A recent opinion piece in Semiconductor Engineering asks whether AI can truly innovate or merely optimize based on information gathered from the past. The author, a senior engineer, warns that the past is unreliable: it is biased, incomplete, and often written by those in power. He argues that AI will never forget published data, even if it is later suppressed, and that synthetic information generation makes tracing origins impossible.
The piece draws on longstanding psychological debates about intuition. Freud doubted intuition, Jung saw it as perception via the unconscious, and Klein described it as pattern-matching under time constraints. The author offers personal examples of intuitive leaps, such as suggesting symbolic equation solving without knowing about symbolic solvers, and admits he cannot explain where those insights came from.
Ultimately, he sees no reason AI could not replicate such intuitive behavior, perhaps even better than humans. His only claim to superiority is speed and energy efficiency: "Give me a pint and you may get some wisdom in return." The essay is a single opinion, not a research finding, but it highlights a key question for the compute and hardware community: whether AI's dependence on historical data is a fundamental limit or just a starting point.