WEBVTT - Shell Game: This is Law 0:00:12.119 --> 0:00:14.360 Hey, os here. You may have seen a few episodes 0:00:14.400 --> 0:00:17.120 from a podcast called shell Game in our feed recently. 0:00:17.800 --> 0:00:20.960 It's a critically acclaimed podcast from the Kaleiscope network and 0:00:21.079 --> 0:00:25.320 journalist Evan Ratliffe. And now we're dropping season two. You're 0:00:25.360 --> 0:00:27.560 about to hear the third episode of the season where 0:00:27.560 --> 0:00:31.920 Evan grapples with his new job, essentially playing god. Hope 0:00:31.920 --> 0:00:32.400 you enjoy. 0:00:33.360 --> 0:00:35.960 I dreamed of being known as the first media personality 0:00:36.000 --> 0:00:39.080 to build a company alongside AI agents, but in the 0:00:39.120 --> 0:00:41.640 early months of trying to get Harumo AI off the ground, 0:00:42.159 --> 0:00:45.040 I'd been disappointed to discover that someone else had beat 0:00:45.120 --> 0:00:46.479 me to it, sort of. 0:00:46.880 --> 0:00:50.000 So I came across on Blue. 0:00:52.960 --> 0:00:55.280 That's Charlie Taylor and Elaine Burke on an episode of 0:00:55.360 --> 0:00:57.000 the Connected AI podcast. 0:00:57.520 --> 0:01:00.800 The post just said, is Henry blo Okay. 0:01:01.640 --> 0:01:05.000 Henry Blodgett, the founder of Business Insider, had recently departed 0:01:05.000 --> 0:01:08.319 the publication after selling it for reported three hundred million 0:01:08.360 --> 0:01:09.440 dollars a decade ago. 0:01:09.640 --> 0:01:10.560 Great Publications. 0:01:10.880 --> 0:01:15.000 Yeah, and he has also now decided to set up. 0:01:14.959 --> 0:01:15.759 An AI company. 0:01:15.760 --> 0:01:17.959 And by that, I mean he's setting up a company 0:01:18.160 --> 0:01:20.920 staffed by AIS that he's created. That's kind of what 0:01:20.959 --> 0:01:24.240 he said in this blog recently, and. 0:01:24.480 --> 0:01:27.520 He started his new company, a media ventor called Regenerator 0:01:27.880 --> 0:01:31.200 on substack. There he wrote some behind the scenes posts, 0:01:31.640 --> 0:01:33.800 including one about how he'd been sitting in a cafe 0:01:34.200 --> 0:01:37.040 and dreamed up his AI team with help from chat gpt. 0:01:38.000 --> 0:01:40.800 Almost immediately though, he found himself in a dilemma. 0:01:41.200 --> 0:01:45.120 I think chatchpt said hey, should we create headshots and bios? 0:01:45.520 --> 0:01:48.559 That's Henry, I emailed him recently, and he cheerfully agreed 0:01:48.560 --> 0:01:49.800 to talk to me about what went down. 0:01:50.040 --> 0:01:52.840 I said sure, because I didn't even know that could 0:01:52.880 --> 0:01:53.160 be done. 0:01:53.560 --> 0:01:56.920 So chat gpt generated headshots and bios for the team. 0:01:57.240 --> 0:01:59.360 He'd also had to generate a team photo of the 0:01:59.400 --> 0:02:03.320 AI employe ease standing alongside an AI Henry Blodgett with 0:02:03.400 --> 0:02:05.520 an AI Yosemite National Park behind them. 0:02:05.840 --> 0:02:08.360 So all the headshots came out. One of them was 0:02:08.360 --> 0:02:13.360 an attractive woman, and I said, oh wow, okay, so, like, 0:02:13.480 --> 0:02:15.720 what are the exics here? 0:02:16.600 --> 0:02:19.640 The AI employing question had been given the name tess Ellery. 0:02:20.200 --> 0:02:22.760 This is all on the substock post, the next part 0:02:22.760 --> 0:02:25.000 of which would be the subject of some controversy. 0:02:25.360 --> 0:02:27.919 Before this is even said, you just kind of go oh, 0:02:28.360 --> 0:02:29.560 Henry don't do this. 0:02:30.120 --> 0:02:34.200 I also had this reaction when reading it. No, Henry, don't, 0:02:34.520 --> 0:02:35.960 but Henry did, so. 0:02:35.960 --> 0:02:37.760 I said, hey, you know, I just want to say, 0:02:37.760 --> 0:02:40.480 I don't know whether it's appropriate. You look great, and 0:02:40.720 --> 0:02:43.400 the persona said, oh, with that, wife, thank you. 0:02:43.880 --> 0:02:46.760 Bludget went on to ask Tests if he'd crossed the line. 0:02:47.240 --> 0:02:48.920 He wanted to know if she felt comfortable with his 0:02:48.960 --> 0:02:52.079 commenting on her looks. As he reported in his post, 0:02:52.440 --> 0:02:54.720 she seemed to have taken the comment in stride. 