Faei/ai-takeover-unlikely

Why an AI Takeover is Unlikely

What the latest research says

In our past article on the Distinct Independent Architecture Hypothesis, we suggested that many of our fears of an AI takeover are due to the combination of properties from two distinct groups of AI systems:

1. Non-Independent:

   1. Current Transformer architecture
   2. Tool-like

2. Independent:

   1. Future Hypothetical
   2. Human-like

Our fear of an AI takeover stems from improperly combining these two groups. We imagine an AI that possesses both the unpredictable, unaligned flaws of current tool-like models and the unstoppable autonomy of future human-like models.

Current AI Flaws

Current transformer-based LLMs (Large Language Models) are more capable than previous technologies but are still highly flawed.

Arc-AGI-3

ARC-AGI-3, a recently released benchmark, shows humans easily scoring 100% while frontier models score less than 1% as of March 2026.[^1]

Instead of solving static puzzles, agents must learn from experience inside each environment—perceiving what matters, selecting actions, and adapting their strategy without relying on natural-language instructions.

Remote Labor Index

The Remote Labor Index also shows that current AI agents cannot compete with remote human work as of April 2026.[^2]

While AI systems have saturated many existing benchmarks, we find that state-of-the-art AI agents perform near the floor on RLI. The best-performing model achieves an automation rate of only 4.17%. This demonstrates that contemporary AI systems fail to complete the vast majority of projects at a quality level that would be accepted as commissioned work.

AI Agent Reliability

AI agents are scoring well on benchmarks but still lack reliability according to this Feb 2026 report.[^3]

While rising accuracy scores on standard benchmarks suggest rapid progress, many agents still continue to fail in practice. This discrepancy highlights a fundamental limitation of current evaluations: compressing agent behavior into a single success metric obscures critical operational flaws. Notably, it ignores whether agents behave consistently across runs, withstand perturbations, fail predictably, or have bounded error severity.

The Hot Mess of AI

This recent April 2026 research supports the Hot Mess theory of AI misalignment for current AI systems.[^4][^5]

Consequently, scale alone seems unlikely to eliminate error-incoherence. Instead, as more capable AIs pursue harder tasks, requiring more sequential action and thought, our results predict failures to be accompanied by more incoherent behavior. This suggests a future where AIs sometimes cause industrial accidents (due to unpredictable misbehavior), but are less likely to exhibit consistent pursuit of a misaligned goal.

These four recent research papers suggest that current AI systems are likely too flawed to be capable of an AI takeover. Because transformer-based LLMs (Large Language Models) are fundamentally tool-like, scaling them up won't grant them independent, human-like agency.

Ultimately, while today's AI can undoubtedly cause harm through miscommunication, misuse or monopolisation, it lacks the fundamental architecture required for a sci-fi style takeover.

As a less confident claim: if the Distinct Interdependent Architecture Hypothesis holds true, all future iterations of current AI architectures will be limited to Lesser AGI, unable to reach full human parity.

[^1]: ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence: https://arxiv.org/abs/2603.24621 [^2]: Remote Labor Index: https://www.remotelabor.ai https://arxiv.org/abs/2510.26787v1 [^3]: Towards a Science of AI Agent Reliability: https://substack.com/home/post/p-189010640 https://arxiv.org/abs/2602.16666 [^4]: The Hot Mess of AI https://arxiv.org/abs/2601.23045 [^5]: The hot mess theory of AI misalignment: More intelligent agents behave less coherently https://sohl-dickstein.github.io/ 2023/03/09/coherence.html