BAMΞ’s AI Consciousness Sprint is devoted to research on artificial consciousness, also called synthetic consciousness. This is research on, among other things (cf. below), the question of whether artificial systems in general, and Artificial Intelligence (AI) systems in particular, can have conscious experiences. Because of consciousness’ ethical relevance, and because of the increasing ad hoc attributions of conscious experiences to LLMs, this question is of particular societal importance.

There is a large and growing body of research devoted to artificial consciousness. BAMΞ’s AI Consciousness sprint is meant to explore, assess, and expand, this research from a mathematical perspective. Because AI systems are mathematical systems—they are defined by formal or mathematical structures, after all, both on the level of programming and the level of machine code—it might be the case that rigorous answers to the AI consciousness question require a mathematical perspective.

Research Questions

1. Are AI systems conscious, or can they be? And if so, which AI systems can, and which cannot? This is the question of whether AI systems can have conscious experiences at all, independently of what these experiences are.
2. Are there tests for whether AI systems are conscious? This is the question of whether there are empirical procedures that can be applied to AI systems so as to infer whether they are capable of having conscious experiences. As of late, such procedures have come to be known as C-tests.
3. Which theoretical means exist to assess whether AI systems are conscious? Here, ‘theoretical’ includes both scientific and philosophical methods. A particularly important example are theories of consciousness. Are the theoretical constructs and empirical evidence across theories of consciousness rigorous and detailed enough to warrant an application to the AI consciousness question, in particular in societal contexts?
4. Which conscious experiences can AI systems have, if any? This is the question of the phenomenal character of the conscious experiences of AI systems: what is it like to be an AI system? Are the conscious experiences of AI systems anything like human conscious experiences? If so, what are the similarities and differences? If not, what can be said about AI's experiences?
5. Which types of computation are necessary or sufficient to support AI consciousness? This includes the question about forms of computation that can support consciousness—can a Turing computation, say?—as well as the question of which specific properties a computation would have to exhibit to support artificial consciousness, if any does.
6. Does consciousness matter for existential risk? This is the question of whether conscious AI systems in general, and self-conscious AI systems in particular, pose a particular problem for alignment or existential threats.

We are currently in the process of organizing workshops, invited talks, and discussion sessions for this sprint. Please reach out if you would like to contribute to this sprint.

Updates about activities related to this sprint will be distributed via our mailing list.