EP #39 Should We Be Fearful of Artificial Intelligence? The AI Panel w/ Emad Mostaque, Alexandr Wang, and Andrew Ng

0:00We're all gonna have some version of Jarvis, right? An AI that is in your ear, on your body, and so forth. Frankly, even if someone in AI, I feel like, man, it's tiring. There's so much going on. Physical things in the real world are much further from being disrupted by AI. And it's gonna create wealth faster than anything we've ever seen. [♪ OUTRO MUSIC PLAYING [♪

0:25The beauty of AI is that it can analyze information faster than any person. We're just literally starting to explore these. All of us are engaged with governments. We're trying to help them. It's a bit difficult when they're just catching up to the internet right now. Now it's got to the point where this technology could be potentially dangerous. So not everything needs to be open, but we do need to be transparent and careful when stuff gets out of our control. After I've met, I'm not letting Chaggit be tell me where to invest at this moment. In school, a lot of kids have been cheating and they've been using Chaggit to write essays and book reviews and do their math.

0:57Cheating implies as a contest. School should not be a contest. You will never not be without the say -i as you grow up. My question is about content creation especially in music. You will have perfect music models by the end of the year. We believe this technology could be existential and we will treat it as such. One of the greatest drivers for peace on the planet is making sure every mother has their children with the best help and the best education. A more peaceful world for us is a world where you uplift everybody, which is what this technology has to do to deal.

1:31Everybody, this year at abundance 360, given the meteoric rise of AI, I decided to put together an entire day on AI with three of the most extraordinary thinkers. You've met one of them, Eamod Musdak, the chairman, founder, CEO of Stability AI. I also brought to the conversation Alexander Wang, who's the founder and CEO of Scale AI. It's a big data machine learning platform that accelerates the development of AI. He founded it when he was at MIT at age 19 as a student, dropped out after the first year to become the youngest self -made billionaire ever. I also brought Andrew Eng, the managing partner at AI Fund.

2:12He's co -founder and chairman of Cousera, the founder of Google Brain, I love the name of that. And one of times 100 most influential people between Emoid, Alexander, and Andrew, you can get an overview of this field, how fast is it moving, how disruptive is it? Is it something you should be excited about or fearful of? My massive transformative purpose is to inspire and guide entrepreneurs to create a hopeful, compelling and abundant future in humanity. And that's what I'm doing with this podcast, to open up every relationship, every conversation I have with you, to inspire you, to support you in going big, in helping uplift humanity.

2:57If that's of interest to you, please subscribe to this podcast, allow me to share with you the wisdom that I'm learning for the most incredible moonshot entrepreneurs on the planet. Let's jump into this episode. It's a conversation that everyone needs to be having in your company, in your family and Definitely at the heads of every nation Enjoy Thanks Peter and thanks to our amazing guests here today We've been talking about this actually for years when we look at the rapid acceleration of AI and This this disruption that's kind of happening very very quickly from this point forward and And when we look at industries and workers and whether they're getting the understanding, the knowledge and the rescilling that needs to happen to approach this current world, in what ways do you see or do you believe we can do that faster and work with governments and larger corporations and education systems to help enable this.

4:06Because we're all talking about positivity and abundance. I think we're very privileged in this room. The question is for the rest of the world, what are we doing? And in what ways are you involved as leader? By the way, we've got 40 minutes here, so lots of great questions. We asked who wants to take that on. You know, so is very important points. I think the world is changing so fast. It's difficult for a lot of people to just keep up with what I don't know if it's going on. Frankly, even as someone in AI, I feel like, man, it's tiring. There's so much going on. So I think maybe one of the things I'm doing years ago, I started co -founded Coursera, which to this day offers all of courses, often works with governments to train up different people.

4:46I think that one nice thing is with online education, with digital education, we can speed up the rate at which we hope people figure out, learn the new technologies and figure out what to do with it as well. And I think Meteor also has huge important roles to play. And eventually, I would love if we can build tools. Like, if you look at Google, the UI is really simple. The backend is insanely complicated. It's just type what you want. You get answers. So I hope that AI tools can evolve to the point where, you know, you could just use it without nearly as deeper education needed. And I think a lot of the generative AI tools are hidden at the same time.

