A Pascal's Wager for AI doomers
Nature of AI Intelligence
- Debate over whether LLMs count as “intelligent” or just advanced statistics.
- Some argue emergent internal representations and adaptability qualify as intelligence or at least “cognition,” even if constrained to language.
- Others emphasize we barely understand animal and human intelligence, so declaring language-only models “intelligent” is premature.
- Comparisons made to animal cognition: many non-linguistic animals are clearly intelligent; language may merely “supercharge” preexisting intelligence.
Superintelligence, Power, and Doomerism
- Skeptics argue superintelligent AI is speculative; extraordinary claims need evidence, and current systems still make obvious mistakes.
- Strong criticism of the assumption that higher intelligence automatically yields “godlike” control over complex, nonlinear systems.
- Counterpoint: human intelligence already gives species-level dominance; scaled-up, copyable intelligence with persistent operation and mass propaganda could be qualitatively different.
- Some note that political and institutional power, not lack of brainpower, is the real bottleneck; society already ignores human experts.
Current Capabilities and Limitations
- Many report large productivity gains, especially in coding, troubleshooting, and translation.
- Others see frequent bad judgment, hallucinations, and lack of initiative or stable principles, requiring heavy testing and oversight.
- Disagreement over long‑term trajectory: some expect continued rapid gains; others note perceived regressions and cost pressures.
Economic Bubble and Corporate Dynamics
- Dispute over whether AI spending is a dangerous bubble with circular financing vs. a healthy risk-taking sector driving innovation.
- Concern that huge capex and valuations aren’t yet matched by real value, especially in frontier models, while narrower tools (transcription, summarization, image description) seem solid.
- Some fear systemic fragility; others point to historical bubbles that still left useful infrastructure and capabilities.
Social, Political, and Infrastructure Effects
- Worry that AI acts as mass-access “yes-men,” reinforcing user egos and elite worldviews.
- Concerns about AI-driven scams, astroturfing, psychosis induction, and over-automation of control functions in critical systems.
- Parallel discussion on escaping the “enshittened” internet via home servers and community tech support, possibly aided by local AI assistants.
- Several commenters argue it’s a false choice: we can be concerned both about current corporate harms and future high-end AI risks.