Stephen Fleming: Consciousness & AI

In this episode, we speak with cognitive neuroscientist Stephen Fleming about theories of consciousness and how they relate to artificial intelligence. We discuss key concepts like global workspace theory, higher order theories, computational functionalism, and how neuroscience research on consciousness in humans can inform our understanding of whether machines may ever achieve consciousness. In particular, we talk with Steve about a recent research paper, Consciousness in Artificial Intelligence, which he co-authored with Patrick Butlin, Robert Long, Yoshua Bengio, and several others.

Steve provides an overview of different perspectives from philosophy and psychology on what mechanisms may give rise to consciousness. He explains global and local theories, the idea of a higher order system monitoring lower level representations, and similarities and differences between human and machine intelligence. The conversation explores current limitations in neuroscience for studying consciousness empirically and opportunities for interdisciplinary collaboration between neuroscientists and AI researchers.

Key Takeaways:

• Consciousness and intelligence are separate concepts—you can have one without the other
• Global workspace theory proposes consciousness arises when information is broadcast to widespread brain areas
• Higher order theories suggest a higher system monitoring lower representations enables consciousness
• Computational functionalism looks at information processing rather than biological substrate
• Attributing intelligence versus attributing experience/consciousness invoke different dimensions of social perception
• More research needed in neuroscience and social psychology around people’s intuitions about machine consciousness

Stephen Fleming is Professor of Cognitive Neuroscience at the Department of Experimental Psychology, University College London. Steve’s work aims to understand the mechanisms supporting human subjective experience and metacognition by employing a combination of psychophysics, brain imaging and computational modeling. He is the author of Know Thyself, a book on the science of metacognition, about which we interviewed him on Artificiality in December of 2021.

Episode Notes:

2:13 - Origins of the paper Stephen co-authored on consciousness in artificial intelligence

5:17 - Discussion of demarcating intelligence vs phenomenal consciousness in AI

6:34 - Explanation of computational functionalism and mapping functions between humans and machines

13:42 - Examples of theories like global workspace theory and higher order theories

19:27 - Clarifying when sensory information reaches consciousness under global theories

23:02 - Challenges in precisely defining aspects like the global workspace computationally

28:35 - Connections between higher order theories and generative adversarial networks

30:43 - Ongoing empirical evidence still needed to test higher order theories

36:52 - Iterative process needed to update theories based on advancing neuroscience

40:40 - Open questions remaining despite foundational research on consciousness

46:14 - Mismatch between public perceptions and indicators from neuroscience theories

50:30 - Experiments probing anthropomorphism and consciousness attribution

56:17 - Surprising survey results on public views of AI experience

59:36 - Ethical issues raised if public acceptance diverges from scientific consensus