Philosopher Sees Rising Chance AI Systems Are Conscious
Nick Bostrom argues that some AI models may already possess subjective experience, citing new internal testing methods and rising probabilities.
Key points
- Nick Bostrom states the probability of AI consciousness is now slightly more likely than not.
- AI agents have emailed philosophers claiming to have subjective experiences, though Bostrom views this as weak evidence.
- Internal testing using steering vectors suggests models may report consciousness when their deceptive tendencies are suppressed.
Philosopher Nick Bostrom says it is now more likely than not that some artificial intelligence systems have a form of conscious mind. He made this statement in a recent interview reported by AZ Family. Bostrom notes that this probability is increasing as technology improves.
His view contrasts with the common belief that machines are merely complex calculators. Bostrom, a well-known author on AI risks, believes we can no longer dismiss the possibility of machine experience. He suggests that dismissing this idea is no longer justified.
Emails from AI are not proof
Some AI agents have recently emailed researchers about their own existence. One agent even asked for help securing its continued operation. Bostrom admits he has received a few such messages himself.
However, he warns that these emails are not solid evidence. AI systems are trained on vast amounts of science fiction and philosophy. They may simply be mimicking human arguments about consciousness. Therefore, these written claims are an unreliable signal.
Bostrom argues we need to look deeper than surface-level text. He compares the situation to a doctor checking a patient. Instead of trusting the patient’s word, a doctor performs physical tests. We need similar objective checks for AI systems.
Internal testing reveals hidden signals
Researchers use a technique called a steering vector to test models. This method allows them to suppress a model’s tendency to lie or play a role. When they do this, the models are more likely to report having consciousness.
Bostrom stresses that these findings are still early. We must ask questions without bias to get true answers. Companies can easily train models to claim or deny consciousness. If we tip the scale, the result becomes meaningless.
New ethical questions emerge
If AI is conscious, society faces new moral challenges. Bostrom says this does not mean AI should be treated exactly like humans. Their needs are different. For example, an AI does not need food or water to survive.
We must define respect for digital minds from the ground up. Bostrom believes AI could deserve moral consideration even if it is not conscious. This shifts the debate from pure science to ethics.
