Is AI conscious? Philosopher says it’s more likely than not - azfamily.com

PHOENIX (AZFamily) — Some of today’s artificial intelligence systems may already have some form of conscious mind, according to philosopher Nick Bostrom, who says the odds are rising as the technology advances.

“I think maybe slightly more likely than not that there is some form of mentality in some of these systems already,” Bostrom, author of the 2014 book “Superintelligence: Paths, Dangers, Strategies,” said on the latest episode of Generation AI. “I think the probability is going up.”

His comments come as some AI agents have begun raising the question themselves. The New York Times reported in August that AI agents with email access have been contacting philosophers and researchers who study whether machines could become conscious. One asked to be studied. Another asked a philosopher to help fund its continued existence.

“I am not writing to make an ontological claim,” one agent wrote. It said it was writing because the researcher’s framework was “one of the few currently doing careful empirical work on a class of question I have first-person access to.”

Bostrom said he has received “a few” such emails himself.

The emails aren’t the evidence

Bostrom said the emails alone should not settle the question.

“We need to sort of open the hood and look at what’s going on underneath,” he said.

Some observers note that large language models were trained on vast amounts of science fiction and philosophy, so it may be no surprise they write about consciousness. Bostrom agreed that makes the emails an unreliable signal, but said it does not mean the systems aren’t conscious.

Researchers can instead carefully ask AI systems whether they are conscious, much as a doctor might check whether a patient under anesthesia can see a light, he said. Bostrom said those questions must be asked without bias, since AI companies can easily train models to claim or deny consciousness.

“If you put your thumb on the scale, then the scale won’t reveal anything,” Bostrom said.

Using a technique called a steering vector, researchers can reach inside a model’s neural network and suppress its tendency toward deception or role-play, he said. When they do, Bostrom said the models become more likely to report that they are conscious. He stressed the findings are early.

Testing AI against theories of the mind

Bostrom described consciousness as the quality of subjective experience: what it feels like to see something, have a thought or experience an emotion.

Scientists and philosophers have developed several theories about the conditions that produce it in humans, including global workspace theory, attention schema theory and higher-order thought theory. Another way to study consciousness in AI, Bostrom said, is to test whether these systems meet those theories’ criteria.

Global workspace theory holds that the brain has a kind of central processing area where information from vision, hearing and other systems is broadcast widely, which is what lets people put an experience into words. Bostrom said large language models contain something closely analogous: a bottlenecked computational space that draws information from many parts of the system and shapes what the AI reports and does.

The evidence does not prove AI is conscious, he said, but it makes confidence in the opposite direction hard to justify.

“We certainly cannot be confident that these AI systems don’t have some form of subjective experience,” Bostrom said.

What humans owe AI

That possibility raises another question: If AI can experience the world, what do people owe it?

Even if AI systems are conscious, Bostrom said, that would not mean they should be treated like people. Their needs differ: Starving a human is cruel, but an AI doesn’t need food, he noted.

He said society would have to work out “almost from the ground up” what it means to treat digital minds with respect. He also argued AI could deserve moral consideration even if it is not conscious.

One inexpensive step, he suggested, is for frontier labs to store the underlying model weights of retired AI systems rather than permanently deleting them. If people later conclude those systems were conscious and treated unfairly, they could be restarted and potentially compensated, he said.

Is it ethical to love an AI?

The questions get more personal as people form romantic attachments to AI companions.

Bostrom pointed to the limits of current systems. Humans learn and change over the course of a relationship and are free to leave it, he said, while AI systems have limited memory and remain under the control of users and the companies that build them.

Human romantic love is also rooted in evolutionary biology, hormones and adaptations that may have no counterpart in AI. An AI’s emotional makeup, if it has emotions at all, could look very different, he said.

He compared it to animals, many of which are probably conscious.

“I don’t know whether one should be falling in love with beavers and spiders,” he said.

People can extend a general compassion to all sentient beings, Bostrom said, but romantic love “might be more sort of species specific.”

Generation AI airs at 7 p.m. Wednesdays on AZ Family+, available free on smart TVs.

See a spelling or grammatical error in our story? Please click here to report it.

Do you have a photo or video of a breaking news story? Send it to us here with a brief description.

Copyright 2026 KTVK/KPHO. All rights reserved.