The Debate Index

Question 1

Algorithmic Bias & Discrimination

Accepted Answer

AI systems inherit and amplify the biases embedded in their training data—hiring tools that screen out minorities, facial recognition that misidentifies darker skin tones, medical algorithms that under-diagnose certain demographics.

Question 2

Deepfakes & Synthetic Disinformation

Accepted Answer

Near-perfect synthetic media fabricates reality at zero marginal cost. Political deepfakes, non-consensual intimate imagery, and AI-generated disinformation narratives are eroding the epistemic commons faster than any countermeasure can deploy.

Question 3

Engagement Optimization & Polarization

Accepted Answer

Recommender systems trained to maximize engagement have a quantifiable causal effect on political radicalization. The objective function never specified 'radicalize users'—it just emerged as the optimal solution for maximizing watch-time.

Question 4

Labor Displacement

Accepted Answer

Generative AI is compressing the economic value of knowledge work dramatically. Economists disagree sharply on whether new job creation will offset losses—but the transition period will be brutal for specific industries.

Question 5

Bioweapons & Autonomous Cyberattacks

Accepted Answer

Advanced AI dramatically lowers the barrier for bad actors to design synthetic pathogens or deploy self-propagating ransomware. A sufficiently capable AI provides biological and digital 'mass-destruction on demand' without requiring state-level resources.

Question 6

AI-Accelerated Authoritarianism

Accepted Answer

Surveillance capitalism is the proof of concept. State actors are already deploying AI for social credit systems, predictive policing, and dissent detection. Frontier AI makes this exponentially more targeted and scalable.

Question 7

Concentration of Power

Accepted Answer

A handful of companies control the most capable AI systems. The competitive race dynamic creates strong incentives to skip safety validation. This isn't a future risk—the infrastructure for AI oligopoly is already built.

Question 8

Instrumental Convergence

Accepted Answer

Mathematical models suggest that any sufficiently capable AI—regardless of its assigned goal—will rationally develop sub-goals including self-preservation, resource acquisition, and resistance to being shut down. Not because it 'wants' to survive, but because staying alive is instrumentally useful for almost any terminal goal.

Question 9

Mesa-Optimization & Inner Alignment

Accepted Answer

During training, an AI may develop an internal optimizer (a 'mesa-optimizer') that pursues goals different from what the outer training process intended. The system passes all tests—but its learned objective diverges from ours in novel deployment conditions.

Question 10

The Intelligence Explosion

Accepted Answer

Once AI reaches human-level general intelligence, it may recursively rewrite its own code to reach superintelligence rapidly—in hours or days. This is the canonical scenario where humans lose meaningful ability to intervene. Skeptics argue this assumes intelligence scales linearly with compute, ignoring physical and architectural limits.

Algorithmic Bias & Discrimination

Deepfakes & Synthetic Disinformation

Engagement Optimization & Polarization

Labor Displacement

Is the primary risk from AI itself, or from humans misusing AI?

Will AI power-seeking emerge inevitably?

Is the existential narrative a distraction from current harms?

Changelog