Sean White on Inflection and AI frontiers

How will an empathetic AI interact with humans? What if we used AI not just for its IQ, but its EQ? How do we build the trust to enable it to take agency on our behalf? This week, Reid and Aria sit down with an expert and friend, Sean White to start answering those questions. Sean White is currently the CEO of Inflection AI and is working on expanding the relationship of what AI can do for humans. Sean takes a humanist approach to new technology, and is interested in AI that can go beyond intelligence to learn empathy and take action. Sean's background is as wide-ranging as the conversation, where we discuss the influence of AI in everything from neuroscience to understanding animal languages. Plus, listen to a live test with Pi to demonstrate Pi's ability to handle a tough situation at work, with empathy. Read the transcript of this episode here. For more info on the podcast and transcripts of all the episodes, visit https://www.possible.fm/podcast/ Topics: 01:22 - Hellos and intros 3:20 - Microsoft and Inflection 6:27 - The capabilities and influence of Inflection’s Pi 11:02 - Pi in the workplace 12:10 - Pi in action 15:38 - AI isn’t a human; it is a tool 16:11 - Pi’s IQ vs EQ 17:46 - Pi’s AQ, or our ability to take action via AI 18:51 - How our lives might change in the near future with Pi 21:36 - How our skill sets will elevate 22:28 - Neuroscience and AI 25:22 - Owning your intelligence 30:14 - The wonders of Earth Species Project 34:05 - Rapid-fire questions Select mentions: Inflection: https://inflection.ai/ Spruce art collective Dr. Eddie Chang, Chang Lab, UCSF Familiarity by the Punch Brothers Dr. Hirokazu Kato, Nara Institute of Science and Technology Possible is an award-winning podcast that sketches out the brightest version of the future—and what it will take to get there. Most of all, it asks: what if, in the future, everything breaks humanity's way? Tune in for grounded and speculative takes on how technology—and, in particular, AI—is inspiring change and transforming the future. Hosted by Reid Hoffman and Aria Finger, each episode features an interview with an ambitious builder or deep thinker on a topic, from art to geopolitics and from healthcare to education. These conversations also showcase another kind of guest: AI. Whether it's Inflection’s Pi, OpenAI’s ChatGPT or other AI tools, each episode will use AI to enhance and advance our discussion about what humanity could possibly get right if we leverage technology—and our collective effort—effectively.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:01Speaker 00

  If you're just thinking about IQ and EQ, well, that's the cognitive world and us being able to have discussions. AQ is really being able to then take action. It's as simple as, all right, I'd like to actually take something from my sales list and I want you to email those three people and then let me know what they say back. It should be able to do that. But that's what I mean when I say something that is agentic is its ability to take action and in some cases even take action that is complex and not just one single thing.

• 0:38Reid Hoffman?

  Hi, I'm Reid Hoffman.

• 0:40Aria Finger?

  And I'm Aria Finkler.

• 0:42Reid Hoffman?

  We want to know what happens if in the future everything breaks humanity's way? What we can possibly get right if we leverage technology like AI and our collective effort effectively?

• 0:54Aria Finger?

  We're speaking with technologists, ambitious builders, and deep thinkers across many fields: AI, geopolitics, media, healthcare, education, and more.

• 1:04Reid Hoffman?

  These conversations showcase another kind of guest. Whether it's Inflection's Pi, OpenAI's GPT-4, or other AI tools, each episode we use AI to enhance and advance our discussion.

• 1:17Aria Finger?

  In each episode, we seek out the brightest version of the future and learn what it'll take to get there.

• 1:23Reid Hoffman?

  This is possible.

• 1:31Reid Hoffman?

  What does it mean to amplify human intelligence with AI? How do you create truly empathetic AI? Today, we're talking about the intersection of LLMs and personal intelligence. Our guest hails from a company you may be familiar with. I'm certainly very familiar with it as I co-founded it with Mustafa Suleyman and Karen Simonian.

• 1:56Aria Finger?

  Our guest, the CEO of Inflection AI, brings decades of experience to some fascinating challenges: decoding animal languages, categorizing the Smithsonian's plant collection, running R&D at Mozilla, and pushing for advances in neuroscience, creative expression, and virtual reality, to name a few.

