Reproduce “An Approximation of the Universal Intelligence Measure” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shane Legg, Joel Veness
AIQ approximates the Universal Intelligence Measure by sampling Brainfuck programs as environments and averaging agents’ rewards. Under one Brainfuck reference machine, scores rank simple reward-seeking learners below capable reinforcement-learning agents. Adaptive sampling reduces the share of programs with 10 or fewer instructions from almost 40% to 20%. These findings support the approximation here, not a machine-independent measure of intelligence.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.