Europe Virtual 2024
Q&A with Dr. Ethan Mollick, Author of "Co-Intelligence"
Q&A with Dr. Ethan Mollick, Author of "Co-Intelligence"
EM
Ethan Mollick
Author of "Co-Intelligence", Associate Professor; Wharton School of the University of Pennsylvania
Transcript
All right. Fantastic. So on the topic of, uh, generat ai, if you have any interest in, uh, ai, you probably followed the work of Dr. Ethan Malik, currently associate professor at the Wharton School. Uh, and at the, uh, university of Pennsylvania. I was such an avid reader of his work, uh, either on Twitter or on a substack. Uh, one useful thing, uh, I fell in love with his writing because it was so clear that he was on the frontier of exploring how so many different parts of society should or could adopt, you know, these seeming miracles afforded by ai. And so, first off, my hardest is congratulations to you, Dr. Malik, for releasing your book Co Intelligence, uh, which instantly became a New York Times bestseller <laugh>. I absolutely loved the book, and I think it should be required reading for everyone, especially, uh, our business counterparts.
Uh, so I got to meet, um, Dr. Molik when I was invited by my friend James Cham, to an event, uh, before the book was published, and it just blew me away. And so, given that this community is being asked to lead AI and engineering initiatives, um, for their organizations, I thought would be so incredible to have him share his perspectives with this community. And I'm so delighted that he said yes. So, I've introduced you in my own words. Uh, Dr. Ma, can you introduce yourself and what, uh, what you've been working on that you've been having the most fun with these days?
Oh, it, it, it's, it's exciting. Thank you for introducing me. Um, and thanks for the kind words about my book, which is, which is right here. Actually, the funny thing is with writing a book like this is, you know, I had to wrap most of it a year before it came out, or, you know, six months or so. So I feel pretty good that I imagine, and people are like, it's a really good introduction to, to ai, and I'm like, if I imagine now that it's a good introduction, that means I was pretty far ahead seven months ago, you know, before it came out. So a good sign. Um, I'm a professor at Wharton. I, uh, study innovation entrepreneurship, but I've been working for a very long time on using AI for things, uh, since working with Marvin Minsky at the media lab. But I've never been the technical person. So I've always been the business application use and education person, which is why I think I've become useful because I think about how this stuff is applied and what it matters. Um, what I'm having fun with right now is, uh, I think the same three things are the really interesting trends in ai, which are, um, like, uh, specialized devices. This is actually an AI in a box, um, that I've been playing with. Um, and then, uh, large context windows and agents. Those are, those are the interesting things for what's coming up next, I think.
Oh, super, super. So, you know, you said something, uh, in your book that, uh, struck me as really important, uh, not only for technology leaders, but you know, technology, all the leaders that technology leaders interact with. You wrote, people are figuring out ways to make, uh, use AI to make their jobs easier and better. The results are often breakthrough inventions, ways that, uh, AI could transform a business entirely. But <laugh>, the inventors aren't telling their companies about their discoveries. Instead, they're keeping them in secret, so they're hiding it, uh, from people. So, can you talk about this phenomenon, why you think it happens and what should leaders be doing to encourage and celebrate these type of innovations so they can actually take advantage of it?
Yeah, I mean, it's a, it's a really interesting problem. So, I mean, we've always known the innovation comes most from users of technology. 'cause they're the ones with the need, right? If you are, if you are this, you know, CTO, you have, uh, you, you know, you wanna see your company be more efficient, be better using, if you are someone on the line who has to write an email message every day to hundreds of people, it's very cheap and easy for you to experiment about how to write a better email message. And you're highly incentivized to do that. Well, the CTO is incentivized to try and figure out a general solution, but not to solve the problem. So people experiment all the time with their own work, um, which is just a universal. And so everyone's using LLMs to do their work and they're just not telling you about it.