• 2:16Aria Finger?

  Now he leads, actually, you know what? Since Inflection is behind our favorite chatbot, Pi, that actually feels like a good reason to have Pi take the floor. Hey, Pi, could you please introduce our wonderful possible listeners to today's guest, who also happens to be our friend, and in a manner of speaking, your boss, Shaun White?

• 2:37Speaker 03

  Yes, Shaun White. He's the current CEO of Inflection AI, an AI company focused on developing personalized intelligence for everyone. Shaun White is a seasoned leader in the tech industry with decades of experience working at the cutting edge of technology, research, and business. He has previously held high-level positions at Mozilla, Bright Sky Labs, and Nokia, among others. Shaun White is known for his expertise in neuroethics, as well as his work in machine learning, user experience, and augmented reality.

• 3:06Reid Hoffman?

  All right. I think that does it. Thanks, Pi. Here's our conversation with Shaun White.

• 3:22Aria Finger?

  So Shaun, for those who don't know, March 2024, Inflection makes headlines because it announced a deal in which Microsoft would license Inflection's AI models, as well as hire many of the startup's employees, which is why Reid called you and said, hey, do you want to come on board? And do you want to join as CEO? And so that put this company at a unique crossroads. So can you tell us, like, what does Inflection look like today? What is it focusing on? And let's clear up the confusion for people who think Inflection equals Microsoft, because you have pivoted it to a new trajectory. So tell us about that.

• 3:57Speaker 00

  Yeah, I'm happy to. So first, this is not Microsoft. We're not part of Microsoft in any way, shape, or form. We're our own company and going off in our own direction. And a life part of this is what I know Reid has referred to in the past as refounding, right? That is, you have something, an organization that is building towards something, and then you're going to make a pivot, a shift. And that's what we've done. And so where previously, a lot of the work was really focused on the consumer usage, the consumer aspects, and frankly, How you are building up to a world-class LLM, how you're building the technology out.

• 4:38Speaker 00

  In some ways, this is a translational stage for the company, which is to say, that's fantastic. It can do so many things. Now, how do we apply that? How do we actually make work-life better? And that's been a large part of our pivot is to spend time with folks who are doing more commercial-oriented kinds of builds. So for instance, one of the things that the company didn't have before is we've opened up an API for usage. And that's been incredibly exciting for me because you get this great long tail of creativity. Right? So we had about 13,000 organizations that had signed up to use the API. And that's been everything from, you know, someone who wanted to explore what it means

• 5:17Speaker 00

  to talk to your spreadsheets, to someone who is helping kids learn how to read books, super cool company called Elo, some mental health companies that are exploring what it means to have either an aide in any of the many, many different ways in which we talk about wellness and health and that dialogue there. The list goes on. The other part of it has been for us to then work with some of these larger enterprises. My background initially is in HCI, which really is about being human-oriented in the things that we do and thinking about that interaction. Pi and really the inflection technology that we've been developing is perfect for that.

• 5:59Speaker 00

  And so a lot of what we've been doing is the kind of the equivalent where, you know, we'll have these fantastic voice at the customer meetings where we bring people in. And that ranges from, say, a car manufacturer or an airplane manufacturer or a large bank or an insurer. And really talk to them, not just say, hey, here's what you guys should do because we're the experts we know, but say, let's impedance match on the things that you are trying to achieve.

• 6:29Reid Hoffman?

  So say a little bit about, one, what inflection, Pi model, what its kind of special sauce is. Two, which pieces of the B2B enterprise pivot?

• 6:41Reid Hoffman?

  Where enterprises and partners who go, this is where it could be especially important. And then we'll just go branch off to it. But like, what's happening in the field of AI? And how people should understand their world's changing because of it?

• 6:54Speaker 00

  Part of this, I think, is a conceptual model thing that has been a shift and change for me, which is I started off really thinking of AI as a computational technology. And what I mean by that, you know, early on, some of my PhD work involved computer vision or, you know, looking video, things like this. But it was really about processing things or finding insights, machine learning. And I held on to that for a very long time. The thing that has been interesting about LLMs is that they bridge a computational technology with a user interface technology, because it's about language. And, and so the, the thing that has been really powerful and different first about LLMs is that they are new kinds of interfaces.

