WEBVTT 1 00:00:00.160 --> 00:00:03.920 We've spent decades treating the human brain as a biological 2 00:00:03.920 --> 00:00:07.760 computer. And today, we talk about artificial intelligence as a 3 00:00:07.760 --> 00:00:11.120 digital brain. We use the exact same words. 4 00:00:11.120 --> 00:00:12.960 We say a machine is learning. 5 00:00:12.960 --> 00:00:14.800 We say it predicts the word. 6 00:00:14.800 --> 00:00:17.120 We say it hallucinates when wrong. 7 00:00:17.120 --> 00:00:20.720 But relying on this interchangeable vocabulary is a trap. 8 00:00:20.720 --> 00:00:24.960 It's a broken metaphor, actively damaging clinical psychology and our 9 00:00:24.960 --> 00:00:27.280 ability to assess AI safety. 10 00:00:27.280 --> 00:00:31.280 Brains and language models share a mechanism called predictive coding. 11 00:00:31.280 --> 00:00:34.240 Both guess the world and use error signals to correct 12 00:00:34.240 --> 00:00:37.920 themselves. But the physical reality is not identical. 13 00:00:37.920 --> 00:00:41.440 The brain operates bounded by strict biological limits. 14 00:00:41.440 --> 00:00:45.120 Mathematical logic scales cleanly to silicon processors. 15 00:00:45.120 --> 00:00:49.040 A biological framework requires a living, vulnerable body. 16 00:00:49.040 --> 00:00:51.600 The brain does not sit passively in the dark, waiting 17 00:00:51.600 --> 00:00:54.080 for your eyes and ears to send it information. 18 00:00:54.080 --> 00:00:57.680 It actively guesses the world into existence, running ahead of 19 00:00:57.680 --> 00:00:59.920 reality and using sensory data only to correct. 20 00:01:00.280 --> 00:01:04.920 its mistakes This data visualization shows that biological cost. 21 00:01:04.920 --> 00:01:08.040 The human brain accounts for just 2% of total body 22 00:01:08.040 --> 00:01:11.560 weight, yet it burns through 20% of your daily metabolic 23 00:01:11.560 --> 00:01:15.560 energy. Every single guess is paid for in glucose. 24 00:01:15.560 --> 00:01:19.400 This process is an organismic prophecy in an indifferent, dangerous 25 00:01:19.400 --> 00:01:23.080 universe. Constantly projecting what is about to happen next is 26 00:01:23.080 --> 00:01:27.320 the absolutely cheapest survival strategy available to any vulnerable living 27 00:01:27.320 --> 00:01:31.960 creature. For humans, prediction is never a theoretical math exercise. 28 00:01:31.960 --> 00:01:35.160 It is a constant act of survival, funded entirely by 29 00:01:35.160 --> 00:01:38.680 physical energy. This brings us to allostasis. 30 00:01:38.680 --> 00:01:42.440 Unlike homeostasis, which reacts to a physiological problem after it 31 00:01:42.440 --> 00:01:45.960 happens, allostasis is the process of predicting what the body 32 00:01:45.960 --> 00:01:49.240 is about to need and deploying resources before the deficit 33 00:01:49.240 --> 00:01:52.440 occurs. Your heart rate climbs a moment before you stand 34 00:01:52.440 --> 00:01:55.600 up. Your cortisol levels rise just before your morning alarm 35 00:01:55.600 --> 00:01:58.720 sounds. Your brain acts as an internal treasurer. 36 00:01:58.720 --> 00:02:03.600 It relentlessly runs a body budget, auditing your metabolic spend 37 00:02:03.600 --> 00:02:07.600 against the environmental demands it expects you to face. 38 00:02:07.600 --> 00:02:11.040 When an organism can no longer predict its own needs 39 00:02:11.040 --> 00:02:13.920 accurately, that budget overruns. 40 00:02:13.920 --> 00:02:17.680 The internal treasurer fails, and the body pays the debt 41 00:02:17.680 --> 00:02:20.640 through chronic physical wear and tear. 42 00:02:20.640 --> 00:02:25.440 Through this economic lens, psychological trauma takes on a specific 43 00:02:25.440 --> 00:02:29.200 physical meaning. It is a structural failure. 44 00:02:29.200 --> 00:02:33.360 The brain's predictive system encounters an event far too massive 45 00:02:33.360 --> 00:02:36.400 to anticipate, and the system shatters. 46 00:02:36.400 --> 00:02:40.160 The brain learns the world is unpredictably lethal. 