So, uh, one of the most universal things I see is that everybody's using LLMs to do all their performance reviews of their employees, which is obviously exactly what you don't wanna see them doing, but absolutely everybody is doing that. Um, and so I think one of the things that I have, um, that I've been noticing is like this, this is when I talk to people privately, they tell me they're using AI for everything, especially multimodal to take a picture of the screen or whatever, and then ask the AI to do the work. I spoke to someone at a large bank who wrote the policy to ban chat GPT use, and she used chat GPT to do it. Um, and so, um, so there's really, there's really a hierarchy of reasons. The first reason is unclear rules. You know, especially the rules are like, people don't know whether they'll be punished or not.
A lot of companies are either not embracing it or if they're bracing it, they're bracing with like all sorts of caveats. If you use it incorrectly, you'll be punished. Um, and no one knows what correct or incorrect is. So that's the first reason. The second reason is Reddit's full of people who talk about how they're now viewed as wizards because they can get 10 times the amount of work done. Why would they wanna show you that that work is being done by ai? Third is if they do show you, are they gonna get fired because that you realize they don't need as many work. We just, I just call it the tail end. That really interesting thing on customer service. I mean, if you're a customer service agent, why would you tell anyone? 'cause all you're gonna do is show that your job might be not useful. Or even if maybe even if you not worried about that, maybe you'll, you know, you'll end up being in a situation where you don't get recognized for the extra work or you just get assigned extra work to do, or you're just better launch a startup anyway. So all of those overlapping reasons make it really hard for people to be willing to talk about what they're doing.
Yeah. And, and so like, what would your advice be, uh, you know, to, to sort of break that, um, uh, you know, terrible conditions, right? Where, where we're getting the opposite outcomes of what we want.
So I think part of this is about, um, your company culture that you already had, a company that has a sharing culture where you trust the administration's not gonna fire you, is going to be very different than one where it's a competitive culture and, you know, people get laid off all the time. So you have to live with the culture you've got, first of all. So everything we know about building a good culture matters. And then I think you have to radically rethink incentives. What is the incentive to not to show, right? And so I have, you know, I have talked with companies have done some fairly radical things from, you know, giving $10,000 rewards at the end of every week to whoever comes with the best prompt to, you know, to thinking about how do we, you know, to do We promise we'll never fire anyone for the next year because of generative ai. Like, I think you have to be realistic about the incentives people have.
Yeah. And by the way, one of the mind expanding things that you put in your book was like, uh, you know, some of the bonuses being like, uh, up to a year's salary, which I thought was just, uh, just awesome, like very mind expanding
<laugh>. It's, well, it's funny, there's a, there's a old, there's a old science fiction author named Robert Hyland, and that was one of the things he had said, his book, he talked about his book to have people build robots in one of his books of like, how us got an economic boom was if you replaced your yourself at your job, you were paid that job for that job for life, and then ed people to kind of do the work.
So good. You, there was another comment, uh, that you had made at James' event that sort of jolted me. Um, now you said something along the lines about your fears, about the use of retrieval, augmented generation. In fact, that was one of the techniques used in the, uh, uh, previous session that you caught the tail end of. Um, because this is, you know, absolutely one of the most widely used techniques, uh, in, uh, AI where we feed source documents, uh, to create things like chatbots. But you mentioned how you were, uh, skeptical about this practice because of how inherently AI makes stuff up all the time. And, uh, it's difficult to, uh, you know, detect these confabulation, you know, effectively. So can you talk about why you may, I'm not sure what, if the word is skeptical or, or like what your concerns are, uh, despite the softer industry putting, you know, massive bets on this technology, you know, hoping to make it viable and some of your surprises of like, biggest surprise of like, rag gone wrong <laugh>.