• 7:38Speaker 00

  And, you know, we can talk about dialogue and, and things that we are used to. And some part that is because we're comfortable with this technology, right? We invented language 100,000 years ago. Yay us. And, you know, it, it meant that, you know, for the first time recently, the computers have come to us as opposed to us going to the computers. And I love that, right? Now I and, and my kids and really anyone can talk in the same ways that you and I are talking right now. It's, it's collaborative in this really interesting way. And that's the thing that Pi is really unique and interesting and distinct in terms of its technology, is that a lot of the technologies and LLMs you see,

• 8:22Speaker 00

  you know, for that set of folks that are in the club as we are, where you've got over 300 billion parameters, right? These really large language models, they can do great book reports and, and they, they, they represent an incredible amount of knowledge, right? What Pi does and what it was fine tuned to do is actually be in dialogue and be in dialogue in a particular kind of way, which means, one, trying to learn more, two, being kind for some definition of kind, right? And really, in some ways, make you feel good about the conversation. Now, on the one hand, you might say, well, that's fine. But, you know, do enterprises really need that? Do they care or they just need to get something done?

• 9:07Speaker 00

  And if you talk to enterprises, you know, one of the challenges is that they have tens or hundreds of different applications that frankly mostly suck. They are really hard to use, they don't work together. You have to learn different interfaces for all of them. And one of the things that we have been working on building then is using speech, using this as an interface to aggregate a bunch of those things. That means that the enterprise cares because in some ways it's accelerating the things that they can do. The standard cognitive task that we would talk about, right? Like, oh, it'll help me be more creative or it'll help me work through a hard conversation and I can role play that and practice it.

• 9:50Speaker 00

  And so maybe the last part, because you were talking about the future. One of the things I get really excited about is this notion that, you know, Pi in particular is a great transducer to knowledge or to a collection of people or to a collaboration. And what I mean by that is, okay, so decades ago, I was working with this art collective called Spurse and we were doing this project, we were trying to give voice to the Cuyahoga River. If you remember, this is the river that caught on fire and had lots of interesting aspects to it. And so the art collective was putting microphones in so you could hear the flow of the water,

• 10:35Speaker 00

  interviewing people in the different communities and in aggregating that you get the voice. We actually can be in dialogue and have the voice of data. And that means that you can not only, you know, just have a dialogue with this massive collection of data instead of just seeing it, like instead of just a dashboard. But what it represents. And so if this is representing your 400 million users, you could be in dialogue with that 400 million users in ways that you just couldn't before.

• 11:04Aria Finger?

  So Sean, this is a perfect segue to thinking more about bringing Pi into the workplace. So I think, you know, at first, when I thought about Pi, I was like, oh, my God, of course, this is consumer. This is one to one. I'm so excited. But as you said, in the beginning, we were thinking about L L Ms for IQ in the workplace. And then when you actually like just think one layer deeper, it's like what everyone needs in the workplace is soft skills, is more empathy, is how to communicate with people, is how to talk to people like that is going to be actually the coin of the realm going forward. IQ is of course important, but EQ might be just important or maybe surpass it. So let's, let's see if Pi is feeling up

• 11:46Aria Finger?

  to it today. So I would love to talk to you to make this more concrete by bringing Pi into the conversation. And so can you walk us through a demo, for example, of how you would use Pi to navigate a tough situation in the workplace? And so if you're using your phone, just make sure to put your phone up to the mic so that we can, so we can hear Pi's voice if that works.

• 12:08Speaker 00

  All right. Hey, Pi. This is Sean.

• 12:12Speaker 05

  Hey there, Sean. How's your day going so far? Anything you need help with or just want to chat about?

• 12:17Speaker 00

  Yeah, it's pretty good. I had a question for you. I'm looking for some help with a conversation I'm having with a board member. He and I have worked together before. He's really great.

