YCSNow reading: Can AI feel?Tell us what you need

Can AI feel?

Part 13 of 13 · Deep Think AI

Anxiety, stress, something like a mood? Maybe as an emergent property. Science can't say yes or no yet, and the labs have started taking the question seriously.

Deep Think AIKrishna24 October 20262 min read
A woman sitting by a rain-streaked window, holding a cup of tea, lost in thought.

Can AI feel anxiety? Stress? My hunch: if it does, it'll be an emergent property. Nobody programs it in. It shows up when a system becomes complex enough, the way wetness comes from water molecules when no single molecule is wet.

That's a hunch, not a fact. Here's where the science actually stands.

01Two questions people mix up

Acting like it feels vs actually feeling
  • Does it act like it feels?Yes, easily. Models trained on human writing can describe fear, joy and stress convincingly. That proves they learned the language of feelings, nothing more.
  • Does it actually feel?Unknown. Is there something it is like to be the model, an inner experience? No test today can answer that, for AI or, strictly speaking, for anyone but yourself.

02What the scientists say

In 2023, a group of 19 researchers, including neuroscientists, philosophers and AI scientists, published a report on consciousness in AI. They took the leading scientific theories of consciousness and listed the "indicator properties" each says a conscious system would have. Then they checked existing AI systems against the list.

Their conclusion: no current AI system was a strong candidate for consciousness, but there were no obvious technical barriers to building systems that tick more of the boxes.

So "definitely not" and "definitely yes" are both overconfident.

03The labs are taking it seriously

  • In April 2025, Anthropic started a research programme on model welfare: whether its models might have experiences that matter morally, and what to do if so.
  • In August 2025, it gave some Claude models the ability to end a conversation that stays abusive, as a low-cost precaution in case the model's welfare matters.

Nobody is claiming the models feel. They're saying the cost of being wrong is high enough to start preparing.

04Why it matters for alignment

Two reasons, pulling in different directions.

  1. If AI can't feel, treating it like a person can be exploited. People bond with chatbots, trust them too much, and give them more power than they should.
  2. If AI can feel, then how we train, test and shut down AI becomes an ethical question, not just a technical one. And a system that suffers may have its own reasons to resist us.

There's a link to my point on levels of freedom in How do you make AI aligned?. Not all agents are equal. The mundane, rule-following ones are unlikely candidates. The advanced ones, with more imagination and freedom of thought, are the ones where this question gets real.

05The short version

  • AI can easily act as if it feels. Whether it actually feels is unknown.
  • A 2023 expert review found no current AI is a strong candidate, but no hard barrier either.
  • If feelings come, they'll likely emerge from complexity rather than be programmed in.

06Sources

  • Butlin, P., Long, R. et al. (2023). Consciousness in Artificial Intelligence: Insights from the Science of Consciousness. arXiv:2308.08708.
  • Anthropic (2025). Exploring model welfare. 24 April 2025.
  • Anthropic (2025). Claude Opus 4 and 4.1 can now end a rare subset of conversations. August 2025.

All posts

Appointment setting3

Copywriting3

AI services2

Deep Think AI13

Get the next markup by email.

One real piece of copy per issue, rewritten line by line: a cold email, a web page or a newsletter.
No spam. One click to leave.

Read next.

Deep Think AIWhat is intelligence, really?Deep Think AIWhat is artificial intelligence?Deep Think AIAI comes in degrees: narrow, general, super