5:22Yeah, I think it's super natural. I mean, that's why super space natural. like Chatchy PT was so successful because you just type and it just came. Stable diffusion, just type and then have images, right? And so I think making these things more and more usable as led to the explosion because it kind of comes from the natural language understanding of humanity itself. It's trained on these large corpses of an entire media set. I think on the policy side, I think all of us are engaged with governments. We're trying to help them. It's a bit difficult when they're just catching up to the internet right now.

5:54But by the way, the most linear organizations on the planet, right, if not sub -linear. They are. And again, the typical approach to this is where we get minimax, you know, like stop it, ban it. But it's inevitable now. And again, it's kind of a global phenomenon. One of the problems as well is that, you know, you've had an acceleration but you haven't had a maturation. So there's lots of excitement. How much implementation is there now? There's going to be a bit of a pause because it has to be in the current wave, even as research keeps going and you go into engineering and application and so I think we're not sure exactly where things are.

6:26I think Andrew talked a bit about where the value ladders on the stack and again we are all kind of at different areas of this stack so we ourselves haven't figured this out and we're doing our best with the community to try and do that to communicate the policy makers and make education courses and others accessible. Michael, good to see you. You're next pal. So this is probably a question more to M .I. but anybody's perspective is welcome. I am done lots of programming and usually like to think about abstractions or concepts. You have a concept, let's say, draw a bullet and fit for some purpose.

7:01That's what it can do that thing. Now with AI, we are using English language, we are using Chrome, so definition of what you are asking for is a lot more vague. So when it's vague, it may have edge problems. You have to clarify things a lot more. And so my question is, do you still see a value in having well -defined structural uniquely named things which you can compose for well -undiced to the result? Or do you think we'll diverge more to everything is a bit fuzzy, everybody just provides a lot more text And it just somehow the system figured out for you. Well, I think a section of question.

7:45I think a lot of the applications right now have been very surface level without the contextual awareness of like, you know, Game of Thrones in South Korea or something like that. Amazon had this thing where they had the six page memos, right, rather than presentations. But another thing they used to do was write the press release beforehand. And so if you write the actual overview of the program in the comments now for kind of co -pilot and things like that or a GPT -4, it will write the program based on the comments. You know, so again, I think this is a failure for us to understand prompting and some of these other things.

8:16There's something that Andrew and I have been talking about. You know, again, it's gone so fast we've got the surface level. We have to explore. Another way to think about this is like game consoles. When the Wii first came out, games were a bit crap. By the end it was fantastic. like we're just literally starting to explore these. So it's simple words now and then we'll compose and build structures around it to get the most out of these weird thingies. Everybody, I want to take a quick break from our episode. Tell you about a health product that I love and that I use every day. In fact, I use it twice a day.

8:44It's called seed health. Now your microbiome and gut health are one of the most important and modifiable parts of your health plan. Your gut microbiome is connected to your brain health, your cardiac health, your metabolic health. So the question is what are you doing to optimize your gut? Let me take a second to tell you what I'm doing. Every day I take two capsules of seeds daily, symbiotic. It's a two -in -one probiotic and prebiotic formulation that supports digestive health, gut health, skin health, hard health, and more. It contains 24 clinically studied and scientifically backed probiotic strains are delivered in a patented capsule that actually protects it from the stomach acid and ensures that all of it reaches your colon alive with a hundred percent survivability.

9:30Now if you want to try Seeds Daily Sinbiotic for yourself, you can get a 25 % off your first month supply by using the code Moonshots at checkout. Just go to seed .com, backslash moonshots, and enter the code Moonshots at checkout. That seeds .com, backslash moonshots, and use the code Moonshots to get 25 % off your first month of seeds daily symbionic trust me your gut will thank you all right let's get back to our episode what are going to zoom next Ned all say coffee Ned where are you on the planet and what's your question? Hi guys can you hear me okay? Yes, we can hear you great Perfect I'm Chicago.

