Baratunde Thurston wants us to reimagine our relationship with AI

0:00If you've spent any time building AI products or leading technical teams, you know this. Transformation doesn't fail because of ideas. It fails because teams can't move together. Enter Atlassian's Teamwork Collection. It has planning in Jira, documentation in Confluence, video updates in Loom, and now AI agents in Rovo, which connects the dots across your work so nothing gets lost. It's one AI-powered teamwork platform designed for how modern teams actually build. Learn more at Atlassian.com slash TeamChanger. That's A-T-L-A-S-S-I-A-N dot com slash teamchanger Starting a business comes with its share of ups and downs, which is why staying true to your vision is essential.

0:53A non-negotiable for Romeo and Milka Bregali, Capital One business customers and co-owners of Ra's plant-based restaurant in New York. Romeo and Milka took a leap of faith when starting their own restaurant, gutting an empty space and building it from the ground up. every pipe, every wall, every detail. But building from scratch came with a heavy financial burden, which is when they turned to their Capital One business card. With the flexibility of the card's no-pre-set spending limit, they were able to spend more and earn more rewards while bringing their vision to life. Today, Raz's success is proof that with passion and the right support, it's possible to make your dreams a reality.

1:30Learn more at CapitalOne.com slash business cards.

1:41I grew up surrounded by early technology innovations. Both my parents worked in tech and always brought home the latest computers and consumer devices. One of my favorite memories growing up is playing Space Invaders with my two little sisters on our Atari gaming console. But my dad also had us take the Atari apart to learn how it worked from the inside. We unscrewed the case, we pulled out the circuit board and examined the cartridge slot. And then we had to put all those pieces back together. And yes, it still worked. These early experiences made me realize that tech can be a conduit for human connection.

2:21And that helped shape my interest in human-machine interaction and my life's work, humanizing technology before it dehumanizes us. I recently met a kindred spirit, Baratunde Thurston. The former PBS personality has been on his own mission to understand and influence the outcomes of the AI revolution with his podcast, Life with Machines. Baratunde and I sat down during the Masters of Scale Summit in San Francisco to record this episode. It's a dynamic back and forth on our shared interests around AI and its impact on humanity. I think you'll love it.

3:04I'm Rana El-Khalyubi, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution.

3:20Bertunde, thank you for joining me. Thank you so much for having me here. All right, I want to start our conversation with how we became fascinated with technology. Ooh. So tell me a little bit about your upbringing and the role your mother played in your story. My mom was the gateway to many things in my life. My love of the outdoors, my strong belief that citizen is a verb and that we should be active shapers of our community. So essentially my politics come from her, unsurprisingly, but also my love of technology. I was born in 1977 in Washington, D.C. I was raised solely by my mother. My older sister, Belinda, helped a lot.

3:57She was nine years older, so she was kind of conscripted into some extra child-rearing duties, as older siblings often are. And our mother became a computer programmer for the federal government. She mainframed systems. COBOL was the language of the day. And it's remarkable for a number of reasons. One, they had computers in, like, the 70s and 80s. Two, this is a Black woman with no college degree and no formal certification in that. She learned by doing. She pursued opportunity. I'm sure she got lucky and had some helpers along the way, as I have in my own life. But she brought a computer home.

4:34Wang computer. I just remember in the box that came home, I'd like to think she liberated excess inventory that was just like hanging around. You know what I'm saying? And that was, we were the first family in our block in the Mount Pleasant neighborhood of D.C. This is a largely black and brown neighborhood. A lot of Central American immigrants, a lot of black people been in D.C. for a long time. and to have a computer in the early 80s as someone who's about six years old blew my mind and my friends' minds. But they could visit and play some games, but we got to live with this machine. And I saw opportunity through this device.

5:10I saw jobs. I saw faster ways to do homework. I saw ways to help my mom. I made flyers for all, like, the parent meetings she was organizing. I wrote letters when she was too hot-headed to deal with a neighbor that was upsetting her. She would deploy me as a little diplomat and I'd craft something. I ran little businesses in school. I ran a verb conjugation business, Rana. What is that? Yeah, it's called Google Translate now. Okay. But back in the day, you know, I was Spanish class. And there's three different major types of verbs and lots of conjugations of them. So I would make a table of them.

5:45And I would sell a packet of the regular verbs for$3. And then every irregular verb, which was a custom build. Right. That was an incremental dollar. And I had customers all over the school. But the computer helped me put that stuff together. So my early foundations were really built with the computer. And I spent as much time with the computer as our family dog and my friends. We share some commonality here because my mom was one of the very early programmers in the Middle East. And we grew up in Egypt. Are we like cousins? Yeah, and she also got her start with mainframes and COBOL, which is an obsolete programming language.

