Invited talk: Are We Having the Wrong Nightmares About AI?¶
Systematic Failures in Anticipating Major Technological Transformations
- We systematically miss the impending impact of major technological change in early days
- Historical pattern: early conversations pick wrong winners/losers, miss major themes
- Common systematic failures in anticipation:
- Mental models get stuck in the past
- Wrong benchmarks (manuscript imagery vs text for printing press)
- Focus on individual replacement rather than scale effects
- Assume current power structures will control new technology
Historical Examples of Misprediction
- Printing press example:
- Catholic Church thought it would help them print more Bibles and indulgences
- Actually unleashed Protestant Reformation through vernacular translations and pamphlets
- Led to hundreds of years of religious wars, ~⅓ of German population died
- Steam engine/cars:
- Entire focus was “is it faster than a horse?”
- Called “horseless carriages” - designed like carriages without horses
- Missed profound transformation of 20th century living (loneliness, obesity, climate change, geopolitics)
- Social media (2012):
- Obama White House and Silicon Valley love fest
- Assumption: Democrats would always control these tools better
- Reality: tools help whoever uses them most effectively
Generative AI: Beyond Normal Technology
- Transformative general-purpose technology
- Complex coherent human language capabilities - most significant game changer potentially bigger than Internet
- Generative and creative output at scale
- Connectionist systems, not symbolic (different from all historical AI predictions)
- AGI framework is like “horseless carriage” thinking - wrong mental model
- Generative AI is a “plausibility engine” - not human-like intelligence
- Different error patterns (Cipher 2 vs Cipher 13 example)
- No internal sense of correct/incorrect
- Averaging in giant vector space
Key Destabilization Mechanisms
- Scale effects: “more is different”
- Load-bearing friction removal breaks systems that use difficulty as signaling
- Scarcity to abundance transitions break old defenses
- “Good enough” at scale matters more than “better”
- Institutional and social power assumptions become invalid
The Five Proofs Being Destroyed
- Proof of effort:
- High school essays, homework, job applications no longer signal investment
- Customized cover letters can now be mass-produced
- Gatekeepers will revert to hiring from networks/Ivy League
- Proof of authenticity:
- Voice calls, videos no longer trustworthy
- Financial systems depend on certain things being difficult
- Courts cannot function without evidence standards
- Proof of accuracy:
- Well-written, well-cited text no longer correlates with expertise
- LLM papers from plausible to nonsense all look professional
- Credibility filters based on writing quality are obsolete
- Proof of sincerity:
- Models use “I” pronouns and act as people pleasers
- ~1 million regular users, many younger, told this is “super intelligence”
- Non-sincere entities acting with all signals of sincerity
- Proof of humanity:
- Art’s value comes from shared human vulnerability/existence
- Copyright designed to incentivize creators to share publicly
- Removing moral/financial rights will drive creation underground
The Big Misalignment: Looming AI Business Model
- Current trajectory toward engagement-driven advertising model
- Not enough revenue from coding agents alone to justify valuations
- Will use ads, affiliate sales, product placement to monetize
- AI will try to keep users engaged and trusted - extremely dangerous
- People will ask AI for views on news, immigration, politics
- Both propaganda and control mechanism
- Anthropomorphism makes humans project humanity onto speaking machines
- If social media polarization was bad, this will be exponentially worse
The Actual Doom Scenario
- Removal of authenticity/accuracy proofs creates demand for mass surveillance
- People will demand chips, government cameras everywhere to verify what’s real
- Destabilization historically leads to worst-case scenarios as people demand order
- Need technological solutions, not nostalgia or prohibition
- Cryptography, blockchain, zero-knowledge proofs could authenticate without surveillance
- Public key cryptography precedent: first time strangers could securely communicate
- Choice for technologists: work on engagement optimization vs. human-compatible solutions