Investigating Digital Minds

Investigating Digital MindsHow can we tell whether an AI system has consciousness, sentience, agency, or other morally significant capacities? And how should we treat AI systems that might have these capacities? In this talk, we will discuss a “marker method” that we can use to test AI systems for moral patienthood. We will also discuss how we can make decisions about AI development, deployment, and scaling under uncertainty about whether AI systems matter, how much they matter, or in what ways they matter. As we will see, there are substantial risks involved with false negatives (mistakenly treating subjects as objects) and false positives (mistakenly treating objects as subjects). Fortunately, we can mitigate these risks by developing rigorous evaluations and by seeking co-beneficial policies for humans, animals, and AI systems.