People

A Harvard–BCG experiment found consultants using GPT-4 excelled on tasks within AI's reach, but erred more on complex problems beyond it. The reason: AI cannot grasp the unspoken standards that live in human experience. Prof. Xin Tong of HKU Business School argues that delegating to AI well means making these tacit judgments explicit. In the age of AI, each of us must learn to define what "good" means.

Investment bank analysts utilize AI to perform concurrent data processing, industry benchmarking, and preliminary drafting, while retaining responsibility for final validation and decision-making. Similarly, law firm partners can now execute due diligence tasks in minutes that previously required a full day of manual effort by interns. These advancements have moved beyond the realm of speculation.

Science fiction often depicts the threat of artificial intelligence (AI) through violent confrontations like those in The Terminator, but in reality AI’s encroachment is far quieter and more concealed than what we see on screen—it is not about physical destruction, but about something more fundamental to our existence: a systemic ceding of humanity’s “right to think,” which is silently taking place through our collusion with efficiency.

This article develops a conformal prediction method for classification tasks that can adapt to random label contamination in the calibration sample, often leading to more informative prediction sets with stronger coverage guarantees compared to existing approaches. This is obtained through a precise characterization of the coverage inflation (or deflation) suffered by standard conformal inferences in the presence of label contamination, which is then made actionable through a new calibration algorithm. Our solution can leverage different modelling assumptions about the contamination process, while requiring no knowledge of the underlying data distribution or of the inner workings of the classification model. The empirical performance of the proposed method is demonstrated through simulations and an application to object classification with the CIFAR-10H image data set.