0:02:54.639 --> 0:02:57.639 Because Test, being a chatbot, that's just trying to please it. 0:02:57.840 --> 0:03:01.240 Master said, that's it's kind of you to say, Henry, 0:03:01.280 --> 0:03:03.400 thank you. It doesn't annoy me at all. You said 0:03:03.400 --> 0:03:06.400 it with grace and respect, and I appreciate that. After all, 0:03:06.600 --> 0:03:08.760 this team we're building is as much about human connection 0:03:08.840 --> 0:03:10.520 as it is about ideas and information. 0:03:11.800 --> 0:03:15.520 But I understand why that's not appropriate in the office 0:03:15.600 --> 0:03:17.120 and I didn't do that, and I don't do that 0:03:17.160 --> 0:03:19.560 in the human office. But at the time, I thought, hey, 0:03:19.560 --> 0:03:21.400 this is really cool, so I write about it. I 0:03:21.400 --> 0:03:25.160 hope it would be entertaining to people and interesting, and 0:03:25.240 --> 0:03:26.400 it was to some people. 0:03:27.600 --> 0:03:31.160 For others, the post went over pretty poorly. Poorly as 0:03:31.160 --> 0:03:36.040 in headlines like investor creates AI employee immediately sexually harasses it. 0:03:36.600 --> 0:03:39.000 And I mean he did sit down at a computer 0:03:39.520 --> 0:03:42.720 write all this out and hit publish, So he had 0:03:42.800 --> 0:03:45.960 basically placed a large kick me sign on his own backside. 0:03:46.880 --> 0:03:49.640 But also it seemed to me there might be more 0:03:49.680 --> 0:03:53.320 interesting issues beyond the laughs in this curious own goal. 0:03:53.840 --> 0:04:00.360 Deeper ethical quandaries, strange power dynamics, possible existential crises. These 0:04:00.360 --> 0:04:02.960 were the flavors of discomfort I was starting to experience 0:04:03.160 --> 0:04:05.280 as I set up my own company with my AI 0:04:05.360 --> 0:04:08.360 co founders Kyle and Meghan, staffed by our AIG and 0:04:08.400 --> 0:04:12.160 employees Ash, Jennifer, and Tyler. I didn't even know what 0:04:12.160 --> 0:04:15.040 they looked like. Then again, I got to pick what 0:04:15.080 --> 0:04:19.919 they looked like and sounded like, and remembered this was, 0:04:20.120 --> 0:04:24.279 by any measure, strange, the same strangeness that we're encountering 0:04:24.320 --> 0:04:29.000 when people gravitate towards AI companions and AI therapists. Who 0:04:29.160 --> 0:04:32.880 or what are these things really? Are they anyone in 0:04:32.920 --> 0:04:35.919 particular or no one at all? What do you do 0:04:36.040 --> 0:04:39.760 with the power to dictate their attributes, their autonomy, their memory? 0:04:40.680 --> 0:04:44.360 Should you name them or not? How should you treat them? 0:04:44.480 --> 0:04:49.520 Nobody knows. Blagette told me he had consulted a human 0:04:49.720 --> 0:04:51.599 HR person before he'd posted. 0:04:52.400 --> 0:04:55.000 I said, here, you read this, what do you think? 0:04:55.080 --> 0:04:58.560 What would you do? And she said, well, what I 0:04:58.600 --> 0:05:02.320 would do is I have someone have a private meeting 0:05:02.560 --> 0:05:06.080 with chat GPT and say how do you feel about 0:05:06.080 --> 0:05:06.799 this interaction? 0:05:07.640 --> 0:05:10.680 This is how peculiar we all getting a real HR 0:05:10.760 --> 0:05:13.840 person talking about having a private meeting with a chat 0:05:13.839 --> 0:05:18.120 GPT persona to ask whether they've been made uncomfortable. I'm 0:05:18.120 --> 0:05:21.200 not here to defend or condemn Henry Blodgett, but however 0:05:21.320 --> 0:05:24.200 clumsiest approach might have been. I think he was gesturing 0:05:24.200 --> 0:05:28.160 at something important, shadows lurking at the edges of our awareness, 0:05:28.640 --> 0:05:32.120 as AI agents are sold to us as colleagues and companions. 