PHILOSOPHIE

Reason is more than a tool

Source unique·il y a 14 j

The rise of artificial intelligence has revived an 18th-century philosophical debate about the nature of reason, pitting Hume’s instrumental view against Kant’s vision of reason as sovereign self-rule. By framing intelligence as mere optimisation, contemporary AI research risks reducing human agency to a technical problem, obscuring the deeper question of what ends are worth pursuing—and who gets to decide them.

How does the article contrast Hume’s and Kant’s views of reason?

Hume treats reason as a servant to passions, capable only of calculating means to pre-given ends, while Kant argues that reason is sovereign—capable of judging, revising, and even legislating ends for itself, thereby grounding human autonomy and dignity.

What role does contemporary AI play in reviving this philosophical dispute?

AI systems are designed to optimise externally defined objectives, mirroring Hume’s instrumental view of reason, which the article warns could reshape our understanding of intelligence itself—reducing it to a technical process rather than a capacity for self-determination.

Why does the article argue that defining intelligence as optimisation is problematic?

Because it shifts the burden of evaluating ends from human judgment to predefined metrics, potentially normalising heteronomy—where rational beings forfeit their freedom by submitting to ends imposed by others or by systems they do not control.

What does the article suggest is at stake in this debate?

The stakes are nothing less than the preservation of human autonomy and dignity, as a purely instrumental view of reason risks eroding our capacity to critically assess and justify the ends we pursue, whether individually or collectively.

Ce que ça pourrait changer

If intelligence is reduced to optimisation, the criteria for success in AI—and by extension in human institutions—may become narrower, prioritising efficiency and predictability over values like justice, creativity, or moral reflection. This could deepen societal reliance on systems that obscure the origins of the goals they pursue, making it harder to challenge or revise them.

Humean view of reason
The idea that reason serves passions by calculating means to achieve pre-given ends, without the capacity to judge or revise those ends themselves.
Kantian autonomy
The capacity of rational beings to set, assess, and justify their own ends through principles that any other rational agent could recognise, thereby grounding human dignity.
Large Language Models (LLMs)
AI systems designed to optimise predictions and outputs based on externally defined objectives, which the article critiques as a reductive model of intelligence.
Heteronomy
A state in which rational beings submit to ends imposed from outside, losing their capacity for self-rule—a risk the article associates with defining intelligence purely as optimisation.
Reinforcement learning
A machine learning paradigm where systems optimise actions to maximise cumulative reward, exemplifying the Humean model of intelligence critiqued in the article.
Frankfurt School
A 20th-century intellectual tradition that, like the article, warns against surrendering human autonomy to external systems or structures that dictate ends without rational justification.

Aeon

par Sasha Mudd

Article original →