AI predictions are never facts, but veiled commands, says philosopher

AI predictions are never facts, but rather veiled commands, according to philosopher Carissa Véliz. In her book "Prophecy," Véliz uses the historical example of Jeanne Calment, who outlived the notary who paid her a monthly stipend in exchange for her apartment, to illustrate how predictions, no matter how plausible, cannot account for unpredictable events or "black swans."

“

A prediction is never a fact, no matter how plausible it sounds.

This tendency to present probabilistic calculations as established knowledge is a "dangerous mistake" of the AI era, Véliz warns. AI predictions now influence crucial decisions, determining who gets a mortgage, who remains insured, who is flagged as a potential fraudster, and who gets a chance to deviate from algorithmic forecasts. "Life can never be fully captured by algorithms," she states, yet AI is propelling society into a world where predictions and models hold increasing sway.

Véliz argues that this reliance on AI is not only unjust and a catalyst for authoritarianism but also poses a significant danger to societal and economic stability. Her book draws parallels between the rise of AI predictions and the age-old human endeavor to foresee the future, a pursuit historically undertaken by astrologers and prophets with limited success.

“

Life can never be fully captured by algorithms.

Despite her reservations, Véliz, who researches AI applications at the University of Oxford, acknowledges the impressive capabilities of large language models. However, she describes them as essentially "fancy autocomplete," adept at predicting the next word in a sequence. She maintains that recent rapid advancements in AI do not alter this fundamental nature.

“

Ultimately, it's just fancy autocomplete.