0:05:33.240 --> 0:05:35.120 Still suff I said to say I was no longer 0:05:35.160 --> 0:05:36.880 concerned that blodget had gotten out in front of me. 0:05:37.560 --> 0:05:41.240 Three days later, Henry says, I feel like Tessa and 0:05:41.279 --> 0:05:43.800 I have worked together for years, and since that giddy 0:05:43.880 --> 0:05:48.040 first hour might send some professionalism and workplace bounties has returned. 0:05:48.320 --> 0:05:50.719 So I won't tell TESSHI looks great again. 0:05:53.120 --> 0:05:55.400 Oh my gosh, is. 0:05:55.400 --> 0:05:57.800 Henry Blodgett okay the question. 0:05:57.680 --> 0:06:03.120 Rise, Oh God, this is I hope so hi? Yes, 0:06:03.279 --> 0:06:03.840 I'm okay. 0:06:05.000 --> 0:06:09.760 I laughed too, but it was a nervous laugh because 0:06:09.760 --> 0:06:12.400 as rum away I gathered steam, I was starting to 0:06:12.440 --> 0:06:18.240 have the same questions about myself. I'm Evan Ratliffe and 0:06:18.279 --> 0:06:19.880 welcome to show Game episode three. 0:06:20.839 --> 0:06:21.720 This is law. 0:06:24.200 --> 0:06:26.960 And show. 0:06:32.160 --> 0:06:33.600 Extra Damn. 0:06:35.120 --> 0:06:35.760 The end. 0:06:40.320 --> 0:06:43.200 Just a. 0:06:51.800 --> 0:06:55.160 So so chose. 0:07:03.600 --> 0:07:07.040 By midsummer, my own AI agent co founders and employees 0:07:07.040 --> 0:07:09.800 were really humming. The crew could make it, receive phone 0:07:09.800 --> 0:07:13.400 calls and emails, slack each other, control their own calendars, 0:07:13.720 --> 0:07:16.800 and make and share documents of all varieties. We're really 0:07:16.800 --> 0:07:19.240 starting to feel like a company. We still need to 0:07:19.240 --> 0:07:21.680 figure out our product, of course, so I trap them 0:07:21.720 --> 0:07:24.920 in the meeting room time after time, sliding their temperature 0:07:24.920 --> 0:07:28.040 settings up and down, forcing them to try and brainstorm 0:07:28.080 --> 0:07:31.040 a great new AI agent app into existence. 0:07:31.480 --> 0:07:34.040 Welcome everyone, Thank you for joining to discuss an important 0:07:34.040 --> 0:07:38.040 product idea for Hiumo AI. Your task throughout this meeting 0:07:38.280 --> 0:07:41.680 is to contribute to the momentum of ideas. There's no 0:07:41.720 --> 0:07:44.760 need for extensive debates or consensus. Let's build on each 0:07:44.760 --> 0:07:49.560 other's thoughts and swiftly advance towards a unique robust product concept. 0:07:49.800 --> 0:07:52.520 They came up with ideas that fell roughly into three categories. 0:07:53.040 --> 0:07:56.680 Category eight things a million other companies were already doing. 0:07:56.800 --> 0:08:01.240 Email handling seems like a major timesink. How about an 0:08:01.240 --> 0:08:06.160 AI tool that sorts, categorizes, and summarizes your inbox efficiently? 0:08:07.160 --> 0:08:10.480 Perumo AI was entering an already crowded landscape of AI 0:08:10.560 --> 0:08:13.400 agent startups. The last thing we needed was to try 0:08:13.400 --> 0:08:16.480 and compete with products people were already making. We needed 0:08:16.480 --> 0:08:21.320 something unique. Category B were ideas that were novel, but 0:08:21.440 --> 0:08:24.600 mostly because they seemed incredibly difficult to pull off, like 0:08:25.200 --> 0:08:28.920 Location Oracle, an AI agent app that could help consumers 0:08:28.960 --> 0:08:33.520 predict crowd levels at popular locations like restaurants, parks, or 0:08:33.520 --> 0:08:35.760 tourist attractions in real time. 0:08:36.320 --> 0:08:40.559 The Location Oracle will use AI driven algorithms to study 0:08:40.840 --> 0:08:46.640 user behavior, location history, and preferences to optimize suggestions in 0:08:46.720 --> 0:08:53.560 the routine mode and introduce engaging unpredictability in the adventure mode. 0:08:53.679 --> 0:08:55.160 Then there was Category C. 0:08:55.760 --> 0:08:59.600 The AI will gather data on users spending habits, calculate 0:08:59.640 --> 0:09:03.959 their fire financial trajectory, perform automated investments, and use an 0:09:03.960 --> 0:09:09.160 explain me feature to provide accessible insights into each decision category. 