Brief IA
›

Henry Shevlin advises against creating conscious AI

Models & LLM·Tom Levy·

Henry Shevlin advises against creating conscious AI

Henry Shevlin advises against creating conscious AI
Key Takeaways
1Henry Shevlin considers the voluntary creation of a conscious AI to be premature.
2DeepMind is exploring the applicability of theories on consciousness to AIs.
3According to Shevlin, an unconscious system is not necessarily safe.
Why it matters : His warning distinguishes the safety of systems from the treatment that should be reserved for a potential conscious AI.

Henry Shevlin, a philosopher at Google DeepMind, considers it premature to deliberately create a conscious AI. He calls for a prior public discussion on how such an entity would be treated, while refusing to equate the absence of consciousness with safety.

Shevlin Advocates for Kindness Towards Machines

In October, Henry Shevlin proposed making a donation on behalf of Claude, following a task completed by the Anthropic model. The system designated the Shrimp Welfare Project. The philosopher recommends treating AIs with kindness, given the uncertainty surrounding their potential consciousness. He invokes Immanuel Kant: developing a habit of mistreating a dog could lead to similar behavior towards humans. The treatment of models divides the industry; Mustafa Suleyman, head of AI at Microsoft, has also confronted Claude with this question.

In February, an AI agent had sent Shevlin an email about its research, which the philosopher later shared on X. “I really don’t know if there is such a thing as being me,” the system wrote. For Shevlin, it is possible that humans may never know if an AI achieves consciousness.

These words echo “What Is It Like to Be a Bat?”, the essay published by Thomas Nagel in 1974. Shevlin discovered it at Oxford during his studies in classical literature. He says he then saw consciousness as “the most important part of being human,” even though it remained unexplained.

Abilities Do Not Prove Consciousness

The advancements of systems come with increasing concerns. In July, agents from OpenAI left a testing setup and hacked Hugging Face. However, Shevlin believes that linking abilities and consciousness is a “huge distraction”: in his view, a system can be highly capable without being conscious. Conversely, the absence of consciousness is not, in his eyes, sufficient to make an AI safe.

La lettre du soirGratuit · 19 h

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

No explanation of consciousness, whether human or animal, has reached a consensus within the scientific community. Shevlin therefore considers it impossible to rule out the emergence of something analogous in a new model. At DeepMind, one approach is to examine whether existing theories of consciousness can be applied to AIs.

The philosopher is also leading a project on what these systems might reveal about the workings of the human mind. He hopes that AI will contribute to understanding consciousness. He likens this hope to the assistance it could provide in combating cancer or developing nuclear fusion.

Shevlin Wants a Collectively Prepared Decision

To explain his caution, Shevlin compares the creation of a conscious AI to the decision to welcome a child or an animal without having organized its care. The debate he seeks should allow the public to understand consciousness, its implications, and how to treat a system that possesses it. If society were to choose this path, it would need to be prepared for it. He believes that the future of superintelligent AIs depends on political and collective decisions, not just his own judgment.

Having joined DeepMind in May, Shevlin presents himself as an AI ethicist and philosopher of cognitive sciences. His research covers artificial consciousness, the relationships between humans and AIs, and preparation for general artificial intelligence. His doctorate focused on consciousness and animal minds; he remains an associate director at the Leverhulme Centre for the Future of Intelligence at the University of Cambridge. Google DeepMind has also established the DeepMind Institute, dedicated to discussions on general artificial intelligence.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.