10:11I am in healthcare and I love the idea of healthcare and longevity I maybe you're only practicing physician here I have the idea of joining this group because as a practicing urologist we are getting humbled and as Peter and AI and at HealthFan go on, most urologists are going to see people who are older. I think all you guys are going to be a patient of either me or my colleagues here soon. The point I'm making here is that there is an overflow of patients and weas clinicians are getting bombarded and the question is, in light of workflow shortages with our staff and the incoming number of patients, how to use AI most efficiently in terms of point of care for patients getting into our offices, as well as having AI be part of the evaluation process because as many people may know, 90 % of the time we see the same 10 conditions.

11:12So the question to the panelists is number one, how would you advise a physician or healthcare provider to set this up and do you think that this is something that is easily attainable? I liked Alex's thought about starting small and then the second question is, if you were to do it, is it the sort of thing that would prevent, I mean, is there anything in there to prevent others from doing the exact same thing so that if you are spending a lot of time and money into this that nobody else just kind of scoops it from you? Thank you for taking my question. I'll just say real quick, if you have a concrete idea for what you like to do in neurology, let me know.

11:54I think teams could help, maybe my team could help execute that. One example, something AI funded, we were a fantastic founder, Alison Darcy, and now, different CEO, Michael Evans, support building up WorldBought, which is a digital mental healthcare chat bot. You could actually installed on your phone that from the data published understand that seems to be able to take effect relatively quickly in terms of treating symptoms of, for users of symptoms of anxiety and depression. So I think that there are actually lots of opportunities to apply AI in interesting ways to healthcare. But figure out those concrete use cases that allows us or others to execute on it.

12:33Yeah, I'll say two more things, which is that I think the first is that getting to market first is actually an advantage. You know, certainly other people can copy the idea, but I think, you know, this is one of the things that's well studied in business, but there's going to be a mode from whoever is able to establish the use case first and then able to use that pull position to get more and more data into the system. So let's say you built, you know, the best possible system for identifying some sort of, some sort of disease or condition using AI, then you're going to just by launching that service get far more data than anyone else and race ahead of where anyone else could possibly be.

13:12So the currency of the realm of AI really is data. And if you have a niche dataset or a unique dataset, that's going to give you an initial advantage. And if you're able to build a strategy, continue and ask more and more data, that'll keep you ahead. We're going to go next to John here at Mike too. Let me remind you please look at the questions in the slide. I'll vote them if you're here in the audience or on Zoom. We'll be pulling them. John, please. What's your question? And keep the question short if you could. Sure. So I'm a medium to small business owner. I'm a physician. I own a diagnostic laboratory.

13:47So I hear a lot of buzzwords. You need to implement AI. You need to create, you know, you need to put your dad on a balance sheet. That sounds awesome. But my initial question is how do you get started going down that road of, you know, we have a ton of data, but how do you put it in a package that you can actually monetize it, put it on a balance sheet? So, what are the process of just going about doing something like that? And then number two, implementing AI is something that I think is a very concrete, like like a case point for our company is digitizing pathology and allowing an AI over engine to help pre -diagnose or triage cases or whatnot.

14:28How do you go about implementing something like that in a company that's my size by either pairing with someone else or do you try to bring a team in to develop it? So what kind of playbook just to get you off the ground and move your company? I think you're moving that. Yeah, so we have tools. You're a scale .com. We have data set management tools that you can upload all of your data and basically build that into a data set. That's step one is to get all of it into one tool, build up an entire data set, and then you can do a bunch of things with that. You can label and annotate the data, turn into a data set that you can train a new model on top of.

15:00You can bring in a host of companies, we could be one of them, or there's a bunch of other vendors out there that can take that data and turn it into a customized model for you, and then launch it and figure out how much impact can you possibly have. And you find out who that data is viable to. Ultimately. We're going to go to Slido. There's a great question here. I'm going to send to you, Imad. It says here, is there any commercial activity which AI will never be able to disrupt? What will be the factors to understand that? I mean, it's an interesting question, right? It's like the question asked of Bezos.

