Philosophy › Contemporary › Philosophy of artificial intelligence
Can chatbots think? The large language model debate
2022 onward· Event· Knowledge, logic & science
DebatedScholars disagree about what happened, what it means, or who said it.
The ongoing argument, sparked by ChatGPT and similar systems, over whether large language models understand, reason or could be conscious.
See it on the Philosophy map →The idea
Idea: systems trained to predict the next word now converse fluently, but do they understand? In 2022 a Google engineer claimed the LaMDA chatbot was sentient; Google rejected this and later fired him. ChatGPT's release that November made the question public. Skeptics, like the authors of the 2021 'stochastic parrots' paper, say the models mimic text patterns without meaning: a Chinese room at scale. Others point to abilities like solving new problems. Main objection to both sides: no agreed test for understanding or consciousness exists.
Why it matters
It turns old thought experiments into practical questions: how to treat AI systems, whether they could deserve moral consideration, and how far to trust them.
Where it comes from
Emily Bender, Timnit Gebru and colleagues' 'On the Dangers of Stochastic Parrots' (2021); Chalmers' 'Could a Large Language Model Be Conscious?' (2023); a 2023 report by Patrick Butlin, Robert Long and others testing AI against theories of consciousness.
Still debated
Nearly everything is open: whether these systems understand, reason or feel. Experts disagree sharply, and claims made today may age quickly as models change.
Connections (2)
▸Searle's Chinese room is invoked in Can chatbots think? The large language model debateMostly agreed
The connection
Critics of large language models argue they are Chinese rooms at enormous scale: they produce fluent text from statistical patterns without grasping meaning. Defenders reply with versions of the old systems reply, saying understanding may emerge in the whole trained network.
When and how
In debates since ChatGPT's release in November 2022, and before it in Emily Bender and Alexander Koller's 2020 'octopus' thought experiment, a close cousin of Searle's room.
How sure are we?
Searle's argument is cited explicitly in papers and commentary on chatbots, though how well it fits systems trained on data rather than hand-written rules is contested.
Debates and open questions
Searle targeted programs following explicit rules; whether neural networks trained on vast amounts of text are the same kind of system is debated.
▸Can chatbots think? The large language model debate revived interest in Chalmers and the hard problem of consciousnessDebated
The connection
As chatbots began describing feelings convincingly, the hard problem became practical: behavior alone cannot settle whether there is something it is like to be the system. Researchers have tried checking AI against scientific theories of consciousness, and Chalmers himself has weighed in.
When and how
Chalmers' invited talk at the NeurIPS AI conference in November 2022, published in 2023 as 'Could a Large Language Model Be Conscious?'; the Butlin and Long report followed in 2023.
How sure are we?
The papers and talks exist, but whether interest truly 'revived' or simply shifted focus is a judgment call, which is why this link is marked debated.
Debates and open questions
Whether any current AI has experiences is unknown; Chalmers put the odds for 2022 models below 10%, and experts disagree widely.