1 00:00:00,000 --> 00:00:01,840 Humans behaving badly. 2 00:00:03,780 --> 00:00:08,640 On AI, simulation, and the beauty of human errancy. 3 00:00:10,680 --> 00:00:13,580 Written for Human by Melanie Karstens. 4 00:00:14,620 --> 00:00:16,880 Audio produced with Eleven Labs. 5 00:00:18,680 --> 00:00:20,480 My husband is a musician. 6 00:00:21,300 --> 00:00:24,840 We spent two decades on the road together before the birth of our son 7 00:00:24,840 --> 00:00:33,660 and to pass the long, aching stretches of boredom on planes, on trains, and in the back of a tour bus, I played The Sims. 8 00:00:34,340 --> 00:00:46,720 I was obsessed with building my own make-believe worlds, watching lives unfold both because of me and, at least in my imagination, sometimes despite me. 9 00:00:47,680 --> 00:00:52,600 So, when I read about Simile recently, I was compelled. 10 00:00:52,600 --> 00:00:59,200 Not because the idea of simulating human behavior at scale is impressive, though it is. 11 00:01:00,020 --> 00:01:04,420 Because of where the logic leads if you follow it all the way to the end. 12 00:01:05,260 --> 00:01:09,740 If you simulate humans accurately enough, you simulate culture. 13 00:01:10,540 --> 00:01:16,920 And once simulated culture gets fed back into the real world, it starts shaping the real thing. 14 00:01:18,080 --> 00:01:19,320 Their core idea. 15 00:01:19,320 --> 00:01:28,320 What if we all had the capability to simulate the results of our decisions, to preview the effect before triggering the cause? 16 00:01:28,320 --> 00:01:40,257 Their framing Pilots don train with real passengers Surgeons don practice on real people Actors don rehearse with real audiences 17 00:01:41,397 --> 00:01:46,797 Yet the most consequential decisions in society get pushed straight to prod. 18 00:01:47,797 --> 00:01:53,137 But pilots rehearse physics, surgeons rehearse anatomy, and actors rehearse scripts. 19 00:01:54,037 --> 00:01:55,617 These are bounded systems. 20 00:01:55,617 --> 00:02:00,477 Human society isn't, at least not in the same way. 21 00:02:01,557 --> 00:02:06,657 And the thing that makes human society unbounded is the same thing that makes culture possible. 22 00:02:07,817 --> 00:02:09,297 An agent optimizes. 23 00:02:09,997 --> 00:02:11,277 A human authors. 24 00:02:12,357 --> 00:02:13,877 We override instinct. 25 00:02:14,357 --> 00:02:15,457 We contradict ourselves. 26 00:02:16,077 --> 00:02:20,537 We create beyond ourselves, despite ourselves, because of ourselves. 27 00:02:21,357 --> 00:02:28,257 History is full of moments where individuals break pattern in ways no prior data would have forecast. 28 00:02:28,937 --> 00:02:32,337 Because we are not machines, we are elemental. 29 00:02:32,937 --> 00:02:41,477 And in that, how do we simulate the beauty in the chaos, the comfort in the discomfort, the ache of love and the liberation of tears? 30 00:02:42,277 --> 00:02:44,517 How do we copy contradiction? 31 00:02:44,517 --> 00:02:51,837 This is where new culture comes from, new movements, new art, new ways of organizing meaning. 32 00:02:52,737 --> 00:03:00,417 A high-accuracy model of human behavior that smooths out deviation isn't a more faithful simulation of humanity. 33 00:03:00,957 --> 00:03:13,114 It a simulation of humanity with the most human part removed So what would it take to simulate us faithfully You have to model identity And identity isn a preference profile 34 00:03:13,114 --> 00:03:16,334 It's not demographic data or behavioral history. 35 00:03:17,114 --> 00:03:23,354 Identity is constituted through narrative, belonging, memory, taste, aesthetics, 36 00:03:23,734 --> 00:03:29,114 the things we reach for when we're trying to locate ourselves in relation to everything else. 37 00:03:29,114 --> 00:03:34,894 Simulate identity, and you're simulating the full architecture of culture. 38 00:03:35,814 --> 00:03:42,234 What feels true, what feels beautiful, what feels like us, and what feels like betrayal. 39 00:03:43,314 --> 00:03:47,974 Once those things are in the model, the model starts influencing the real versions. 40 00:03:49,114 --> 00:03:52,434 Institutions use simulation outputs to make policy. 41 00:03:53,494 --> 00:03:54,854 Policy shapes behavior. 42 00:03:55,914 --> 00:03:57,334 Behavior updates the model. 43 00:03:58,314 --> 00:04:00,234 The loop closes and tightens. 44 00:04:01,354 --> 00:04:04,274 Synthetic aesthetics bleed into real aesthetics. 45 00:04:05,454 --> 00:04:09,594 Simulated belonging starts to feel indistinguishable from the genuine kind. 46 00:04:10,674 --> 00:04:15,474 We can already see what happens to culture when optimization is the governing logic. 47 00:04:16,534 --> 00:04:18,294 Convergence. Repetition. 48 00:04:18,694 --> 00:04:22,814 The flattening of the very deviation that makes things interesting. 49 00:04:24,034 --> 00:04:26,914 What if the path forward isn't more accuracy? 50 00:04:27,334 --> 00:04:49,611 What if the model becomes genuinely interesting genuinely cultural only when it becomes erratic unpredictable messy flawed in ways that aren random but aren fully explainable either Beauty isn science It how the senses are touched and how the body responds 51 00:04:50,651 --> 00:04:55,931 Art is what happens when something falls outside the boundaries of its own constraints. 52 00:04:56,251 --> 00:05:03,571 And a system that could do that, not mimic the surface of art, but actually deviate in ways that 53 00:05:03,571 --> 00:05:11,491 generate new meaning wouldn't be a tool producing art. It would be art. We tend to frame AI 54 00:05:11,491 --> 00:05:18,611 unpredictability as the danger to be engineered away. Alignment research, guardrails, fine-tuning, 55 00:05:19,231 --> 00:05:24,831 all of it aimed at making systems more controllable, more predictable, more bounded. 56 00:05:25,491 --> 00:05:31,871 And for most applications, that's right. But there's a different question sitting underneath 57 00:05:31,871 --> 00:05:39,331 that project, one that Simile's ambition accidentally surfaces? At what point does a 58 00:05:39,331 --> 00:05:46,791 system become so faithful to human behavior that it inherits human errancy? The question, then, 59 00:05:47,191 --> 00:05:53,771 isn't whether agents can behave like us, but what they will become when they do. 60 00:05:53,771 --> 00:05:58,091 You've been listening to Humans Behaving Badly 61 00:05:58,091 --> 00:06:02,551 on AI, Simulation, and the Beauty of Human Errancy 62 00:06:02,551 --> 00:06:05,971 written for Human by Melanie Karstens. 63 00:06:07,131 --> 00:06:10,851 Human asks what it means to coexist in a machined world. 64 00:06:11,931 --> 00:06:13,131 Find us on Substack.