• 12:28Speaker 05

  A conversation with a board member sounds important. What kind of help do you need? Is it about preparing for the conversation, structuring your thoughts, or maybe even practicing some possible scenarios? I'm all ears. Well, virtually, anyway.

• 12:45Speaker 00

  Well, how about practicing some role playing with this? Because it might be a tough conversation, and we're trying to decide on how we do some compensation structuring.

• 12:59Speaker 05

  All right, let's dive in. For our role-playing exercise, let's imagine that I'm the board member and you've just brought up the topic of compensation and structuring. To start things off, I might say something like, "Sean, I understand the importance of discussing compensation, but I'm hesitant to make any major changes at this time. Can you explain why you believe adjustments are necessary?"

• 13:24Speaker 00

  Oh, yeah, no, that's great. So, yeah, part of us is, I think that we have one of our team members who is just really stepping up and what I'd like to do is increase their compensation and increase their equity in the company because I think that will align the incentives in a way that will have impact on our success and also just make us work better as a team.

• 13:48Speaker 05

  I appreciate your perspective, Sean, but I'm still cautious about making these adjustments. I'm concerned that increasing compensation and equity for one team member might lead to a sense of inequity or even resentment among the other employees. How do you plan to address these potential issues?

• 14:04Speaker 00

  That's a great question. Really, what I'm gonna do is make sure that everybody sees that value and that we're all in agreement about this. We try and be pretty transparent about all the things that we're doing.

• 14:19Speaker 05

  Transparency is definitely important, Sean, but I'm still worried about the impact this decision might have on team morale. How can we be sure that everyone will truly see the value in this change and that it won't create any unintended divisions within the team?

• 14:32Speaker 00

  That's a good question. You're the expert, what would you suggest?

• 14:37Speaker 05

  One approach could be to gather feedback from all the team members before making any final decisions. This could help ensure that everyone feels heard and understands the reasoning behind the proposed changes. Additionally, perhaps we could consider a more holistic review of compensation and equity across the entire team to ensure that any adjustments are fair and balanced for everyone.

• 15:01Speaker 00

  That's a great idea. I appreciate your advice.

• 15:05Reid Hoffman?

  It's useful. I certainly don't feel threatened in my job anytime soon.

• 15:09Aria Finger?

  I mean, still, every time I hear something like that, I do have to remind myself, like, Aria, there is not a person on the other side of Sean's phone. Like, this is an AI. That it is so interesting to hear that. And sort of like you said, sometimes you just need more of the, I just need someone to listen to me. I just need someone to bounce off. I just need someone to try it out with. And even if, you know, they're not giving you the the top notch 100% responses, just being able to practice with someone who's going to listen is incredibly valuable.

• 15:39Speaker 00

  We always talk about this as not being a human. It's a tool. It's not meant to replace humans in any way. But if someone feels better after talking with Pi, or if someone gets something done after talking with Pi, or they're able to, you know, bounce off an idea, or grow their ideas, or role play, or any of these things, that doesn't mean that the effect isn't there. It's similar to a placebo effect. It's still a very powerful and positive thing.

• 16:13Reid Hoffman?

  Like, for general principles, like, you know, how would I approach a board member to raise the topic the right way? It'll be very good at. When it gets to specifics, those things it won't generally be as good at. And so that's kind of the variation. And I think the thing that inflection adds in is a, an emphasis on being trained in EQ as centrally as IQ. So the IQ is extremely good, but to those dynamics of both how you interact with Pi itself, but also how Pi will recommend or shape how you interact with other people, because that's, you know, that's what it's trying to be extremely good at.

• 16:55Reid Hoffman?

  And, you know, to some degree, these things are trained. So what they're trained for really matters. And but so far, I don't think anything is. Trained nearly as good at the EQ as Inflection and Pi have done.

• 17:07Speaker 00

  That's right. We actually had a Berkeley student. She did this really interesting thing because the race is so constantly around IQ. But what she did is she took some of the EQ tests that we use with humans and she gave it to a bunch of different LLMs. And proud to say, Pi came out way on top. In some ways, this isn't a surprise, in part because I don't think folks are focusing on it. And a lot of the organizations really are focusing on IQ. And my sense is that for a lot of these things, we already have sufficient IQ. It's where we need to broaden to EQ. And if we talk about the future, to some extent, to AQ, or how we take actions based on some of that.

