Can AI Be Conscious? Why Science Needs New Words for Machine Minds

Table of Links

Abstract and Introduction

1. Extents and ways in which AI has been inspired by understanding of the brain

   1.1 Computational models

   1.2 Artificial Neural Networks

2. Embodiment of conscious processing: hierarchy and parallelism of nested levels of organization

3. Evolution: from brain architecture to culture

   3.1 Genetic basis and epigenetic development of the brain

   3.2 AI and evolution: consequences for artificial consciousness

4. Spontaneous activity and creativity

5. Conscious vs non-conscious processing in the brain, or res cogitans vs res extensa

6. AI consciousness and social interaction challenge rational thinking and language

Conclusion, Acknowledgments, and References

6. AI consciousness and social interaction challenge rational thinking and language

Starting from evolutionary and neuroscientific data, we have listed above a number of features that AI research might take into account in its attempts to advance the development of artificial consciousness. Some of the limitations of current AI that we have outlined had already been identified by other authors, such as the role of embodiment, emotions, and motivation in machine consciousness, and the issue of robot experience (Chella & Manzotti, 2009; Montemayor, 2023), while others appear less explored. Among these, the capacity for interaction between individuals in a social group which gives rise to a number of humanspecific dispositions, including theory of mind, cultural experience and ethical responsibility. In principle, the fact that AI constructs have a different physical structure and a different functional organization than biological systems that are probably conscious (even if at different levels, e.g., humans vs. other species including Drosophila (Grover et al., 2022), does not imply that AI systems are not or cannot become conscious. The above analysis has revealed some factors that data indicate are correlated with conscious processing in the human brain, but this is not sufficient to conclude that they are necessary conditions for conscious processing in both humans and in other biological or in artificial systems. Notably, the issue of the actual existence of artificial forms of consciousness is importantly empirical: we need reliable indicators to identify non-biological conscious processing in AI systems (Bayne et al., 2024; L. Dung, 2023; Pennartz, Farisco, & Evers, 2019). Yet the challenge is not only empirical, but also conceptual: how can we conceptualize a non-biological kind of conscious processing? Probably the linguistic and conceptual categories that we normally use to refer to consciousness are not good (e.g., they are too anthropomorphic and anthropocentric), and should be replaced. As (Pennartz et al., 2019) outline, we may use the same term and expand our context of application accordingly, or we may coin a new term to denote machines. This choice is epistemically very important. In fact, “linguistic innovations must be well-justified in order to avoid conceptual inflation. Yet, if we are too bound by common usage traditions, we hamper development of new thought. Our languages need to evolve to enable the expression of new ideas, knowledge and normative systems.” (Pennartz et al., 2019)

We do not presume to re-invent the terminology, but rather wish to recommend an improved terminological and conceptual clarity when referring consciousness to AI. This necessity may be interpreted as part of the more general call for conceptual adaptation as a consequence of the development of new technologies, as expressed in the conceptual engineering (Hopster & Löhr, 2023). An additional, specific justification of our recommendation derives from the multidimensional nature of consciousness, which makes crucial for any attempt to develop artificial consciousness to specify which is the dimension and/or the form of consciousness that it targets. Whilst the extension of other mental terms like attention and intelligence to non-human animals and AI systems may appear less problematic than consciousness, we should distinguish two semantic dimensions when referring any mental term to AI: the epistemological dimension (i.e., what is the nature of the feature indicated by the term?), and the functional dimension (i.e., how does the feature indicated by the term work? What is the relevant architecture, that is the adequate combination of structure and functionalities?). While the extension of the first dimension to AI is problematic, the extension of the second dimension may be more straightforward provided that relevant structures and functions may be replicated at the adequate scale or level of details, as argued above. In any case, a neuroscience-based linguistic and conceptual development appears plausible.

In conclusion, caution in attributing consciousness to AI, and particularly improving the conceptual clarity when making this attribution, are more than a matter of communication. It is also a matter of how thoughts develop, for language shapes thought both epistemologically, in terms of what we can think and know, and normatively, in terms of the values we develop. Therefore, these questions present us with conceptual choices that need to be made both philosophically (in terms of clarity, simplicity and logical coherence) and empirically (in terms of scientific justification, experimental validation and explanatory power).