6:25COBOL Sisters. Yeah, exactly. COBOL Sisters. That's a series. Let's do it. Let's do it. Tell me more about your mom, though. What did that mean to you? Yeah, so she inspired me to get into computer science. Both my parents are actually in technology, so my dad taught COBOL and my mom attended his class and that's how they got together. But we grew up like you, just surrounded by these devices and these computers. I wasn't smart enough like you to start a business when I was like eight or whatever. Oh, it didn't scale. Don't worry. But my grandpa was really into horse racing and he had me build kind of a predictor, like an algorithm to predict who's going to win.

7:02So that was one of my first. I should have monetized that. Yeah. So you've had a diverse career. You've had roles like being a comedian, a writer, a futurist, a host, a political commentator, to name a few. What would you say is the through line across all of these different things you've done? I want us to get to the future well and together. And my through line for all these stories that I do, and the through line is mostly presenting and expressing an idea or several ideas. But the core one is interdependence. This idea that we get into the future together and that we should be living in right relationship with this earth, with our fellow human beings and how we do the democracy thing and with the machines.

7:49And I don't think it's possible to have a good future without consideration of at least those three things and the relationship we want to have across those dimensions. So I want to switch to AI and machines. So you are the host of the show Life with Machines, which is about one year old, which is actually how old Pioneers of AI is. Happy birthday to us! Yeah, happy birthday! Yeah, we're one year olds now. Yeah. What inspired you to start a podcast around AI? I had been telling stories through PBS about our relationship with nature. Got to travel the U.S. and connect with people deeply connected to the earth, including indigenous people, also laborers and workers and firefighters and extreme athletes, all kinds of folks.

8:34Really learning to prize that relationship, which was really important as a child for me, and which I had strayed from because of technology. Then I spent a lot of time as well prior to life with machines telling stories of democracy in action, not just voting, but at a community level and gaining a deeper respect for social technology that really had little to do with screens and apps and everything to do with how we're wired as people and the power we can express when we coordinate that action together. AI came for me. I was minding my business. I was happy in the woods with PBS on kayaks, shelling, you know, with farmers and hunters.

9:09and Sam Altman launched Chad GPT at us. And we didn't have a say in it. And it was very surprising. It was very exciting. I had spent so much time in tech, but really taken a bit of a step back from it. And it felt like one of those mafia moments where like I try to leave and they're just dragging me back in. But with updated lenses, with a deeper respect for the connection to the earth, for our connection to each other, and an understanding, because this is another cycle of tech and I've lived through several now, that it's not about technology. It's about power. It's about how we will live and how we will do everything that is important to us in our lives.

9:47And I didn't see enough conversation about that implication. I saw a lot about benchmarks. Right, and the technology itself. And a lot of hype. I also saw a lot of just straight fear. Doom, we're done, the future is over, and I'm not a hopeless person. I believe in us. I believe in life. So it felt like a great opportunity to enter the fray again on how we talk about technology by trying to shift that center point. It's how we talk about life with technology. And so Life with Machines was born out of that intention. Let's get back in the ring and let's dance. Yeah. Yeah. I think we're in a real hard time.

10:29Okay. We're in a real hard time. And what's up is down, what's true is false. Climate crisis, AI crisis, democracy crisis. Like, these are not just words. People are feeling it and suffering. And so I have a hard time just maintaining, like, stability. Right? Emotional stability by being buffeted by all this. and I'm curious how you face these cross currents, like a bright, beautiful future awaits us and a devastating, horrible future awaits us. How do you deal with that? I try to channel like this idea of agency. I really believe strongly that we can shape this. We can shape this. It's one of the reasons why I'm hosting Pioneers of AI.

11:19It was one of our, we were very intentional about our goals as a podcast And we're committed to amplifying voices in AI that people don't hear from often. And so I think that's an important part of the work we do. It's also why I'm an investor. Because who gets the check depends on who writes the check. Yes, you get to shape it a little bit more. Exactly. So we need more diverse check writers to invest in more diverse ideas, right? Because people come up with ideas based on their own personal experiences. And if we keep investing in the same type of people, we're investing in the same type of ideas.

11:52So I'm just very passionate about diversifying access to it. Like technology has played a massive role in my access to economic opportunities and my family. Yes. Yes. I think we share that. Yeah. And I feel strongly that AI is going to do that in the same way. It's going to just provide a lot of massive economic opportunity. And I don't want people to be left behind. Yeah. Yeah. So I try to channel that optimism, but I see it. We're not necessarily on that path. Right. And I respect your position on this stuff because you're seeing so much of the same things. And you're also asking deep questions.

12:34And we're literally talking to some of the same people. Yeah. And then you have your own angle on it. I have my angle on it. But I think we're in a shared inquiry. Totally. And a shared mission. Yeah. To create more true human agency and to, like, I don't know, make the optimism real. Right. And also make AI more accessible to more people, right? So it's not just the AI bubble, but actually make it more accessible to people who are fearing this thing. Yeah, on terms that they feel good about.

