Clara Colombatto

Why do politicians endorse nationalistic policies like vaccine hoarding? In our new paper @PsychScience we test public & expert opinions on vaccine policies, w @mollycrockett JimEverett @SennJulien @MichelAMarechal and in collaboration w @apoliticalco 1/n doi.org/10.1177/095679…

🦀🐝🧠🤖 How should we make decisions when faced with systems of uncertain sentience? My new book The Edge of Sentience: Risk and Precaution in Humans, Other Animals, and AI is available online for free at academic.oup.com/book/57949. Please download it! Just click on PDF to get a PDF of the whole book. Individual chapters are also available as PDFs - just click on whichever chapters you want. The book is completely free to everyone and I'm grateful for all shares and forwards.

@cMadan @OfficialCSBBCS I will be there too! 🧠 camp reunion!

Canada friends, let me know if you'll be at @OfficialCSBBCS. I'll be there and am happy to chat :)

New postdoc opening at @imperialcollege to work with me on #AI topics, especially at the intersection of AI and philosophy, as part of @LeverhulmeCFI: jobs.ac.uk/job/DHQ544/res…

Using a computational model to quantify difficulty in reconstructing images from compressed codes, @qi_lin7 et al. show that reconstruction errors explain aspects of perception and memory performance. @YalePsychology @WuTsaiYale nature.com/articles/s4156…

The New York Declaration of Animal Consciousness was signed today by leading figures in philosophy and science, reflecting our current scientific knowledge of consciousness in vertebrates and invertebrates. Join us! nydeclaration.com

@action_brain @BMI_WesternU Thank you — very happy to be here 🇨🇦🇨🇦 and looking forward to visiting soon!

@davidchalmers42 The "human-like" labelling was inspired by the text of the question which takes humans as examples of experiencers; and yes, 23% are above 50 if you take that as cut-off!

@ClaraColombatto it looks to me that the key takeaway is: 10-20% of respondents say that LLMs are at least somewhat experiencers (i'm not sure of the exact figure, i'm just eyeballing answers 50 or above in figure 1A). that's interesting but 10-20% is less than 67%.

The spread of LLMs has accelerated debates on whether AI could be conscious. While experts have weighed in on this issue, in a new study (preprint: osf.io/preprints/psya…) we surveyed the general population on whether they attributed consciousness to LLMs. A thread: (1/n)

@AndrewCurran_ ah, good point -- we didn't specify the version, so our participants could have had either in mind; we should add this in follow-ups, thank you for suggesting it!

@ClaraColombatto Hi, thank you for your work. I couldn't find if you used 3.5 or 4 for testing? The paper just says ChatGPT, unless I missed it.

@Dzoldzaya_J @davidchalmers42 Yes, this phrasing is more precise; thank you for the suggestion!

@MicheleFarisco Yes that's right, we mean inclination, not a motivation-thank you for pointing this out! As for ethics, we are thinking this might matter for how people think about AI responsibility eg given evidence that experience/agency map onto moral types, eg tandfonline.com/doi/full/10.10…

@ClaraColombatto Very interesting indeed. What specific ethical implications do you anticipate? Also, you talk about willingness to attribute consciousness: why not simply “inclination”? “To want to”implies clear motivations in my mind, while it seems to me it is more a matter of feeling

@ahron_maline We actually tried to avoid this by modifying the original question to have as a midpoint not confidence ("unsure") but degree of experience; we had a separate question probing confidence, and people at the extremes of the scales were more confident in their answers

@ClaraColombatto The question can be interpreted as "how likely is it that the AI is an experiencer" or as "where is the AI on a continuum from 'no experience' to 'human-level experience". If most of those who said 10-20% meant those as probabilities, then this isn't such a bad error

@DanielCappell @MatthiasMichel_ I agree that in general they will correlate, but note that only attributions of experience-related mental states predicted consciousness attributions, not all mental states -- see #5 in thread

But these high rates of consciousness attributions suggest that the general population is willing to attribute consciousness to AI – perhaps surprising given experts in neuro & psychology/philosophy estimate that LLMs are unlikely to be conscious, eg arxiv.org/abs/2308.08708 (7/n)

Of course, these are preliminary results: this pattern of attributions might differ with different experimental measures or prompts (eg with indirect behavioural markers), and in different samples (eg not recruited online, beyond the US). (6/n)