47 00:02:40.160 --> 00:02:44.160 As a result, it begins assigning extreme threat levels to 48 00:02:44.160 --> 00:02:45.920 minor sensory noise. 49 00:02:45.920 --> 00:02:49.280 A system stuck in this state of hypervigilance can no 50 00:02:49.280 --> 00:02:54.280 longer afford. to ignore false alarms Checking every single floorboard 51 00:02:54.280 --> 00:02:58.520 for danger requires a profound, exhausting amount of energy. 52 00:02:58.520 --> 00:03:01.880 This is a complete and devastating collapse of a body's 53 00:03:01.880 --> 00:03:07.000 metabolic budget. Large language models also build highly complex internal 54 00:03:07.000 --> 00:03:11.480 structures to predict the world, but their underlying physical reality, 55 00:03:11.480 --> 00:03:14.280 their substrate, is entirely different. 56 00:03:14.280 --> 00:03:16.280 This chart shows the difference. 57 00:03:16.280 --> 00:03:20.280 A biological organism pays for errors in tissue damage, social 58 00:03:20.280 --> 00:03:21.800 rupture, and death. 59 00:03:21.800 --> 00:03:24.760 An AI pays in mathematical loss functions. 60 00:03:24.760 --> 00:03:27.320 The constraints are entirely detached. 61 00:03:27.320 --> 00:03:30.680 The language model operates without thermodynamic burden. 62 00:03:30.680 --> 00:03:33.560 Its limits are compute power and server bills. 63 00:03:33.560 --> 00:03:37.000 An artificial network has no continuous living self. 64 00:03:37.000 --> 00:03:40.280 Because it does not inhabit a physical body, it completely 65 00:03:40.280 --> 00:03:43.640 lacks the biological imperative to ensure it survives to see 66 00:03:43.640 --> 00:03:46.960 tomorrow. We can engineer a robot to have a battery, 67 00:03:46.960 --> 00:03:49.760 but those stakes are externally assigned by us. 68 00:03:49.760 --> 00:03:54.160 For humans, the stakes are intrinsic, inescapable, and existential. 69 00:03:54.160 --> 00:03:57.760 A patient told debug a thought is often left trying 70 00:03:57.760 --> 00:04:00.080 to fix their logic while their body remains in a 71 00:04:00.080 --> 00:04:02.320 state of high alert exhaustion. 72 00:04:02.320 --> 00:04:06.720 Because the underlying budget crisis hasn't been addressed, these logic-based 73 00:04:06.720 --> 00:04:09.600 treatments often fail to produce lasting change. 74 00:04:09.600 --> 00:04:14.000 Real clinical success requires respecting the body's physical substrate. 75 00:04:14.000 --> 00:04:17.520 It means finding ways to explicitly lower the metabolic cost 76 00:04:17.520 --> 00:04:19.040 of existing in the world. 77 00:04:19.040 --> 00:04:22.640 Social attachment functions as a mechanical tool for this purpose. 78 00:04:22.640 --> 00:04:26.160 A safe relationship acts as a buffer that literally reduces 79 00:04:26.160 --> 00:04:28.560 the brain's anticipated energy spend. 80 00:04:28.560 --> 00:04:32.480 Lasting recovery happens through physically restoring the body's energy budget 81 00:04:32.480 --> 00:04:33.920 to a state of safety. 82 00:04:33.920 --> 00:04:37.520 The vocabulary we use also distorts our approach to AI 83 00:04:37.520 --> 00:04:40.600 policy. When we look at a mathematical misfire in a 84 00:04:40.600 --> 00:04:43.720 language model and call it a hallucination, we are making 85 00:04:43.720 --> 00:04:45.000 a dangerous choice. 86 00:04:45.000 --> 00:04:49.720 Borrowing words from clinical pathology accidentally smuggles human vulnerability into 87 00:04:49.720 --> 00:04:53.960 cold statistics. By treating algorithms like fragile minds, we blind 88 00:04:53.960 --> 00:04:57.400 ourselves to the actual structural risks these systems pose. 89 00:04:57.400 --> 00:05:00.120 The human brain is a metabolic prophet. 90 00:05:00.120 --> 00:05:04.520 It is tethered to mortality, and its predictions carry semantic 91 00:05:04.520 --> 00:05:08.200 meaning entirely because of the threat of death. 92 00:05:08.200 --> 00:05:11.160 The next time someone tells you their artificial model is 93 00:05:11.160 --> 00:05:15.080 genuinely thinking, ask them what the machine is physically paying 94 00:05:15.080 --> 00:05:16.600 to have that thought.