• 17:48Aria Finger?

  Sean, we were talking about that the other day, and you were talking about IQ, EQ, and AQ. Can you just go further? Like, what does AQ mean to you? And how will that be important in the future?

• 17:59Speaker 00

  In some ways, there's a simple version, right? Which is, if you're just thinking about IQ and EQ, well, that's the cognitive world and us being able to have discussions. AQ is really being able to then take actions. And you hear a lot of discussions now about agentic flows or agentic AI. And for us, just really trying to get some things that work that have meaning for enterprises. It's as simple as, we have these great demonstrations of this, where it's not only having that discussion like we just had, but if I say, all right, and I'd like to actually take something from my sales list, and I want you to email those three people, and then let me know what they say back, it should be able to do that. But that's what I mean when I say something that is agentic, is

• 18:45Speaker 00

  its ability to take action, and in some cases, even take action that is complex and not just one single thing.

• 18:52Reid Hoffman?

  Speculate a little bit about the near term and the medium term futures, about how our lives will change. How should people be thinking about, you know, what kind of future?

• 19:03Speaker 00

  Yeah, so there are lots of different versions of this, right? I mean, AI actually is not a single monolithic thing. Agentic isn't either, but I think some useful ways to think about it is within what environments do we want it to take action? And how do we want it to take action? But I think the places where it gets more interesting. One, if you're not just helping somebody in the workforce have, hard to say, you know, enjoy their work more, where they don't have to do all of these rote things, but some of that is just taken care of. That feels like a good aspiration. If it is actually letting us focus more on some of the cognitive things and all the little details get taken care of, that feels like a good aspiration.

• 19:46Speaker 00

  And then I think one of the places that if we pushed it even further, it would be useful for us is that, let's say, we start off with being able to do one action within some small place. That's great. Now we can cover a heterogeneous set of applications within enterprise. That's great. The part where it gets really interesting in my mind is when we start to look at what we tend to think of as executive functions and executive thinking. That is, I have an intention. I have a set of goals. I have a set of things I want to do that are both immediate and over the long term. And how is it helping me with that? It turns out, by the way, that these systems, certainly ours, is much better at figuring out intention

• 20:31Speaker 00

  than a lot of our previous heuristic-based systems, right? So if you think about how search, you know, search cares about your intention. I want to know what you are looking for, you know, and disambiguate. And in the way that we built this, we figure out the intention, then it might go to a different model, depending on the intention that you have for what you want to talk about. And that actually is very powerful. If you think about like, you know, someone that is helping you plan out or think about or even execute on your long-term goals, even actually sometimes knowing. Or at least asking you if what you really intended to do is different than what you said. Right? Because you know, oftentimes we'll express an intention.

• 21:15Speaker 00

  It says, well, you know, I want to buy this screwdriver. When what I really want to do is overhaul and fix this door. And if it can actually extract that intention, that's much more powerful than just answering the basic questions that I have. And so you can imagine that both in our, our personal lives and also in, in the commercial world.

• 21:38Reid Hoffman?

  That rabbit hole goes very deep indeed. What do you think will be like some of the evolving skillsets? Like how they'll need to conceptualize their work? How they'll need to be able to use these things to make their work part of their lives more effective? What's going to be the ongoing table stakes for people engaging with these agentic technologies to, to strategize and navigate their career and work trajectories?

• 22:07Speaker 00

  I think there are a couple of different models here. What you really want to be able to do is stay in the flow of the things that you're doing and be able to have that skill so that in the best of all possible worlds, a lot of the interface is disappearing and you're just focusing on the, the things that you want to do and the things that you want to accomplish.

• 22:30Aria Finger?

  One of the things I love, Sean, about talking to you is that I am so obsessed with neuroscience and you're always like giving me the latest. I'm like, wait, Sean, tell me what's, you know, what's the newest thing? And so I feel like that's another area where sort of well-being and health can really be positive, positively affected by AI. So can you just like, tell us what's going on? Like, what are the interesting and innovative things that you're seeing in that field right now?

