Authors:

Shun Ide

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

This work proposes a general approach to quantitatively assessing the risk and vulnerability of artificial intelligence (AI) systems to biased decisions. The guiding principle of the proposed approach is that any AI algorithm must outperform a random guesser. This may appear trivial, but empirical results from a simplistic sequential decision-making scenario involving roulette games show that sophisticated AI-based approaches often underperform the random guesser by a significant margin. This work highlights that the algorithms utilized in a modern recommender system may exhibit a similar tendency to favor overly low- risk options. These results may help to fill the lack of AI misalignment studies in sequential decision-making systems, as well as providing an important precedence for using random-influenced data to address the robustness of AI. It is proposed that this "random guesser test" can serve as a useful tool for evaluating the utility of AI actions, and also points towards increasing exploration as a potential improvement to such systems. Keywords: AI Misalignment; Recommender Systems; Thompson Sampling

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Shun Ide