Auctioning Intelligence

Auctioning Intelligence: Experiments with AI at the Edge of Reason explores a fundamental question: When placed in strategic, incentive-driven settings like auctions, can large language models such as ChatGPT behave like humans?

While LLMs have demonstrated impressive performance across language tasks and professional exams, their behavior in competitive decision-making environments reveals striking differences from human participants. Through a series of carefully designed auction experiments, this book shows that even the latest versions of ChatGPT, despite their linguistic fluency, consistently diverge from human strategies when real stakes are involved.

The findings highlight a key reality: LLMs are optimized to produce coherent language, not to maximize rewards. Without emotional heuristics like risk aversion or the instinct to win, and drawing on vast but imperfect human-generated data, ChatGPT's actions remain fundamentally different from ours in strategic settings. These results challenge the assumption that more training or broader knowledge alone can bridge the gap between AI and human decision-making.

While the experiments may evoke comparisons to a Turing test, the core focus here is not whether AI can ""appear human,"" but whether it can act human when true incentives and strategic reasoning are required. In doing so, Auctioning Intelligence invites readers to rethink what human-like intelligence really means-and whether AI can ever truly replicate it.

Drawing inspiration from science fiction visions, including Philip K. Dick's Do Androids Dream of Electric Sheep?, this book offers a timely and thought-provoking investigation into the evolving frontier between artificial and human minds.