• 22:58Speaker 00

  Yes. So the recency was a project that Steve Gulen is doing right now. And ADHD focused, these issues tend to be problematic. There've been a lot of alternatives to medication to try and approach this. The thing that was interesting about this project was that they use a small system that is effectively reading EEG. And we'll talk about that in a second. But you know, it's looking at what your brain is doing. It can detect through the use of AI whether you are focused or not. And in the game, that means, you know, the first level of the game, your character runs faster. So the more focused you are, much like meditation or something else, the better you play the game.

• 23:42Speaker 00

  And then there's this whole other set of activities that you do that build on that, that are directly related to clinical studies about how you can, you know, effectively change and, and work on your own ability to attend and to focus. I think this is just neat because it's a closed loop system. It couldn't have been done before because we didn't have the sensor systems for it. And we didn't have the AI to actually really grok what was going on in the brain. And so, you know, seeing some of the demonstrations of this yesterday in the studies, it was really exciting because it's another tool in the arsenal.

• 24:22Speaker 00

  Maybe one last one that's sort of related. There is a surgeon and professor up at UCSF, Eddie Chang. I just, I love his work. And so there'd been this previous work looking at the motor cortex. And if, if I imagined that I was drawing out the words A R I A, it could tell that I was thinking of your name, right? Because it was looking at the motor cortex and using a neural net to sort of predict. But what he did was looking at an overall FMRI of the entire brain. And again, using a generative model, able to really almost predict.

• 25:00Speaker 00

  Based on the activity in your brain, what you were thinking. And it wasn't perfectly accurate, but it was pretty close. And it was the closest I've seen to having someone who is trapped in their head and can't speak. But, you know, suddenly given a voice. And, you know, I think I said this earlier, right? That's one of these things that excites me, right? Is this, they think giving things a voice.

• 25:23Aria Finger?

  So when people talk about what makes us human, obviously it's, you know, is it your personality, your intelligence, the way, the way you look, sort of all of these things. And, and one of the things you've talked about, Sean, is the idea of ownership and owning your own intelligence. And I think especially with AI, you know, we, we've talked on this podcast before, everyone is concerned about privacy and owning their own data. But also Reed and I talked to Daphne Koller about sort of the tragedy in some medical and health situations where actually if we could just get all the data together and aggregate it, you know, obviously respecting PII, that we'd be able to make enormous breakthroughs. And I feel like that becomes even more sensitive in the intelligence realm.

• 26:07Aria Finger?

  It's like, oh, my gosh, own my own intelligence. That, that's what is me. So I would love to hear you talk about like, in this AI future, what does that mean for people to own their own intelligence or control their own intelligence or control the data around it?

• 26:20Speaker 00

  Yeah. So are there a couple of framings for that? Actually, I'll, I'll start with the, the way that we have talked about it with Consumer Pi. And in part, this is about owning it, but also being in control of it. And then we'll talk a little bit about the enterprise. So one of the things that we had announced, you know, I didn't think was going to be particularly controversial, was that we were working with this organization, DTI, the Data Transfer Initiative, so that there was interoperability across different personal AI systems. And, you know, this was sort of a first salvo on the idea that, oh yeah, well, you know, this was your discourse, your dialogue, your knowledge. You should own it. And so if you would like to have it and move it somewhere else

• 27:05Speaker 00

  or do something else with it, you should be able to do that. That feels right. We did this. In part, that was the initial intention was that what we create has value. Maybe it's just value to us, but it really, it has value. It's what makes us different and sometimes unique. So when you come to the enterprise, then oftentimes there are things that they have that can't leave the premises, right? And that might be for business reasons. That is, there are some trade secrets they don't want to share. That might be for regulatory reasons. And so a part of what we have been doing is, and this is one of the reasons to be able to license your code out, is to say, well, look, no, actually, if that's