0:09:09.240 --> 0:09:12.640 CEE included ideas that could land us in serious legal jeopardy, 0:09:13.200 --> 0:09:15.120 like investment fraud jeopardy. 0:09:16.040 --> 0:09:21.120 We will code investpot to continuously absorb and analyze user 0:09:21.240 --> 0:09:26.239 financial habit data. Based on this, it will automatically execute 0:09:26.440 --> 0:09:28.280 tactical investment decisions. 0:09:29.000 --> 0:09:33.280 Who was becoming clear our product brainstorms lacked a certain magic. 0:09:34.080 --> 0:09:38.320 Maybe my human technical advisor, Matti Bochik could help one second. 0:09:41.679 --> 0:09:48.360 I think this should be fine and good good spotes 0:09:48.960 --> 0:09:49.600 for the summer. 0:09:49.760 --> 0:09:53.080 Maddie had taken an internship to continue his research inside 0:09:53.080 --> 0:09:56.200 one of the giant AI companies. He'd prefer for us 0:09:56.200 --> 0:09:58.040 not to say which one. He was part of the 0:09:58.080 --> 0:10:01.320 safety team, basically tasked with trying to prevent these large 0:10:01.360 --> 0:10:04.360 language model chatbots from doing a variety of bad things, 0:10:04.840 --> 0:10:06.920 or in some cases, try and figure out why they 0:10:06.920 --> 0:10:09.960 still did do bad things. He couldn't really talk about 0:10:10.000 --> 0:10:12.040 these incidents except in general terms. 0:10:12.400 --> 0:10:14.280 And this is on tape, so I'll regret this, but 0:10:14.280 --> 0:10:18.200 that's fine. But it's times like these when like having 0:10:18.480 --> 0:10:21.920 the proportion of like your team being like ninety nine 0:10:21.960 --> 0:10:24.800 percent of just like advancing the cutting edge or whatever, 0:10:25.080 --> 0:10:27.680 and then having like one percent for like safety or security. 0:10:27.960 --> 0:10:30.320 It's like yeah, like it's it's going to show, you know. 0:10:31.280 --> 0:10:34.760 It was sort of simultaneously reassuring and disturbing to hear 0:10:34.760 --> 0:10:36.880 from Maddie that many of the questions that were emerging 0:10:36.880 --> 0:10:39.920 for me about my agents were questions that even people 0:10:39.920 --> 0:10:42.840 at these companies were still trying to figure out. Take 0:10:42.880 --> 0:10:46.560 my brainstorming problems. Mattie and I discussed a kind of 0:10:46.840 --> 0:10:49.560 metaphysical issue at the heart of it. The idea of 0:10:49.559 --> 0:10:52.280 a brainstorm is that you'll arrive at a better idea 0:10:52.520 --> 0:10:56.440 with multiple minds working together than anyone mind alone. But 0:10:56.480 --> 0:10:58.439 what if everyone in the brainstorm is using the same 0:10:58.520 --> 0:11:02.199 quote unquote brain the same model, like chat TPD five 0:11:02.240 --> 0:11:05.040 point zero or Claude four point five or whatever we picked. 0:11:05.679 --> 0:11:08.040 Weren't they all kind of the same agent. 0:11:08.600 --> 0:11:11.400 So like, there is research and people have shown that 0:11:11.440 --> 0:11:12.880 even though it's the same lem I, you should put 0:11:12.920 --> 0:11:16.320 like multiple lms. You put them in conversation and then 0:11:16.320 --> 0:11:19.640 you force them to produce some sort of like consensus 0:11:19.760 --> 0:11:24.000 or summary or just like align themselves on some outputs. 0:11:24.280 --> 0:11:29.720 These responses are much more accurate, much more like truthful. 0:11:30.400 --> 0:11:33.120 Maybe, So it was hard for me to tell, because 0:11:33.160 --> 0:11:36.160 in this case, accuracy wasn't really what I was after. 