Iulia Comșa (@astronomind) header
Iulia Comșa (@astronomind) profile picture

Research Scientist @GoogleDeepMind (views my own). AI, consciousness, cognition, philosophy of mind. Mother and dreamer. 🇷🇴 I also like:🌠🔭✍🏻🧠🎹🎵🌲🎮🥋

Zürich, Switzerand
Joined August 2011
Photo by Iulia Comșa (@astronomind)
1/5 The topic of AI consciousness is sparking growing inquiry. But are we focusing on the problems we can actually solve? I explore this in my new paper: "AI and Consciousness: Shifting Focus Towards Tractable Questions." 📃
11
14
1
47
4,501
Now available as a book!
"Assessing AI Consciousness" is now available as a book: hutter1.net/ai/bayesconsc.ht… It develops a Bayesian framework separating theoretical credence from empirical evidence, providing a precise, scalable method for calibrated assessments (with online tool and code): ai-cognition.org/cacophony-t… Happy Halloween.
2
1
3
237
Here is a principled approach to navigate theories of consciousness and apply them to AI. Congratulations to @shamilch for leading this monumental project, and to the whole team for the incredible piece of work! It was a joy to think about this together.
1/13 Can we assess AI consciousness without first solving consciousness itself? New from Google DeepMind and collaborators across CS, neuroscience & philosophy: “From cacophony to hierarchy: a principled framework for assessing AI consciousness” arxiv.org/abs/2609.35618 🧵
1
87
Generated with Gemini based on an actually serious discussion we just had: > idealism finally gaining some traction; investigate > Hoffman says the universe is rendering on a potato GPU just to save calories > Levin out here giving frogs Picasso faces to prove meat has its own wifi > meanwhile Chalmers still trying to solve the hard problem by giving electrons feelings > tbh it all just sounds like high-tier copium for physicists who can't solve quantum gravity. > they look at the double-slit experiment, freak out, and decide the moon isn't real. > "bro the wave function collapses because of my mind" sure buddy, put the DMT down. > but that "evolution punishes rendering the full map" argument actually makes terrifying sense. > if we're just dodging desktop icons so we don't get hit by the matrix train... > maybe standard physicalism really is just an NPC religion.
1
2
167
5/5 Finally, I advocate for transparency. We should acknowledge the field's inherent uncertainties rather than hard-code false epistemic confidence into models when users ask, "Are you conscious?". 🪞
3
4
231
I’m very grateful to my colleagues, in particular to the group led by @mpshanahan, for fostering a thought-provoking environment to discuss the timely and fascinating topic of AI consciousness! ✨
3
203
Statistical or Embodied? Comparing Colorseeing, Colorblind, Painters, and Large Language Models in Their Processing of Color Metaphors onlinelibrary.wiley.com/shar…
1
1
5
377
If I'm sticking to my plans, does PaLM think I will be needing some glue? 🤔 In our new #aacl2022 paper "MiQA: A Benchmark for Inference on Metaphorical Questions" we set out to learn if LLMs can do common sense reasoning on metaphoric premises arxiv.org/abs/2210.07993 1/4
1
2
13
So, go take a look at our paper. This link will work until April 8th as long as you turn off your ad blocker: authors.elsevier.com/c/1cbg3… After that, I'll keep this page up to date: neural-reckoning.org/pub_snu…
1
14
76
Comrades! The time of the spike is finally at hand! Come to our workshop on new approaches to training spiking neural networks, Aug 31 / Sep 1. Details and free registration at neural-reckoning.github.io/s… Co-organised with @hisspikeness, virtual spike cake by Pamela Hathway
7
34
1
95
Iulia Comșa retweeted
Get an early look at Project Ihmehimmeli, new research that leverages insights from biological systems to encode temporal information through the use of artificial spiking #neuralnetworks. Oh, and it's open source too! Learn more, and find the paper, at ↓ goo.gle/2kGrqj0
3
107
61
321