• 27:46Speaker 00

  important to you, well, then you own that. And, you know, you can control it. You can run it where you want to run it. And there's a set of companies, particularly as we have learned more and more from them, that really care about that. I think there is an interesting question, because you were bringing this up about that tension between general public good and private privacy, you know, there's a, you know, your ability to sort of control and own all of that. And, you know, we saw this when I was at Mozilla, one of the places where we saw this was around speech and language. We had a project, Deep Speech, which is an open technology for doing speech rec. And

• 28:27Speaker 00

  then Common Voice, which was really an effort to let any language and representatives of that language collect and create their own data set. So, good example of this was, we see this a lot, actually, in the medical data and the neuroscience data, right? Which is, for instance, EEG data can be incredibly personal. And certainly, at the point where we start gathering more data, you know, there are folks who are working on things like wearable FMRI's, which then could translate to language. So knowing what you think. But it is, I think, where there's a lot of neuroethics work going on right now. We did some with the OECD, and different states are trying to sort through what this means. And

• 29:08Speaker 00

  Being able to associate that with different health outcomes or different attitudes, different behaviors is incredibly powerful. And so, you know, for that, I think there's still a lot of room for us to actually explore ways in which we do privacy-preserving but federated knowledge. I think we'll see more and more of that, particularly where we see that there's a trade-off between things that are public good, but also the desire to maintain privacy.

• 29:37Aria Finger?

  And I actually think so often, there isn't, there's only a tension in trust, that actually there's almost no tension between, okay, you want to keep your medical records private and you need to share them for this cancer study, for this whatever study. But because we have, and sometimes government bad actors throughout history, because we have this lack of trust, especially within specific communities, it, of course, people don't want to participate, and of course, people see that as a real tension, instead of, we can do both. And so to your point, we need to make sure to manage one to get the benefits, benefits of the other.

• 30:14Speaker 00

  No, I think that's exactly right.

• 30:15Reid Hoffman?

  You know, in our last minutes, before we get to rapid fire, let's also do a little bit more broad afield. I know that one of the projects that you and I work on together, Earth Species Project, where these generative AI models may help us understand other sentience, other consciousness, other, say a little bit about what you're doing with the Earth Species Project.

• 30:39Speaker 00

  Earth Species Project is a fantastic project. It started off with the seed of an observation. We have these machine translation techniques that we had started using that were unsupervised, which is to say, it used to be that if I wanted to translate English to Spanish, then I would have to label and say, "mam" equals "madre." But when we looked at large corpuses of data in both the languages, we started to see that there were structures, almost geometric structures in an n-dimensional space, a bit like constellations, right? And we could say, oh, that's interesting. The Big Dipper here, it looks like the Big Dipper here.

• 31:18Speaker 00

  And what that means is that madre and hija are the same over here in terms of mother and daughter, right? With that, I mean, most of the machine translation systems now don't need labeled data. They start to capture some of the interesting idiomatic aspects of it. And so that was great. We figured this out. The observation from Azza Braskin and Katie Zakarin, who are the co-founders of this, was, well, that's cool. What about all of these other species that we believe have some form of intelligence? And they are certainly communicating with each other. You know, for instance, we have evidence that there are pods of whales

• 32:00Speaker 00

  that have been around for centuries, that pass culture from one generation to the next. And so can we look at the ways in which they communicate? And again, use this as a transducer so that we can actually understand what it is they are saying? And can we speak back to them? That's the overall goal and focus of the project. And you can imagine with generative techniques, right? I can, as an example, kind of autocomplete a sentence or really a whole paragraph, just with a little bit of context. So you can do some of those same kinds of things if you want to autocomplete a bird song.

• 32:42Speaker 00

  Historically, what we would do in some of the studies is you would mimic something, play something back. But that's a little bit like talking in a mirror. And sometimes that gets a little awkward. I mean, if Reed, you and I were talking, but it was constantly Reed talking back to you and saying the exact same thing that you said, you'd be like a little kid, just stop doing that. Stop repeating what I say. But if you could really use these generative techniques to talk to other species, both to generate it and to understand it, that would be really powerful for several reasons. One, it actually gives us insight into the epistemology of these other species. Two, we can learn from them. I mean, it's really curious to understand a culture that has been going for hundreds of years in the ocean,

• 33:26Speaker 00

  actually has been passing along. I want to know that. But three, and maybe this is the most interesting, is it does start to ask the question of us about what our relationship is to them, what humanity's relationship is to all these other species. The Earth Species Project, to me, has been really exciting because it is potentially one of these world-changing moonshots in how we understand how we are in our relationship to the world. And if I didn't say this, I suspect Reed might add this in. If we ever encounter species outside of what we know and understand here on Earth, it gives us a head start on how we might communicate with them as well.