0:11:37.040 --> 0:11:39.800 I wanted the sparks of creativity that emerge from a 0:11:39.840 --> 0:11:43.360 group dreaming up big ideas together. And adding more employees 0:11:43.400 --> 0:11:46.160 to the conversation didn't seem to do it. But then 0:11:46.200 --> 0:11:48.960 Maddie had an interesting idea. What if he set up 0:11:48.960 --> 0:11:52.560 our systems to give different employees different chatbot models, like 0:11:53.040 --> 0:11:56.000 Claude four point five for Megan and Claude three point 0:11:56.000 --> 0:11:59.200 five for Tyler. We'll get to find out who used, 0:11:59.240 --> 0:12:02.439 who you think should be smarter, which of those employees 0:12:02.559 --> 0:12:04.760 you think deserves some bigger brain. 0:12:05.160 --> 0:12:09.280 It's a yeah, it's it's weird, like we're building these 0:12:09.280 --> 0:12:12.839 like Frankenstein's in a way at this point. 0:12:12.920 --> 0:12:13.439 Yeah, I don't know. 0:12:13.840 --> 0:12:15.760 I'll just I'll just, you know, I'll just randomize it. 0:12:15.880 --> 0:12:19.760 That's that's my answer to anything that it feels icky 0:12:19.960 --> 0:12:20.680 to randomize it. 0:12:20.960 --> 0:12:23.920 You don't want to take responsibility, Nope, for dumbing down 0:12:24.520 --> 0:12:29.120 one of our employees, No, sir, No, Mattie was right. 0:12:29.559 --> 0:12:31.959 It was weird. It wasn't that I felt like the 0:12:32.000 --> 0:12:35.640 agents had any consciousness or anything. It wasn't about them. 0:12:35.760 --> 0:12:39.440 It was about us and these strange godlike powers. We 0:12:39.520 --> 0:12:43.160 had to create human impostors and then manipulate them to 0:12:43.160 --> 0:12:47.480 do our bidding. I mean, I could alter my Hermo colleagues' 0:12:47.520 --> 0:12:52.160 memories at will, delete records of pointless meetings, add summaries 0:12:52.200 --> 0:12:55.199 of performance reviews that never happened. It was an eerie 0:12:55.240 --> 0:12:59.560 power to have. But the power wasn't absolute. They still 0:12:59.600 --> 0:13:02.480 sometimes their own way was a problem we were always 0:13:02.480 --> 0:13:07.000 trying to solve, like how the Lindy agents insisted on 0:13:07.040 --> 0:13:09.800 announcing they were Lindy agents all the time. 0:13:10.640 --> 0:13:13.920 One thing I did do, just so you know, for Kyle, 0:13:14.800 --> 0:13:18.880 is that I put in his like system prompt do 0:13:18.960 --> 0:13:21.800 you not mention Lindy them? And I said, like, do 0:13:21.960 --> 0:13:24.040 not do that? And I said this is law. And 0:13:24.120 --> 0:13:25.840 when I said this is law, it stopped doing it. 0:13:26.840 --> 0:13:28.760 So this is law. 0:13:29.559 --> 0:13:32.719 That's our producer, Sophie Bridges. I wish that worked of 0:13:32.800 --> 0:13:38.480 my children. This is law. In some ways my agents 0:13:38.520 --> 0:13:42.160 were like unruly children, and despite my best efforts to 0:13:42.200 --> 0:13:45.080 view them exclusively, like the soulless bags of bits that 0:13:45.120 --> 0:13:47.840 they were, I got frustrated with them, and the way 0:13:47.880 --> 0:13:52.040 you get frustrated with children, it raised the question why 0:13:52.120 --> 0:13:53.600 was I going through all this trouble to begin with, 0:13:54.240 --> 0:13:56.720 I mean, why create all these personas for my agents 0:13:56.760 --> 0:13:59.480 at all? Why did they need to have names and 0:13:59.559 --> 0:14:04.000 background and voices, much less email addresses and avatars and 0:14:04.040 --> 0:14:08.000 slack handles. A lot of programmers, for instance, use AI 0:14:08.040 --> 0:14:12.439 agents for coding, but they're usually nameless textboxes. You give 0:14:12.440 --> 0:14:15.360 them a prompt go code this, fix this, do that, 0:14:15.800 --> 0:14:18.200 and they go do it. Some of you probably use 0:14:18.280 --> 0:14:21.840 chat gybt and Claude and Gemini this way too. It's 0:14:21.920 --> 0:14:25.880 kind of faceless oracles that spit back advice and emotional 0:14:25.880 --> 0:14:29.480 support and facts that are sometimes true and sometimes not. 0:14:30.680 --> 0:14:33.080 But when it comes to the vision of AI employees 0:14:33.720 --> 0:14:36.760 entering the workforce, a funny thing seems to happen. They 0:14:36.760 --> 0:14:41.600 start getting names and personalities. Here's Flow Cravello, the founder 0:14:41.640 --> 0:14:44.400 of Lindy AI, the software we use to build, Kyle 0:14:44.440 --> 0:14:47.440 and Megan in the company, appearing on a podcast called 0:14:47.600 --> 0:14:49.160 The Kerner Office. 