Core conclusion
Six cognitive scientists led by Iris van Rooij of Radboud University in the Netherlands have published a paper in Computational Brain & Behavior giving a formal proof called the "Ingenia theorem": achieving human-level AI through learning (AI-by-Learning) is inherently an NP-hard problem.
Line of proof
- The proof grants a hypothetical engineer idealized conditions: perfect data, any present or future machine-learning method, and a very low performance bar
- Even so, the learning task can still be reduced to the Perfect-vs-Chance decision problem that Hirahara proved NP-hard in 2022
- Conclusion: creating systems with human(-level) cognitive abilities is "inherently computationally intractable"
Clash of views
- The authors state bluntly that the "factual AI systems" built in the near term are "at best decoys," and that the claim that "AGI is inevitable in the foreseeable future" is "false and misleading"
- This directly challenges the mainstream narrative that "brute-force scaling will work wonders and inevitably lead to AGI," pointing out that it rests on unproven assumptions rather than theorems
- On the constructive side, the authors do not reject computationalism; they argue for treating AI as a theoretical tool for cognitive science (formal models, thought experiments, complexity constraints) rather than as an engineering program of "rebuilding cognition on silicon"
Source: read the original. This text is a summary for reference only.