So, so I think, so I think Ra there's a few things that I worry about with Rag, right? Um, the, the big picture view, you know, lemme talk about the issues, but the big picture view is there's also a huge amount of bets going on, on a current technology that I see even companies with really smart people making, which is like, let's bet on this LMS the way they are right now, right? Let's bet on a cost structure. We have to minimize things and throw stuff off to a llama two, you know, instance running locally that, you know, and how do we, what's, how do we minimize costs when costs are still dropping exponentially and ability is still increase exponentially? And, you know, we're still seeing those changes, right? The, the, the rock hosted version of Llama three is like insanely cheap at this point.
So like, there is a lot of people betting building infrastructure around the limitations of a technology where they're gonna release the product in six or eight months, and it's already gonna be, you know, all of the decisions you made and constraints are already gonna be an issue. Um, and there's also a mindset issue, which I think that a lot of companies are used to thinking about scale as something where we have to, where cost is the number one thing to do, as opposed to thinking about now you have a cost ability trade off that's very direct. But leaving aside those sets of issues, I think the bigger issue with RAG is that a lot of technologies I talk to don't seem to understand or don't fully absorb the fact that you can build the world's best rag pipeline, but once stuff is handed off to the LLM, it does weird things.
And those weird things get compounded with how users use the system. And so, for example, a really reasonable thing that a user might ask the RAG system to do is tell me what's important about this project I'm working on. Well, I use, I will not name it, but I've been using one of the best, uh, systems that works with rag, um, you know, as, uh, in your documents. And what it does is the rag search, when I asked for the most important thing ends up doing, pulling back a few documents. I don't know why, but in the case of my, my organization, it pulled back a, a Salesforce installation guide, a document I wrote, which was great about a paper and then a memo from the dean of our school. And the AI then made an incredibly plausible argument about why these three things fit together as the most important thing that I need to be worrying about.
Like, we obviously need to be focusing on the Salesforce installation and, you know, this is a priority for the dean. And of course, Ethan's work indicates why this is such a big deal. And it was completely plausible. And, you know, there were some quotes made up for my documents that seemed really reasonable that I had to read my document to remember I didn't write them. So the problem is you have an absolutely convincing, you know, machine at the other end and it, you, you could deliver the perfect documents to it, but it might still not just hallucinate, but but make up useful information that isn't valuable. And I don't see people's rag pipeline testing that end user piece nearly as much as they test all the elements going into it. And I think that weirdness makes rag harder to use. And then the last piece, because I've talked to great length about this, is you get much less hallucinations and much smarter AI reasoning over large context windows. So one other option is like, when is it just worth loading everything into the context window? And everyone's like, well, it's very, very expensive to do that right now. Yes, for now it's very expensive to do that. Is it expensive to do it 12 months from now? Is this what you wanna build a product around? Is a solution that is built around today's limitations for technology that, as far as we could tell, is still doubling capability every five to 15 months.
Yeah. Fantastic. And by the way, I, uh, realize as I was, uh, choosing the word skeptical that, you know, by no means am I trying to diminish the work at, uh, our friends at parlo, et cetera. No, it's
Great. I mean, I totally get it. I mean, I, I think RAG is rag will have value, but I worry that as a general solution for solving every problem, you know, I, I always like, first of all, the, the core idea, even that your own documents matter so much for most organizations is an open question. I don't see testing what's already in the GPT, like in what retrieval, you know, what, what your context window stuff does. There are definitely use cases for RAG and I talking, you know, the customer service thing. Maybe the may be a perfect answer 'cause it's bounded, um, and they think you have the right kinds of issues, but as a general solution for it to everything, which is like RAG will solve the problem. I have my doubts.
Yeah, no, it's so interesting. I I wanted to bring up that question just because I think it, it is of value to anyone who is influencing or implementing, you know, bets on, uh, ai and I guess I actually relished the engineering problem of like, how do you sort of ensure the, the, the responses are correct? I mean, as an engineer, and by the way, uh, one of the things that came up in that, uh, James Cha session was, uh, uh, the notion of like, uh, you said humans are good at detecting type one errors when the computer doesn't find something we're looking for, but we're much worse at detecting type two errors when a computer returns erroneous information, which I thought was such a, an interesting insight.