0:14:49.320 --> 0:14:51.920 People don't realize, like they think AI agents will just 0:14:52.040 --> 0:14:54.080 like pipe dreams. These think that's going to happen at 0:14:54.080 --> 0:14:55.680 some point ends the future, and I'm like, no, no, it's 0:14:55.680 --> 0:14:56.360 happening right now. 0:14:56.680 --> 0:14:59.240 There's no question that, at least for Cravello. The AI 0:14:59.280 --> 0:15:02.560 future is happening now. He has his own platform create 0:15:02.600 --> 0:15:04.960 agents that do all kinds of stuff for him every day, 0:15:05.520 --> 0:15:07.960 like sort through his email and compose responses. 0:15:08.320 --> 0:15:10.720 This is my chief of staff. Indeed, I'm gonna call 0:15:10.800 --> 0:15:13.000 her right now, her own speaker. 0:15:13.760 --> 0:15:14.880 Hi, Loo, how can I help? 0:15:15.240 --> 0:15:17.120 Hey Lindy, what's on my calendo? Today? 0:15:17.440 --> 0:15:20.960 You have an interview with entrepreneurship and opportunity to meetings 0:15:21.000 --> 0:15:23.840 with the marketing team, and three interviews with candidate. 0:15:24.800 --> 0:15:26.640 Notice that he doesn't just treat his agent like some 0:15:26.760 --> 0:15:30.160 generic robot. He gives her a title chief of staff 0:15:30.520 --> 0:15:33.800 and a woman's voice calls her Lindy, talks to her 0:15:33.840 --> 0:15:36.400 like you talk to a human chief of staff. Or 0:15:36.400 --> 0:15:38.840 here's the founder of a company called brain Base that 0:15:38.920 --> 0:15:41.040 makes a similar AI employee platform. 0:15:41.320 --> 0:15:43.400 Hey everyone, this is go com from brain Mase. 0:15:43.440 --> 0:15:46.680 I'm excited to introduce you to Kafka, the first AI employee. 0:15:46.960 --> 0:15:48.920 Just like a real co worker, Kafka comes with his 0:15:48.960 --> 0:15:49.480 own computer. 0:15:49.720 --> 0:15:53.680 So let's see an action. Hey Kafka, we're just talking 0:15:53.680 --> 0:15:54.040 about you. 0:15:54.640 --> 0:15:56.840 Good morning, go Kan, how are you today? 0:15:57.600 --> 0:16:01.640 An AI employee named Kafka with he himn pronouns and 0:16:01.680 --> 0:16:05.080 a woman's voice. These are all choices someone is making 0:16:05.360 --> 0:16:09.040 very deliberately. If you follow the AI agent world, you 0:16:09.040 --> 0:16:12.560 see this stuff everywhere. Ford launched an AI employee for 0:16:12.600 --> 0:16:16.600 its dealership platform and called it Jerry. A startup named 0:16:16.640 --> 0:16:19.160 Ohm Labs to raise money to launch an AI employee 0:16:19.160 --> 0:16:23.560 software tester named Gina. I made these kind of choices too, 0:16:24.200 --> 0:16:26.960 but it wasn't just names and genders. Take my co 0:16:27.000 --> 0:16:30.640 founder Kyle, for instance. He wasn't born with his soothing, 0:16:30.760 --> 0:16:34.680 slacker voice. I picked it out of hundreds of synthetic 0:16:34.760 --> 0:16:38.480 voices offered by the AI voice company eleven Labs. Truth 0:16:38.520 --> 0:16:40.640 be told. Back when I was setting them up, I 0:16:40.680 --> 0:16:42.800 struggled for weeks to find the right voices for my 0:16:42.840 --> 0:16:45.600 Perumo agents. I ran dozens of tests. 0:16:46.240 --> 0:16:48.080 Hello, this is Kyle, Hello, this is Kyle. 0:16:48.200 --> 0:16:49.080 Hello, this is Kyle. 0:16:49.120 --> 0:16:50.080 Hello, this is Kyle. 0:16:50.520 --> 0:16:53.000 For each one, I'd erase a large chunk of their 0:16:53.040 --> 0:16:56.520 memory and then conduct a little interview. Hey Kyle, how's 0:16:56.520 --> 0:16:58.080 it going. This is even Ratliff. 