15:35What would change? There's nothing that will never be able to disrupt it because you'll have autonomous robot agents that basically has a glater runner, right, in strungable from humans. So I can't see any commercial activity, you cannot disrupt. Do you guys agree? Alex, go ahead and go with you. You know, I think at least in the short term, and who know, this stuff could all change. But in the short term, physical things in the real world are much further from being disrupted by AI. So anything that we have to interact with or do something physical in the real world, robotics is relatively behind a lot of the pure digital, pure online AI systems.

16:21So I don't know if the word is never, but at least for a long time, we won't have an AI robot that can do construction or that can do mining or some of these things that are very interesting. I go into the I go into E -Mod's category here, but that's my opinion you guys are actually Andrew. My friends and I, some of my AI tech friends, I used to challenge each other, name an industry that AI will not be the disrupts, we challenge each other to name that, and I had a hard time coming with them until one day I thought, all right, maybe hair dressing in this tree, you know, hair dressing. Hair dressing.

16:58I think that's easy. here. And I used to say this on stage until one day, one of my friends, who's a robotics professor, she was in the audience when I said that on stage. And after she stood up, she pointed at my head, and she said, Andrew, for most people's hair cells, I don't know how to put a robot to cut a hair like that, but your hair's size is so good, we're all going to do that. Oh, that's great. All right, we're going to go to Kiran and Mike for. Do you think with the large language models like GPPT will they ever be able to explain why they generated what they generated? So right now you can ask the models to explain why it said what it did.

17:40It's not clear if those answers are actually correlated at all with the actual reason why they aren't, probably, correlated with why they actually said that. And so there's sort of, you know, humans of this bug, which is that we will do things. And then if somebody asks us to explain why we did the thing, we'll come up with some post -hoc rationalization that oftentimes is not literally the reason why we did a certain thing. Usually it's, you know, there's the analogy of the elephant and the elephant rider, which is that our emotional brain is an elephant and our rational brain is this rider of the elephant that just sort of is trying to explain what the elephant's doing.

18:15So I think the large language models are similar, which are that they're going to go through process by which they make a decision or they say something and then you can ask it to explain and it will come up with something on the spot to try to explain it. You know, there's something that a modern I work just talking about is like how do you get greater levels of interpretability, how do you, how do we do research in the direction of like actually truly understanding where it's coming from, but that's all active research. Nice. Jacelle, you're the next contestant, good to see you.

18:42So given what has happened to social media, I actually rated the way I feel about AI I scared chitlas. Because you guys, you are the leading edge of this, and this is your space. Is anybody doing anything to build parameters around this technology so that what you were saying you met about someone calling you and your mother's voice and selling you stock? That is, we already live in a world where we don't know what truth is, how are we going to fight this? Well, I think you see information curation going through social networks. I mean, this is the Twitter verified, it's Apple Identity, it's kind of a bunch of other things.

19:32We bank content authenticity .org, which is a diabetes thing around visual media, but we do need to have centralized repositories of trust, because the cost of information becomes comes zero and creation becomes zero. And unfortunately, there is nothing that can really regulate this output, like there are laws in various countries that you must identify. Our output was stable diffusion, there's an invisible walkmarked of things, but it's become so accessible that we need to have standards quicker than the system is adapting. So this is why I think information distribution systems need to standardize around this right now, which is something we're pushing a lot.

20:07Yeah, I mean, there's a big question, can we even govern against any of this, right? All of this is bits. We live in a world of poorest borders, data flows. So, at Giselle, we're going to have a lot of these conversations around ethics and around implications. Guys, this is happening right now, right? This is this year, next year, not 10 years from now, five years from now. What gets played out, this is why we need to be paying such attention to it. Giselle, thank you. I'm sure we'll hear from you a bunch more times. Let's go to mic number three in the back there, and then I'll go to Zoom, and we'll go to our slides.

