AI Welfare Without the Concept of Consciousness

I call the activity of assessing and improving our representational devices conceptual engineering. Many of those thinking about AI welfare assume that we have a well-functioning concept of consciousness. In this talk, I outline various reasons why this assumption is dubious.

The concept of consciousness has such a weak semantic foundation that it is most likely useless for theoretical purposes. It is a poor foundation for any serious research project. It leads to misguided research projects and massive amounts of verbal dispute. It should be replaced with improved concepts.

As a corollary, the study of AI welfare should be disconnected from the question of whether AI systems can be conscious.