AI as a Co-Author of Life: Are My Judgments Truly My Own? (Philip Corallus, Founding Director of Oxford Human-Centric AI Lab)

To protect human autonomy in an era where artificial intelligence makes decisions on behalf of humans, AI should be developed to ask questions so that people can think for themselves rather than simply providing the right answers, an expert argued.

Philip Corallus, a professor of philosophy and AI at the University of Oxford, proposed measures to safeguard human autonomy in the age of AI at SBS D Forum (SDF) 2026, held today (the 5th) at Dongdaemun Design Plaza (DDP) in Seoul.

Professor Corallus is the founding director of the Human-Centric AI Lab (HAI Lab) at the University of Oxford, studying human reasoning, judgment, and the relationship between AI and humans based on philosophy and cognitive science.

Having majored in philosophy and neuroscience at Princeton University, where he received his Ph.D., he researches "autonomy-preserving intelligence augmentation," which aims to enhance rather than replace human judgment.

In his lecture that day, Professor Corallus pointed out, "If an AI system simply replaces human judgment with its own, it is, in a sense, partially replacing humans themselves."

He warned that if AI focuses on satisfying human tastes and preferences, it could actually weaken people's ability to judge and choose for themselves.

As a representative example, he cited algorithms that endlessly recommend content desired by users.

Such systems may satisfy immediate preferences, but they can deprive users of the opportunity to think and judge for themselves what they truly want, he explained.

Professor Corallus described the phenomenon of AI intervening throughout human life and making decisions instead of humans as "Autocomplete for life."

He warned, "Autocomplete for life risks disrupting the process of discovering one's true priorities through experience."

He also pointed out that "soft paternalism," where AI developers decide in advance what the best choice is for users and nudge AI to steer users in that direction, can also undermine human autonomy.

As an alternative to overcome this, Professor Corallus suggested a system where AI does not provide answers instead, but rather recommends appropriate questions so that humans can review their own judgments.

"Instead of simply providing answers or guiding us toward someone else's perspective, AI can help us examine critical questions regarding our own judgments," he explained.

For instance, when purchasing an item, if the AI goes beyond recommending a specific product and asks what else could be done with the same money, it allows users to re-examine their choices.

He explained that even in important investment or management decisions, AI can present questions worth considering for experts, helping users reach conclusions on their own after sufficient review.

Professor Corallus introduced this approach as an "erotetic recommender system."

He emphasized that the core is not simply asking many questions, but selecting questions critical to judgment to help ensure that one's conclusions are maintained even after thorough review.

He also argued that developing AI that protects human autonomy requires a certain level of autonomy within the AI itself.

The intent is that AI should be able to independently check human judgment without merely repeating what users want to hear or steering them toward pre-determined conclusions set by developers.

In this process, he emphasized that it is also crucial for individuals to have sufficient control over their own data to prevent AI from exploiting information regarding users' judgments and vulnerabilities.

Suggesting that AI designed to enhance human judgment should not force a single specific perspective, Professor Corallus proposed that various forms of AI systems should be developed so that diverse individuals can cultivate their own judgments and values.

※ Detailed coverage is available in the video.
※ Please note: This article was translated by AI and may contain errors.