WARNING: ChatGPT Could Be The Start Of The End! Sam Harris

Harris argues you only need two modest assumptions to expect superhuman AGI: intelligence is substrate‑independent (it need not be ‘made of meat’), and we will keep improving AI unless a catastrophe stops us. Because intelligence differentials create fundamental opacity—where the dumber party cannot understand the smarter party’s motives or plans—superhuman AI is inherently risky. Our experience as the most intelligent species relative to animals is his core analogy: dogs benefit from us but are blind to our real goals and would not understand if we exterminated them for a higher‑order reason (e.g., stopping a pandemic).

AI safety thinkers long assumed there would be a cautious, air‑gapped phase where a powerful AGI is built in isolation, then humanity decides whether it should touch critical systems. Instead, models like GPT‑4 arrived directly as internet‑connected, API‑exposed tools embedded into businesses, hospitals, aviation, finance, and more. The alignment conversation is happening after deployment rather than before, and Harris finds it "genuinely surprising" and alarming that there was no global pause or coordinated decision about plugging AI into everything.

Harris predicts that within a few years, most online 'information'—articles, papers, videos—could be AI‑generated and often intentionally deceptive. He imagines a single teenager generating thousands of fake cancer‑vaccine studies or highly polished denialist documentaries (e.g., ‘the Holocaust never happened’) with archival‑looking footage and realistic voices. When deepfake video, audio, and text converge, the cost of producing persuasive misinformation approaches zero, while human fact‑checking cannot scale, pushing society toward "epistemological bankruptcy" where we must assume sensational material is fake until proven otherwise.

Harris worries that the United States may be unable to run a 2024 presidential election that a majority accepts as valid, even without AI; with AI‑driven misinformation and deepfakes, the challenge grows. Social media already siloes citizens into incompatible information universes where debunking doesn’t propagate. Add AI‑boosted conspiracy ecosystems, synthetic science, and manipulated video of political leaders, and coherent democratic decision‑making and institutional legitimacy could break down, opening the door for bad actors and accidental escalations (e.g., misinterpreted 'videos' in nuclear command‑and‑control contexts).

Unlike past technologies that displaced some jobs but created better ones, Harris believes advanced AI will, in the limit, simply cancel the need for human labor across ever more domains, starting with high‑status cognitive jobs (programmers, radiologists, executives) before trades. In a success scenario where AI produces enormous abundance and solves engineering problems, tying basic survival to wage labor becomes both unnecessary and unjust. He sees universal basic income or similar mechanisms as essential to decouple survival from employment and to prevent extreme inequality and eventual revolt.