20:50My name is Hajime, running the AI startup in Japan Series D, now getting much closer to the Japanese government, and they're concerning about the disaster and resilience, because there's a huge risk on the earthquake, as well as the font -whosey erosions. So my question is, what could be the interesting, but the practical way to apply AI for those disaster recovery? So we actually worked on something in this direction for in the Ukraine conflict, which is obviously been this horrible thing to happen to a region in a country. And you're able to, one of the things that is really unfortunate in any sort of disaster situation is that seconds really matter.

21:37Every second spent that you aren't deploying resources to help resolve one of the people who are in a damaged structure or tackle a certain area is going to result in other lives lost or further damage of same infrastructure. So one of the things that you can do with a combination of satellite imagery plus drone imagery on in major cities or major areas that are being impacted is use AI to automatically identify damage on a building by a building level and a change in that damage in every area and use that to coordinate humanitarian response. So this could be applicable, you know, this is applicable in any sort of natural disaster, any sort of scenario where fast action really matters, the beauty of AI is that it can analyze information faster than any person.

22:24Yeah, we're doing a very stable diffusion for satellite imagery and time series data as well, and so that I'm sure we'll be using that toolkit. Yeah. Yeah. Amazing. All right. To our Zoom audience, we're going to you next, Nico. Nico, Drinek, where are you on the planet and what is your question? I'm Peter. I'm calling in from Austria, from Europe. Beautiful. I have a question regarding the profile of a GFI officer. So what's private? Profits should this AI officer have to fulfill this for best? should it be technical, should it be another profession, should it be a mix? Because it seems to me that combines a lot of traits that are found in and for the nursing CEOs anyway.

23:08So what else is needed? And I asked this specifically coming from a non -tech traditional background as long as I can. Great question. Let me prep us a second. What my advice as a first step for folks is I like the idea of get your team to experiment and so forth. But my, what's been working for me is finding somebody who knows the field out there understands who to partner with is advising as a strategic advisor to the CEO about the platforms and such as a chief AI officer in that regard. Do you guys agree with that idea and what would be your, you know, what do you think that person should have is background?

23:48So I think I might have been the one that coined the term chief AI officer. I stole it from you. No, but so actually years ago, actually wrote an article in hobby business review. I remember googling if a chief AI officer didn't really exist on the internet. So I actually wrote a piece in hobby business review with my specific recommendations to the chief AI officer profile. Maybe a person needs to be technical enough to understand the tech. And then also business oriented enough to work cross -functionally to figure out what are the valuable business use cases for your specific application. But I have like a multi -page article on HPR talking about.

24:23I mean, I would imagine the CEO or the store, the head of whatever division says, this is what I think is possible. This is what I need. The chief eye officer is sort of an interface to the world out there to bring in a scale, to bring in part of what EMAID, what you're building, is that sound like a reasonable role in a company? I think I'll have to be technical enough to make good judgments and also to make the bi -versus build decisions and often for many companies you should buy a lot and build a little bit. To make those decisions requires both the technical judgment as well as the ability to figure out, understand the business well enough to figure out the use case.

25:01You might, I can't tell if you're agreeing or disagreeing. No, I think so. I think that there's an implementation side of it like an applied engineer is a good thing. I think passion is kind of key. Yes. Because you have to throw yourself, otherwise there's no way you can keep up. Actually my suggestion would be that given this suggestion that you have, there should be an internal A360 repository around the ecosystem, the latest trans and others that can go to all the chief offices here. They can keep on doing that. All right, Steve, you heard it. You're in charge. Wonderful. Thank you. I hope Nico that answered your question.

25:39Let me go to slide on next and thank you. Great questions. Katie, I apologize. I didn't note your name earlier. What is the first step companies can take to evaluate the data they have? How can we get started extracting from the systems we have in place today? So I think the first thing to do right now, the new large models are quite capable of just sort of ingesting in existing data you have. And basically whatever format you have, in some cases it can be literally as simple as uploading a bunch of PDFs into the models and sort of extracting all the information from there. So, by the way, something you just said is very important, right?