Conscious AI: separating the myth from the science
9 steps · about 6 h 22 min · Beginner
- Hard Problem of Consciousness (article, Internet Encyclopedia of Philosophy, 25 min)
Start by learning what 'consciousness' means and why explaining it is hard, since every later argument depends on that.
- Why AI Isn't Going to Become Conscious (video, Anil Seth, TED, 17 min)
A short, accessible introduction to the gap between intelligence and consciousness, and to our habit of projecting minds onto things.
- The ELIZA Effect (article, 99% Invisible, 30 min)
Shows through history that people have attributed minds to chatbots since the 1960s, even when the program was trivially simple.
- The Chinese Room Argument (reference, Stanford Encyclopedia of Philosophy, 45 min)
Introduces the classic argument that manipulating symbols isn't understanding, along with the main replies to it.
- On the dangers of stochastic parrots: Can language models be too big? (article, Emily M. Bender, University of Washington, 20 min)
Gives the linguist's case that models trained only on text form lack access to meaning, which is the 'stochastic parrot' argument.
- The Mythology Of Conscious AI (article, Noema Magazine, 30 min)
Ties these ideas together by asking which psychological biases and assumptions, such as computational functionalism, make conscious AI feel plausible.
- Prospects and Pitfalls for Real Artificial Consciousness (video, Anil Seth, NYU Center for Mind, Ethics, and Policy, 1 h 15 min)
A longer, more rigorous lecture on whether computation is enough for consciousness and what follows if AI seems conscious.
- Consciousness in Artificial Intelligence: Insights from the Science of Consciousness (article, Butlin, Long et al., arXiv, 2 h)
Shows how scientists turn neuroscience theories into checkable indicators for assessing AI systems.
- The Evidence for AI Consciousness, Today (article, AI Frontiers, 20 min)
Ends with a recent argument that AI consciousness deserves serious probability, so you can test it against the skeptical sources.