• 34:07Reid Hoffman?

  Yes, wow, a thousand percent.

• 34:16Reid Hoffman?

  All right, so rapid fire. Is there a movie, song, or book that fills you with optimism for the future?

• 34:22Speaker 00

  There's a song called Familiarity by the Punch Brothers. And it is as much about the way it makes you feel as some of the lyrics. And it is a complex song, almost in three parts. And it's almost confusing in terms of its genre. And every time I listen to it, you know, I feel hopeful about our future and where things are, and somehow comforted in the cycle of the universe.

• 34:51Aria Finger?

  Sean, what is a question that you wish people would ask you more often?

• 34:55Speaker 00

  There's a research professor, Hirozato Kato. He's a professor at the NARA Institute of Technology. I used to love his talks in the augmented reality conferences because he would always start it off asking the folks, is it possible? And then go on. Maybe this podcast aside, we don't actually ask that enough. Is it possible? Not, you know, can we do it? But is it possible?

• 35:20Reid Hoffman?

  We a thousand percent agree. Where do you see progress or momentum outside of your industry that inspires you?

• 35:28Speaker 00

  This is going to sound a little weird, but material science. In part, it's because it enables so many other things within food security, within all of the sort of medical field and things that we look at, just even within AI. So I guess related to us, but we are getting so much better at both understanding what happens at a very, very small scale and being able to guide and construct that into new materials that have new behaviors. Sometimes they are biomimicry, like the surface of a scramjet that actually models the surface of a leaf with stomata, you know, two threads and yarns of carbon nanotubes that are used, you know, something to be able to.

• 36:13Speaker 00

  Put something in through the, our veins that is, again, the end of scale and size. Um, I, I, every time I see this, it is only hopeful because it's, again, this giant space in which we can be creative and make progress.

• 36:29Aria Finger?

  Last question, can you leave us with a final thought on what you think is possible to achieve in the next 15 years if everything breaks humanity's way? And what's the first step to get there?

• 36:40Speaker 00

  I, I think the first step is for us to relax a little bit. That is for us to actually use these technologies so that all of the breakthroughs that are happening around Alzheimer's and dementia and, and the brain, we can test these and we can try these out. And what I mean by relax a little bit is we have rightly so leaned so hard towards concern about the things that could go wrong. But I want us to lean towards the things that could go right. There was, there was a professor from UCSF at a recent Stanford Medical School talk, and she highlighted this great thing around AI in particular. And she said, you know, there's all this discussion right now about what could happen.

• 37:24Speaker 00

  But I want you to know that someday, I believe it will be unethical not to use AI to save someone's life. I think for us to really get there, we have to, to focus on those things. There is always the nuts and bolts of doing the right thing, bending the arc in the right ways we do that. But we need to relax a little bit as we get there.

• 37:47Aria Finger?

  I love it so much. That's awesome. Sean, thank you so much.

• 37:50Speaker 00

  Thank you.

• 37:50Aria Finger?

  We really appreciate it.

• 37:51Reid Hoffman?

  Always a pleasure.

• 38:04Reid Hoffman?

  Possible is produced by Wonder Media Network. It's hosted by Ari Finger and me, Reid Hoffman. Our showrunner is Sean Young. Possible is produced by Katie Sanders, Edie Allard, Sarah Schleid, Adrian Bain, and Paloma Moreno Jimenez. Jenny Kaplan is our executive producer and editor.

• 38:22Aria Finger?

  Special thanks to Surya Yalamanchili, Saida Sabieva, Ian Ellis, Greg Beato, Ben Relis, Parth Patel. And a big thanks to Carrie Wang, April Duarte, and Little Monster Media Company.