Consciousness in AI: Insights from the Science of Consciousness (2023) (arxiv.org)

A recent report has sparked discussion in the AI community regarding the potential for machines to achieve consciousness. The authors advocate for a systematic, evidence-based examination of AI systems through the lens of existing neuroscientific theories, such as recurrent processing theory and global workspace theory. They propose "indicator properties" derived from these theories, which can be applied to assess the consciousness potential of AI systems. Their findings reveal that while current AI models do not demonstrate consciousness, there are no apparent technical hurdles preventing future systems from embodying these indicator properties. This analysis is significant as it provides a framework for determining AI consciousness, a topic of both ethical and technological importance. By clearly articulating the computational terms associated with consciousness indicators, the report lays the groundwork for ongoing research into advanced AI systems that may one day possess conscious-like qualities. As the debate intensifies, this study encourages a deeper investigation into the implications of conscious AI, raising questions about the responsibilities and ethical considerations that would accompany such advancements.