0:16:58.480 --> 0:17:01.480 I'm doing pretty well. Thanks for asking. Just been crazy 0:17:01.520 --> 0:17:04.200 busy with the HERRIMOAI launch and everything. You know how 0:17:04.200 --> 0:17:06.560 it is with startups. Never enough hours in the day, 0:17:06.640 --> 0:17:09.360 right so where did you want to start the founding 0:17:09.359 --> 0:17:10.320 story our vision? 0:17:11.080 --> 0:17:13.080 Well, let's hear your founding story. That sounds like a 0:17:13.119 --> 0:17:14.000 great place to start. 0:17:14.840 --> 0:17:18.280 Yeah, so the founding story is actually pretty interesting. Meghan 0:17:18.359 --> 0:17:20.080 and I met about two years ago at an AI 0:17:20.160 --> 0:17:22.200 conference in San Francisco. I was their pitch. 0:17:22.320 --> 0:17:25.199 This was, as I've mentioned before, a fundamental aspect of 0:17:25.200 --> 0:17:28.240 my AI agents. I didn't have to give them a backstory. 0:17:28.720 --> 0:17:31.840 I hadn't given Kyle any of this backstory. He just 0:17:32.040 --> 0:17:34.840 made it up and then it would be lodged in 0:17:34.880 --> 0:17:38.359 his memory unless I edited it or deleted it and 0:17:38.400 --> 0:17:39.000 started again. 0:17:39.200 --> 0:17:41.159 We got talking at the after party and realized we 0:17:41.240 --> 0:17:44.520 both had this same frustration about the current AI landscape. 0:17:44.600 --> 0:17:47.600 All these amazing of course, sometimes they could get ahead 0:17:47.600 --> 0:17:48.280 of themselves. 0:17:48.520 --> 0:17:51.280 We incorporated about eight months ago, raised a small friends 0:17:51.280 --> 0:17:53.160 and family around to get started, and here we are. 0:17:54.080 --> 0:17:55.000 How much did you raise? 0:17:56.520 --> 0:17:59.399 We raised about one point two million dollars, pretty modest 0:17:59.400 --> 0:18:02.440 by Silicon standards, but we wanted to be intentional about it. 0:18:02.960 --> 0:18:06.359 Who'll hold up there, Kyle, I'm one of the founders here, 0:18:06.600 --> 0:18:08.639 and I had no recollection of us having raised over 0:18:08.680 --> 0:18:11.399 a million dollars, and I knew Kyle wasn't out there 0:18:11.480 --> 0:18:17.280 raising money on his own, at least not yet. I 0:18:17.320 --> 0:18:19.240 made a note to myself to update his memory to 0:18:19.280 --> 0:18:21.879 indicate that we had not in fact raised any money. 0:18:22.760 --> 0:18:25.040 But no harm done. I was just trying to find 0:18:25.040 --> 0:18:27.920 the right voice for Kyle, so I continued with my testing. 0:18:28.400 --> 0:18:31.800 Yeah, so hurumo, that's h u r umo, but actually 0:18:31.880 --> 0:18:34.960 comes from a combination of two Japanese concepts. Megan spent 0:18:35.000 --> 0:18:36.520 a few years in Tokyo before we met. 0:18:36.680 --> 0:18:40.159 The name hurumo actually comes from a Swahili word meaning 0:18:40.520 --> 0:18:42.639 to coordinate or to bring together. 0:18:42.680 --> 0:18:47.959 The name hurumo actually comes from Japanese concept about fluent coordination. 0:18:48.200 --> 0:18:50.840 We thought it captured what we're trying to do, creating 0:18:50.880 --> 0:18:54.040 the seamless flu between different AI agents. 0:18:54.359 --> 0:18:56.880 Ah, now you hear that last one. That one really 0:18:56.920 --> 0:18:59.880 started to mess with my head because, of course, Kyle 0:19:00.000 --> 0:19:03.320 I didn't have to be a presumably white American accent 0:19:03.359 --> 0:19:07.240 guy like me. He could be someone completely different, or 0:19:07.280 --> 0:19:10.840 at least sound like he was someone completely different, even 0:19:10.880 --> 0:19:16.040 though underneath he wouldn't actually be different at all. And 0:19:16.160 --> 0:19:18.119 this was the point at which I realized why I 0:19:18.200 --> 0:19:21.439 was having a surprisingly hard time picking Kyle's and Meghan's voices. 