This section contains archived content made available for reference, research, or recordkeeping purposes. These materials are not actively maintained and may not meet current accessibility standards, such as WCAG 2.1 Level AA. If you need assistance accessing this content or require an accessible format, please contact sohwebmaster@uci.edu. We are committed to ensuring access for all users and will provide accommodations promptly upon request.
AI as a pathway to knowledge?
Artificial Intelligence is increasingly positioned as a transformative tool in the pursuit of knowledge, reshaping how we collect, analyze, and interpret data. But under what conditions can an AI system produce knowledge? Reliability is required for knowledge (or so we shall assume here). A true belief qualifies as knowledge only if it is formed via a reliable belief-forming method or process (i.e. one that generates mostly true beliefs). Thus, true AI-based beliefs qualify as knowledge only if they are formed via reliable AI methods or belief-forming processes (i.e. methods or processes that take AI outcomes as input and deliver beliefs as outputs). We argue that, while reliability may be a necessary condition for AI-based knowledge, it—together with truth—is not sufficient. Reliability must be supplemented by a condition which ensures that the relevant AI belief-forming method or process does not take highly inaccurate inputs. We home in on this condition through considerations on the distinctive nature of the types of errors that AI systems may make.