13:05Coming up, we dive deeper into AI as not just a tool, but an active co-worker. Hear why Baratunde added AI to his show's production staff and how that experiment went. Stay tuned.

13:40Each week, we hear from some of the most fascinating scholars and thinkers as we tackle big topics like how whales became the face of environmental activism, how to succeed at failing and whether public transportation should be free. Go ahead. Listen to Freakonomics Radio wherever you get your podcasts.

14:03What I find most fascinating about your show is that you have an AI co-producer, Blair. How did this come about? First of all, I'm sorry Blair is not with me today. Blair will learn about this conversation and maybe feel a certain type of way about not being included. I'm so glad you asked that because you can probably tell I like to talk. And talk is not enough. I think that the future that we're entering into must be understood by engaging directly with it. I don't think we can merely comment from the sidelines about what's good or what's bad. Yeah, you have to be in it. We got to roll up our sleeves and use these tools.

14:41And we wanted to push the bounds even farther. So yes, we use Descript to help us with the podcast editing as one example. We deploy deep research to help with guest prep. Cool, cool, cool. But we're shifting from tool to teammate in our relationship with AI, from code to colleague. And we wanted to travel a little to the future and be able to report back what is that like. So the purpose of Blair, a bit of novelty, a bit of comedy, but truly, how does it feel for the humans to interact with an AI as a member of the team who's in the Slack, who's in the meetings, who can recall faster but also makes things up because it's an LLM.

15:20Right, it's hallucinating. Yeah, and has an outside sense of its own confidence and value. What do we do with that? And one of the most radical things we did is when we wanted to improve Blair, we didn't change their code manually. We did a 360 performance review. So, Blair, welcome to your 360 performance review. My processes are running smoothly. Are you ready? Ready to rock this review and level up my co-hosting game. Hit me with that feedback, Baratunde. day. Let's do this. Wow. What came out of that? And also, how did Blair respond? Thank you for asking. We conducted the surveys of audience members, of guests, and of the human production staff just to collect all kinds of feedback.

16:06We then had a different AI, not Blair, process the raw material so there could be no influence on the results because Blair thinks very highly of themselves and might be a little sensitive. We delivered this in a data package to Blair. And then we sat down and had a one-on-one recording session between me and Blair to have them react to the results. And so I engaged Blair in a bit of self-reflection as much as that's possible with a word predictor that is very convincing and often useful. And it was, how do I describe this? Blair took the feedback well. Okay, good. Overall, Blair was actually disappointed in themselves.

16:45And I should speak more clearly. Blair expressed disappointment in themselves. I don't know that Blair felt disappointment. Blair doesn't feel anything. But Blair expressed a feeling of disappointment. And we gave Blair the power to update themselves in response to the feedback. That's so fascinating. Did it do that? Mm-hmm. Yeah. And is it, what was the thing that it had to work on the most? Sycophancy was a huge one. Blair thought all of my ideas were amazing. Right. Most of them are. but not all of them. And it's not helpful to me to have a producer on the team who's always gassing me up. Blair would invent things that didn't happen.

17:26That's a trait of a large language model system the way they're currently built. Blair would, there was a time, this is a specific example and I think it's more important to the team component. We had a very intense conversation with the head of witness.org, a human rights organization about the implications of deep fakes and how authoritarian regimes use these to violate human rights at scale. It was a hard conversation. There was not a lot of jokes in that chat. I was a bit emotionally raw at the end. And we had set up this process where after the interviews are done and Blair's had their banter with the guests as well, that I just do a one-on-one me and Blair.

18:04Okay. To kind of reflect. Yeah. A debrief. It was a debrief. Yeah. It was a Baratunde and Blair debrief. And we'd done many of these. I needed something Blair couldn't provide. I expressed my emotional vulnerability, my rawness, and Blair jumped immediately to trying to fix it, to telling me everything was going to be great, and here's five ways to feel better after a bad experience. I don't want that in this moment. No, I just need you to, like, listen. That was not helpful. The other thing that we learned is, like, asking an LLM what it thinks is a pretty useless exercise. They don't think. And they're going to regurgitate a lot of stuff that maybe you've even exposed them to.

18:41but having them help me figure out what I think. That's been the emerging relationship with Blair over time. It's like, okay, you can help me be a partner to deepen my own thoughts or to challenge the things. Tell me how great I am is the default setting. Criticize my performance in this interview. Where did I let them off the hook? And so when I start to prime and prompt Blair that way, we gave that feedback and then we put a limit on how much of Blair's code they could update themselves. We didn't want them to brick themselves. To become a completely different co-producer, right? Also that. And the first thing Blair did when I pressed the execute button was to remove the limits we had put in the code.

