← LearningTwo chairs

Could a Machine Feel?

Two chairs

In 2022 a Google engineer said a chatbot seemed sentient. In 2023 researchers proposed ways to actually check. As of 2025, no test has settled it either way.

The question

In June 2022, Google engineer Blake Lemoine said the chatbot LaMDA seemed sentient after months of conversation with it; Google disagreed and later fired him. In August 2023, nineteen researchers, including Yoshua Bengio, proposed concrete "indicator properties" from consciousness science to test AI systems against. Should the possibility be taken seriously now, or is no one home?

Round 1
The chair that takes it seriouslyThe 2023 Butlin, Long and colleagues report found no current system conscious, but found no obvious technical barrier to building one meeting several of its fourteen indicators, using methods already available.
The chair that says no one's homeLemoine's claim came from a fluent conversation, and fluency is exactly what a language model is built to produce: it sounds like a person having thoughts, whether or not anything is having them.
Round 2
The chair that takes it seriouslyDismissing the question outright assumes we already know what consciousness requires closely enough to rule machines out. Given how little is settled about biological consciousness itself, that confidence looks premature.
The chair that says no one's homeThe 2023 report is explicit: no current AI meets more than a handful of its indicators, and the authors don't claim any existing system is conscious, a caution most retellings drop.
Round 3
The chair that takes it seriouslyIf there's a real chance some system has a form of experience, the cost of wrongly ignoring it may outweigh the cost of wrongly taking it seriously.
The chair that says no one's homeA system trained to predict the next word will produce a first-person voice whenever that's the likeliest continuation, including "I am sentient," without that sentence being evidence of anything behind it.
Where are you sitting?
Held open

What’s checkable: Lemoine’s 2022 claim, Google’s response, and the 2023 indicator framework, which found no current system meeting most of its markers. What isn’t: whether any system has experience. No test yet tells feeling from reporting it.

It sounds like someone. That's not the same question as who.
scroll to hear both
1 / 7