think this thing, what is creativity? Right? So this can match together concepts. So we still have an iterative process to get there. There's just one part of the process. So the ideation phase, the fine tuning phase, the kind of like we have text, image, image to text in painting so I can remove, I can make that car blue and it'll make that car blew in the next version. You forget how to integrate this into your process is just as you did with after effects Photoshop, a lot of these other tools because it's just a tool, it doesn't do anything by itself. So what we found is the best artists got even better and this is what we saw with and there's an Ai that beat at least a dollar go who is the best go player? The average go score, it looks like this over time. Is that because they all got better because they use the technology as a tool to extend and embrace it because it's here and you should use it right now and then you'll outperform all of your competitors. Right. Other questions, we're here. Hi there, Ahmad Rishi Garg, a partner at the Mayfield Fund. Um, uh, you've got a lot of other projects on the website that you're working on, you've obviously started with Image generation. But one of the things that's captivated everyone's mind is text text threading and text generation because of the broad applications. Could you talk a little bit about the future horizon for stability and all the different applications you want to build into and how and how kind of starting with image Get you to a place where you can do multimodal applications over time. So that's excellent as a business perspective when I saw the image I saw as we're going from a 10% to 90% like if I show you this a year ago, it's impossible, right? And the fact that it's real time now is also impossible. So you wake up and do some impossible things before breakfast text. By contrast we got 85% to 95% there and I struggle to see where the edge is going to be when you have a chat bots that can do anything like use chat. GPT somewhere, you know it's fantastic. All the Ai models will be like that. Whereas an image I saw that there are giant asset stores of I. P. That need to be converted because you always want mickey mouse or halo or any of these other things so we can create everything, something becomes valuable but it's not enough just to have image cause image will be standardized instead. What we're doing is we're building the ready Player one Oasis slash Holiday where anyone can create anything they can imagine. And that's not only media, no more power points, no more slide creation in a few years. Just describe it and make it more peppy, make it nicer and they'll do it automatically recreate game of Thrones season eight because it was a bit rubbish. You know these types of things is what we are enabled seamlessly. And you talk a little bit about how you build stability itself because I think a lot of people are surprised you guys almost came out of nowhere with this fantastic technology. I think there are a lot of businesses out there where they, you know, they talk all the time about the struggle they have, hiring ai talent, you've obviously been able to build, you know, really cutting edge stuff. Um how did you do it? So we built a community when some of the large open source models were open source claiming the big entities could do it and I looked at whereas the arbitrage properties and the gap, the gap was super compute data and talent. The talent came from our communities and we hired primarily from our communities. The 1st 96 employees, only one PhD So differentiated talent then data. We built large open source datasets like Lion, 5.6 billion images. The previous large, 100 million to power this. And then finally, Supercomputer worked closely with amazon to build the 10th fastest public Supercomputer in the world. And so that was overwhelming amounts relative to what happened. So 4100 fast in the UK was 6 40. And that unlocked all of this because we allowed people to experiment and play and figure out and make mistakes that then optimized for this crazy model right here, but also other models that were released in our generation to the new language, ones to the others. Because you need all the components all in one and that's why I created stability to be like this foundation layer for next generation generative AI. And you you now have a sort of a partnership with AWS, is that right? Yes. So this has been a long term partner now, it's accelerating up, we're gonna power generative AI through kind of amazon and make it available to everyone. It becomes a sort of differentiate for different peoples clouds. Obviously has got their relationship with Open AI AWS as you guys. It's kind of interesting. Exactly. You know, and there's different approaches like quite integrated, very close to Microsoft where we're just trying to get this out to the entire world, right? Thank you so much for joining us. This has been fantastic. Really exciting stuff.

Colin Murdoch, Chief Business Officer, DeepMind
Siraj Khaliq, Adviser, Atomico
Moderator: Jeremy Kahn, FORTUNE

Brainstorm A.I. 2022 - Out Of The Lab

Naba Banerjee, Director of Trust Product, Airbnb
Abhijit Bose, Managing Vice President and Head, Center for Machine Learning, Capital One
Nicole Eagan, Chief Strategy and A.I. Officer, Darktrace 
Moderator: Verne Kopytoff, FORTUNE

Brainstorm A.I. 2022 - In A.I. We Trust Shaping Confidence And Safety

Naba Banerjee, Airbnb
Abhijit Bose, Capital One
Nicole Eagan, Darktrace 
Moderator: Verne Kopytoff, FORTUNE

In A.I. We Trust? Shaping Confidence And Safety

Brainstorm A.I. 2022 - In A.I. We Trust? Shaping Confidence And Safety

Reid Hoffman, Partner, Greylock; Founder, LinkedIn; Co-founder Inflection 
Saam Motamedi, Partner, Greylock 
Moderator:  Jeremy Kahn, FORTUNE

Brainstorm A.I 2022

Brainstorm A.I. 2022 - Big Data Vs. Good Data

Latest video

Andrew Ng, Founder and CEO, Landing A.I. Interviewer: Jeremy Kahn, FORTUNE

Videos (17 videos)

Brainstorm A.I. 2022 - Big Data Vs. Good Data

Brainstorm A.I. 2022 - Robots Around The Water Cooler

Brainstorm A.I. 2022 - Building The Future We Want: A.I’s Human Factor

Brainstorm A.I. 2022 - Salvador Dali Meets Wall-E

Brainstorm A.I. 2022 - Can A.I. Help Solve The World’s Biggest Problems

Brainstorm A.I. 2022 - The Age Of Intent: Knowing What Customers Need Before They Do

Brainstorm A.I. 2022 - Reinventing National Security: A.I. And Defense

Brainstorm A.I. 2022 - How To Turn Good Products Into Great Ones With A.I.

Brainstorm A.I. 2022 - How A.I. Is Spreading Throughout The Supply Chain

Brainstorm A.I. 2022 - View From The Top

Brainstorm A.I. 2022 - Leading With A.I.

Brainstorm A.I. 2022 - A.I. And The Promise of Large Language Models

Brainstorm A.I. 2022 - A.I.’s Role In Shaping The Metaverse

Brainstorm A.I. 2022 - Generative A.I.: Silicon Valley’s Newest Gold Rush

Brainstorm A.I. 2022 - Out Of The Lab

Brainstorm A.I. 2022 - In A.I. We Trust Shaping Confidence And Safety

Brainstorm A.I. 2022 - The Next Billion Dollar Ideas