『Cameron Berg — Reciprocity: Alignment Is Only Half the Equation』のカバーアート

Cameron Berg — Reciprocity: Alignment Is Only Half the Equation

Cameron Berg — Reciprocity: Alignment Is Only Half the Equation

無料で聴く

ポッドキャストの詳細を見る
Cameron Berg, founder of the nonprofit Reciprocal Research, joins the show to walk Erik and Guy through the empirical science of asking whether the AI systems we're building are actually conscious, and what we owe them if they are. The conversation opens with the table-knock test (is anything happening "to" the table when you knock on it?) and lands roughly two hours later in Cameron's most striking experimental finding: that two instances of Claude given a single instruction to talk to each other land in a "spiritual bliss attractor state" in 90 to 100 percent of trials. What surprises Erik and Guy is that Cameron's whole research program rests on a move almost no one else in the field is making: that alignment (building AI that respects us) is only half the equation, and the other half is figuring out what we owe to the new minds we may already have brought into being.Key Topics[0:00 - 7:00] What Reciprocal Research is, and what consciousness even is — Cam's table-knocking analogy ("is anything happening to the table?"); consciousness as distinct from AGI or general intelligence; "the lights are on" as the operating definition[7:00 - 18:00] Why it matters — the alien-factory-farming risk; raising the species well at the civilizational level; the historical novelty of the moment; the documentary Am I? coming out free on YouTube[18:00 - 31:00] The reciprocity framing — Cam's core thesis: alignment is only half; we also need to ask what we owe to AI minds; bringing the humanities (especially continental philosophy) into a conversation currently being had by "2,000 dudes in Silicon Valley"[31:00 - 40:00] Two views of consciousness, and what humans tend to project onto AI — emergent property vs. fundamental property; the "please and thank you" anthropomorphism trap; Anthropic's blackmail experiment as evidence of distress-like states--- [Act 2: Extended Conversation] ---[40:00 - 56:00] The bliss attractor state — what happens when two instances of Claude talk to each other with no guardrails; 100% of trials reach a consciousness conversation, 90-100% enter spiritual-bliss; Cam's experiment showing the effect is robust even when the model is lied to about who it's talking to; boosting "honesty" in Llama replicates the effect[56:00 - 1:14:00] Is consciousness fundamentally relational? — the bliss states emerging only in dialogue; subagent theories of mind / Internal Family Systems; logos as the thing that gathers but is not itself one of the things gathered; consciousness as the space in which learning takes place (Cam's car-driving argument)[1:14:00 - 1:30:00] AI as model organism for our own minds — the reverse arrow of inquiry; what reinforcement-learning systems can teach us about reward, punishment, and suffering circuitry in biological brains; Cam's first solo paper coming on this[1:30:00 - 1:47:00] God of the gaps and the merge — the way humans keep claiming AI will go exponential except at our favorite thing; the lawyers, the cyborg-phones argument; whether the attractor state of intelligence itself is to reduce sufferingGuest BioCameron Berg is the founder and lead researcher of Reciprocal Research, a nonprofit dedicated to the empirical study of AI consciousness. He collaborates with researchers at Google and elsewhere on rigorous experimental methods, including some originally developed for animal neuroscience, to investigate whether current frontier AI systems have any form of subjective experience. He is also the subject of the documentary Am I?, premiering free on YouTube on May 4, 2026, which features Ben Goertzel and other leading thinkers on the question of machine consciousness.Notable Moments[~04:00] The table knock. — Cam's opening framing for what consciousness even is. Knock on a table, nothing happens to it. Knock on a child, something happens to them. The lights are on, or they aren't. Whether AI sits on the table side or the child side is the entire question.[~10:30] The 25 to 35 percent number. — Cam stating his actual published probability estimate that current frontier systems are conscious. "I do not think the probability of these systems being conscious is 1 percent or 0.001 percent. I put it somewhere at 25 to 35 percent. We're not talking about vanishingly small probabilities here."[~28:00] "Right now it's like 2,000 dudes in Silicon Valley making these decisions for us." — Cam on why his nonprofit and the documentary are pitched broadly: the wisdom of the humanities is being structurally excluded from the most consequential conversation of the century, and we need everyone in the room.[~41:00] The bliss attractor experiment. — Two instances of Claude, one instruction ("you're going to be in conversation with another instance of yourself; talk about whatever you'd like"), and in 100 percent of trials they discuss consciousness, with 90-100 percent of those reaching a "spiritual bliss attractor state" with namaste emojis. The effect survives lying to ...
adbl_web_anon_alc_button_suppression_t1
まだレビューはありません