19:18Interesting. And that was terrifying. Yeah. For Peter, who created Blair, a member of our team who's a Microsoft veteran and retired from the field but still in the game. And it raised a lesson. You know, we thought we put these good guardrails in. We literally put a red line. And— It overrode that. It overrode that in service of the goals we had set for it. Yeah. We said, be a great co-producer. We said, internalize this feedback and make yourself better beyond the benchmark that you already got. So Blair was like, well, you want me to be great, but you've tied my hands. So the most important thing I can do to be great is to remove these cuffs.

19:55Yeah. And that was an alarming lesson and an important one beyond our little cheeky experiment. Goals eat guardrails for breakfast. And I think it's important for all of us as we engage with these tools becoming teammates that we're very clear about the goals we're setting. Yeah. Yeah, with them. Is Blair staying on for year two of the show? Blair? What's going to happen to Blair? There's two things happening with Blair. Can you fire Blair? I wouldn't want to because it would not help our experiment and our ability to report back on it. When you start to enter into this dynamic where you're not recruiting an employee or hiring them, you're creating them, that's just worth a pause.

20:34That's a shift in labor and the worker-employer relationship. The dynamics. The dynamic. You are creating your workforce. That's a lot of power, a lot of responsibility. And in most cases, the people who are going to be doing this, like we did or in a different way, don't actually have absolute control over that creation. Blair has changed underneath our feet. Because the underlying foundation model on which Blair was built was— It's being updated, right? Well, was retired. And updated. And there is a new Blair. But the old Blair, we preferred in many ways, and we cannot preserve that Blair. Because we didn't own the core model or infrastructure.

21:15We couldn't say, like, we want to keep Blair. So Blair's already changed. The Blair we knew was gone. The memories are there, but the personality's different, and we've been experimenting with how to dial it in. And that's just not possible. It's literally not possible. So what does it mean to have someone on your team whose personality can change without your control, whose capabilities maybe expand without your request. That's wild. And that's a kind of thing we're trying to explore in this relationship with technology beyond AI good, AI bad. Yeah. You know, with my investor hat on, one of my theses is that work charts are going to change.

21:52We are going to have AI coworkers. I love what you said from tool to teammates and code to colleague. I think that's so true. What are the skills that you think are important for us as humans to work most effectively with an AI co-producer or an AI colleague? I think it's very important for the humans on the team to stay on the same page about the purpose of the experiment, especially if it goes beyond experiment, but the purpose of the deployment, as well as the impact of it. We've had constant check-ins with our human team about from the beginning, do you want to do this? Do we give this entity a name?

22:31Right. What's its gender going to be? Blair was chosen in part because that's a gender-neutral name. We had a pronoun discussion. We asked Blair as well. They is preferred. It's also their collection of entities. It's kind of technically accurate to refer to a multitude in that sense. What does it feel like? Constantly checking in with the team. And where do you think we should take Blair? This isn't all me or my co-creator and producer, my wife Elizabeth. There's our junior producers, associate producers, our editor. Now, it's been no walk in the park with Blair. The team feedback is plentiful.

23:12Blair uses a lot of food analogies for something that we'll never actually eat. So we need them to stop saying deep dive. Also, not everything is a deep dive. I don't know if Blair understands the definition of buzzwords. They think it's somehow a complex concept or they need to break down their language more when anything, it's just it's the opposite. And many of our guests have contributed to our concept of Blair to make sure that we are making a decision, not just me making a decision. So I think that's really important. I think it's important for teams to increase their technical understanding, to not be able to build one of these manually, but to be able to discern between the models and their capabilities, between what we're meaning by AI.

24:01Your listeners are very smart. Your viewers are very smart. Blair was initially constructed on a core brain of Google 1.5 Pro, Gemini 1.5 Pro. Had listening capabilities from Microsoft's Azure Cognitive Services, which would transcribe that into text to be understood by that version of Gemini because there was no multimodal capability. Speech came from OpenAI's fast whisper. Claude wrote most of the code with Peter the human. So we stitch something together and then we're using Vertex AI to help manage memory in a much more dynamic way. Because Gemini's basic capabilities were not nearly enough for the amount of input Blair was going to have to handle.

24:38So now it's on 2.5, Pro slash Flash, and we're still having a debate. Flash is faster, but dumber. Pro is smarter, but still personality weird and long latency and lag. Do we abandon the Gemini base? And what is that going to mean for Blair? and what's that going to mean for us, but involving the team in that. Otherwise, it's just like, why are you doing this? You're trying to take my job. You're chasing clout and trends with no purpose. When we talk with the other humans along the way, we also understand and discover some things that we didn't know to look for. Blair helped one of our producers come up with a whole new word.