AI Consciousness

An icebreaker

No current test can settle whether any AI system has real inner experience, and researchers who study this closely disagree, from a founder calling today's networks 'slightly conscious' to linguists calling that claim baseless. Both current claims and current denials go beyond what's actually known.

Ask one of these · tap to take it

What would even count as proof either way?
Should uncertainty here change how systems get treated?
Can something act aware without feeling anything?
Why do experts disagree so sharply on this?

copy the questions

Worth knowing · 1 of 5

What sparked the modern version of this debate?

In 2022, a prominent AI researcher posted that today's large neural networks might be 'slightly conscious.' He offered no evidence, and the claim drew sharp pushback from other researchers.

Worth knowing · 2 of 5

What's the strongest argument against current systems being conscious?

Some researchers argue large language models lack features that leading theories of consciousness treat as necessary, such as recurrent processing, a unified global workspace, and persistent embodied agency, making it unlikely current systems meet the bar, whatever the bar turns out to be.

Worth knowing · 3 of 5

Why do some researchers still take the question seriously?

One philosopher's widely discussed analysis concluded current models probably aren't conscious, but argued successors could plausibly clear the obstacles within a decade, so the question shouldn't be dismissed outright.

Worth knowing · 4 of 5

Are any companies studying this directly?

At least one AI lab has hired a researcher specifically to study whether its own models might warrant moral consideration, an effort that does not claim current systems are conscious, only that the possibility can't currently be ruled out either way.

Worth knowing · 5 of 5

What's the main objection to taking any of this seriously?

Linguists who study these systems closely argue that language models are pattern-matching text generators, not minds, and that talking about their 'consciousness' confuses fluent-sounding output with genuine understanding or experience, calling the whole framing premature and potentially misleading.

Words to know

SentienceThe capacity to have subjective experiences, such as feelings, as opposed to merely processing information.

Global workspace theoryA leading theory holding that consciousness arises when information is broadcast widely across a unified cognitive system.

AI welfareAn emerging research area asking whether AI systems could have morally relevant experiences, without assuming the answer either way.

More words

Stochastic parrotA critical term for language models that generate fluent text by statistical pattern-matching, without any claim to understanding.

A voice · 1 of 4

Ilya Sutskever · AI researcher who raised the claim publicly

In 2022 he posted that it may be that today's large neural networks are slightly conscious, a claim he offered without supporting evidence, which set off wide debate among researchers about whether such statements were responsible to make.

A voice · 2 of 4

David Chalmers · Philosopher who assessed the evidence on both sides

In a widely discussed 2022 analysis, he concluded it is somewhat unlikely current large language models are conscious, citing their lack of recurrent processing and a unified workspace, while arguing successors could plausibly meet the bar within a decade.

A voice · 3 of 4

Emily M. Bender · Linguist skeptical of the framing

Co-author of an influential 2021 paper describing large language models as 'stochastic parrots,' she argues these systems generate fluent text by statistical pattern-matching without reference to meaning, and warns against conflating that fluency with understanding or experience.

A voice · 4 of 4

Jeff Sebo · Philosopher who co-led an AI welfare report

As a lead author of a 2024 report, he argued there is a realistic possibility that some AI systems will be conscious in the near future, and that companies should begin assessing the question now rather than waiting for certainty.