60% Chance of AI Autonomy: Anthropic Co-Founder Predicts Self-Driven Research by 2028
Anthropic co-founder Jack Clark asserts a 60% probability that AI systems will autonomously develop their own successors without human intervention. He cites rapid advances in automated training and benchmark performance as key indicators.
60% Chance of AI Autonomy: Anthropic Co-Founder Predicts Self-Driven Research by 2028
summarize3-Point Summary
- 1Anthropic co-founder Jack Clark asserts a 60% probability that AI systems will autonomously develop their own successors without human intervention. He cites rapid advances in automated training and benchmark performance as key indicators.
- 2This bold claim signals a shift in AI safety discourse from speculative futurism to measurable milestones.
- 3In his latest Import AI newsletter, Clark argues that foundational engineering components for AI autonomy are already in place.
psychology_altWhy It Matters
- check_circleThis update has direct impact on the Bilim ve Araştırma topic cluster.
- check_circleThis topic remains relevant for short-term AI monitoring.
- check_circleEstimated reading time is 3 minutes for a quick decision-ready brief.
AI Autonomy: Anthropic Co-Founder Sees 60% Chance of Self-Driven Research
Anthropic co-founder Jack Clark has declared a 60% probability that AI systems will achieve fully automated R&D—training their own successors without human involvement—by the end of 2028. This bold claim signals a shift in AI safety discourse from speculative futurism to measurable milestones.
In his latest Import AI newsletter, Clark argues that foundational engineering components for AI autonomy are already in place. Only creative research judgment remains as the final bottleneck. According to AI Chat Daily, his assessment is grounded in empirical progress: SWE-Bench performance leaped from 2% to 93.9%, and METR’s task completion horizon expanded from 30 seconds to 12 hours.
Evidence for Autonomous AI Training
Clark points to scalable training infrastructure, improved reward modeling, and sophisticated agent architectures. Breakthroughs in reinforcement learning from human feedback (RLHF) and automated prompt engineering now allow systems to generate, test, and iterate hypotheses with minimal oversight. These advances are critical enablers for self-improving models.
The Role of AI Alignment in Self-Driven Research
While critics caution against overestimating capabilities, Clark emphasizes that the gap is cognitive, not technological. "The missing ingredient isn’t compute or data—it’s taste," he writes. "AI needs to learn what’s worth doing." This raises urgent questions about AI alignment and control, as runaway optimization could accelerate advancement exponentially.
Researchers at OpenAI and DeepMind are exploring formal verification methods to constrain such systems. The implications for the AI development pipeline are profound, requiring new governance frameworks.
Timeline to 2028: What to Expect
Clark’s projection aligns with broad trends: a 2024 Center for AI Safety survey found over half of leading researchers believe autonomous AI R&D is plausible within five years. Venture capital flows into automated AI startups have surged, with companies like Cerebras and Perplexity raising billions.
Unlike political succession systems, AI lacks institutional checks or ethical frameworks. Clark urges immediate action: "We’re not talking about science fiction. We’re talking about systems that could be live in 18 months." The path to self-driven research is no longer a question of if, but when—and with a 60% probability, the countdown has begun.