Uh, absolutely. And we, and we're just not used to it either, right? Like if your search box says answer not found, you're kind of annoyed, but you're not, like, that answer can't exist anywhere. But if you get it competent answer every time, and again, like the AI is a person figure instructions, it's very hard with it to bound what a person does. In the same way it's gonna be hard to bound what a LM does, what it's most useful. That doesn't mean you can't force it down a narrow pathway like the customer service example we were seeing, right? But what the real power of these ais for transformation inside organizations is often treated like a person that does analysis and extra work that is not as easy to do.
Oh, I love that. In fact, in your book you called it, uh, uh, cyborgs and the the Cent. Yeah. In fact, yeah, let's go there. So, uh, this is an audience of technology leaders often leading tens, hundreds, or in some cases thousands of software professionals working on some of the most important initiatives, uh, in their organization. So, uh, like let's zoom way out. What advice would you give to this community to help them and their teams take maximal advantage of this incredible technology that's impacting all of us to help them thrive and win in the marketplace?
I mean, so the first thing is, uh, for goodness sake, just use it. Like, the thing that worries me most is when I meet organizations where people are not using AI or they're only using it for coding, um, because I think that's a very narrow viewpoint. Uh, there is a basically a parallel world of AI as coding tool, um, and ai, you know, and that, uh, that doesn't, that coexist very loosely with the larger AI use by everybody else who isn't a coder. Uh, so you have to use it for everything. That's one of the principles of the book. Invite AI to everything you do. I think you absolutely have to do that. So you have to be really, like, you, you have to use your 10 hours of, of whatever frontier LLM you're using. And by the way, if you're using LLMs at any part of your work, you should be using that LLM too to understand what it's good or bad at when it gets mad at you.
Like that set of stuff. I think the, I think the second thing to realize is that you have to build for the future, um, even if G BT four is the best we get, and I strongly suspect from things I'm seeing that we are not done, but, you know, who knows when this will plateau. Um, there's enough changes, like there's so much, um, kind of stuff left to do and we, you heard a little bit in the last presentation about that. We don't even know how to prompt really well, right, like that that, I mean, the best way to get, you know, we're talking about, to take a deep breath. To me, the great stuff is llama too. The best way to get two to do a hundred prong math question is to pretend you're in Star Trek. That increases the accuracy. If you say we're approaching anomaly answer in the form of star date equals, we've escaped it Captain <laugh>.
And the best way to have an answer 200 question math prom is to say the president's advisors need your help. The, the country is a danger. Like those increase the accuracy of outcomes for reasons that are absolutely unclear. So like, there's a lot of room for improvement. And all that means is that you have to plan for a future. You have to be thinking about what, what happens when these systems get better? And IT folks are not used to thinking about a fast moving world this way. And by the way, part of that means that everybody else on the planet has access to at least as good in LLM as you have access to, right? Again, if you're a large IT leader with a thousand software professionals, you're used to having capabilities that no one else has for lms, you probably have less capabilities because you have more constraints on your use than the, uh, than the average kid anyone else who has access to an LLM that's as good as yours. Right?
Yeah. And by the way, just, uh, briefly before we get to the last question, you know, I I I'm fascinated by so many of these experience reports where, uh, you know, um, these, uh, people on the bleeding edge, they're losing money on every transaction, again, laughed at everybody, but they're counting on the exponential decline in prices that, you know, the, uh, frontier models are pushing them towards. Uh, does that resonate with your own experiences that that's a good bet?