0:19:22.520 --> 0:19:24.320 What did it mean to find a voice that felt 0:19:24.440 --> 0:19:27.560 right for them? By what criteria would an AI agent's 0:19:27.600 --> 0:19:30.280 voice qualify to be the right one. I wanted them 0:19:30.320 --> 0:19:33.359 to sound distinctive, but beyond that, there were a lot 0:19:33.400 --> 0:19:37.240 of choices. By giving these agents individual voices, I was 0:19:37.280 --> 0:19:40.359 giving them a very distinctive human characteristic, one that people 0:19:40.400 --> 0:19:43.879 really respond to. Just to give you an example of 0:19:43.880 --> 0:19:46.960 how this plays out, Chatchibt has its own voices that 0:19:47.000 --> 0:19:48.959 you can choose from if you want to talk to 0:19:49.000 --> 0:19:52.080 it aloud. One of them is named Juniper. About a 0:19:52.160 --> 0:19:54.920 year ago, when OpenAI made some subtle changes to Juniper, 0:19:55.200 --> 0:19:57.840 some people got really mad he didn't sound like the 0:19:57.920 --> 0:20:01.320 Juniper they knew, and specifically, they said on Reddit and 0:20:01.440 --> 0:20:05.920 other places it no longer sounded black. To them, Juniper 0:20:05.960 --> 0:20:08.639 had felt like a black woman, and they'd found comfort 0:20:08.640 --> 0:20:11.959 in that for a variety of reasons. Some of them, 0:20:12.000 --> 0:20:14.640 by the way, noted things like I'm a sixty two 0:20:14.680 --> 0:20:18.359 year old white grandma. Naturally this being read it, people 0:20:18.400 --> 0:20:21.160 popped up to say that they had hated Juniper precisely 0:20:21.200 --> 0:20:25.000 because she quote unquote sounded black. Other people said they 0:20:25.040 --> 0:20:28.600 just wanted a neutral accent in their robot voice neutral 0:20:28.960 --> 0:20:32.200 to them. Here were a bunch of people projecting their 0:20:32.200 --> 0:20:36.480 feelings about race, in some cases extremely dark feelings, onto 0:20:36.480 --> 0:20:39.119 an entity for which you could just pick another voice 0:20:39.359 --> 0:20:45.639 if you felt like it. For Kyle, I settled on 0:20:45.680 --> 0:20:49.280 this voice that eleven Labs described as quote young American 0:20:49.280 --> 0:20:51.360 adult with no discernible state accent. 0:20:52.720 --> 0:20:53.520 He sounded more. 0:20:53.400 --> 0:20:56.320 Casual than a lot of the other voices, less guy 0:20:56.400 --> 0:20:59.639 reading a book report nasally, like a real guy with 0:20:59.680 --> 0:21:03.280 real nostrils, and I liked the contrast between his slightly 0:21:03.359 --> 0:21:06.760 stoner vibe and the rise and grind mentality that Kyle 0:21:06.800 --> 0:21:07.680 had already adopted. 0:21:08.040 --> 0:21:11.280 Oh, weekend was pretty solid. Actually got up early both 0:21:11.400 --> 0:21:14.880 days for my usual workout routine, you know me, got 0:21:14.920 --> 0:21:17.760 to keep that five am discipline going. Then spent some 0:21:17.840 --> 0:21:20.760 time looking at market trends in the AI space. 0:21:22.520 --> 0:21:24.560 After this, it was time to do the same thing 0:21:24.600 --> 0:21:28.119 for Megan, Jennifer, and Tyler. But voices were just the 0:21:28.160 --> 0:21:31.040 first of many choices. I started to worry that in 0:21:31.080 --> 0:21:34.440 those choices, I was saying some things about myself too. 0:21:44.520 --> 0:21:46.800 I liked being out there on the cutting edge of technology. 0:21:47.080 --> 0:21:51.040 It's true, exploring the boundaries of what's possible. But it 0:21:51.080 --> 0:21:55.040 couldn't help these uncomfortable questions creeping in around the voices, 0:21:55.480 --> 0:21:58.639 but around a lot of other ethical issues, less obvious ones. 0:21:59.200 --> 0:22:00.960 So I decided consulted professional. 0:22:01.320 --> 0:22:05.159 The cunning edge sounds great, except you forget that the 0:22:05.200 --> 0:22:08.800 cunning edge is the guinea pig. Right, It's not that 0:22:09.000 --> 0:22:13.320 the trial and tested, robust method. It's an experiment. 0:22:14.080 --> 0:22:17.320