This paper argues that building stronger intelligent systems capable of open-ended innovation requires creating, stabilizing, and reusing new representational primitives rather than merely searching within fixed frames.
The authors characterize the distance between current AI systems and genuinely open-ended intelligence through two specific gaps:
- The vocabulary gap: the difficulty of inventing and stabilizing new representational primitives instead of just recombining existing ones.
- The verifier gap: the difficulty of judging the value of a new primitive when its full payoff may only be visible after future reuse.
The paper interprets these gaps through a framework of intelligence as cognitive discrepancy reduction, distinguishing between intra-space transformations and generative transformations. It proposes a ladder of innovation autonomy and outlines directions for advancing open-ended AI, including objectives that reward useful representational change, persistent memory architectures, and adaptive verification mechanisms.