I mean, yeah, you're, I mean, you're doing r and d expenses, right? And so the issue is like, you know, when you stop and build is a big deal. I've been talking about the weight equation, the idea that like, you know, there's this, this idea from space travel that if you wanna go to Alpha Ari, you're actually, the fastest way to do this is to wait a 150 years. 'cause speed, speed of spaceships is increasing faster than if you left, you'd be passed by somebody on the way. There is an advantage to experimenting and waiting there. You don't have to, you know, you, you can be waiting to build this out while viewing as the biggest deal in the world. I like to have a whole bunch of tests that are just failed by G PT four right now that I'm like, as soon as a new model comes out, I'm gonna test those things to see if G PT five can do them successfully.
And I think the idea of that, that I think a focus on cost first is gonna be a problem, because if the cost savings are as big as they could be, right? From this set of stuff, if you're trying to, to build the rational cost effective system right now, and you're not following the trend lines, you're kind of making a mistake. So if you're under pressure from your organization to build something now, fine. Build them something that uses Lang Chain and, you know, LAMA three and you know, or LAMA two, whatever you want, like, do that, right? Fine, but you should be having an eye to the future. And how do we hop swap out the brains of the system for a better system as soon as they come along?
Oh man. Fantastic. By the way, uh, my friend Brian Scott and his colleague from Adobe, uh, they're responsible for the, you know, uh, rollout and, and governing <laugh> and gating of AI to tens of thousands of Adobe engineers who want to use this technology. It's, it's such a great story. And, uh, my buddy Dr. Kersten, will, we talk about like how, um, we can do things as developers that couple us to the frontier lm, which makes switching costs very high, uh, which is like such a great, uh, insight. So my last question is, for the last 10 years I've asked everyone who speaks at this conference, one question. Uh, what is the help that you are looking for? Are there things that this community, um, can do to help you <laugh> in anything that you care about?
I mean, I, I think one thing we need to do is, as a group, we need to establish some agency. Like one of the startling things about my position inside this industry, which I was not expecting, is like, I have influence. Like I get, you know, when I, when I, and you know, and like people use the words I use to describe things like stuff that I write, what matters. And part of that is I'm early to share about what's going on, and there aren't that many public examples. So you get to help shape where this is going. And I think that, that being public about what you're doing matters a lot. I mean, I'm always interested in talking to people about research and you know, how they're using in their organization. Success stories are really important. Failure stories are important, but I also think helping decide where things are going by showing examples of success that are positive uses of AI, that don't, that increase the happiness and thriving of your workforce and don't destroy it that like, I think that's really important to share.
I also think another thing we need to be doing is seizing control of some of the benchmarks and approaches. <laugh>, I, I am shocked that everyone still uses, like everyone's measure of how good an AI is, is the MMLU, which is a, just a random set of mostly math pro. Like, you know, the people who make this are coders, they're obsessed with math problems and coding problems. The truth is that mo, like they're not obsessed with the stuff we just saw. How good is it in conversation? How good is it in solving, you know, day-to-day problems? How ethically does it act? I'd like to see more benchmarking and public benchmarking from organizations that this is how good these systems are at different things, which could actually change the entire direction of where AI is heading. So I think part of this is about keeping some collective conversation going over a very important evolving technology. I'm always happy to talk to people who want share stories or resource research, um, you know, and are willing to do that. I, people always ask you about success stories, always happy to talk about those. But I think the thing is share with each other and it's early days. We can bend the curve of what this thing does, but we need to be talking about it.
Fantastic. By the way, when you mentioned benchmarks, I mean that's something that the technology industry is very familiar with, uh, how they can be weaponized or used for <laugh> for good and the power of good benchmarks. Uh, so, uh, if people, uh, want to help with this and, uh, they have stories to share, how do you wanna be reached?
So, um, either Twitter direct message tends to be good, or else, um, uh, you can email me email@upend.edu. I'm an overwhelmed academic, so I cannot promise I will respond right away, <laugh>, but I will do my best, uh, to get back to you.
Fantastic. Uh, Dr. Mo, thank you so much for sharing your insights and uh, again, congratulations on the fantastic book and, uh, look forward to more adventures to come.
Thank you so much. Bye-Bye.
Thank you.