Breakthrough Innovation and the Asymptotic Rationality of Artificial Intelligence

Breakthrough innovation is commonly perceived as a process of search unfolding across known and unknown task environments. This process promises to be upended by the increasing prowess of Artificial Intelligence (AI), capable of enhancing the breadth, depth, and speed with which such environments may be searched, with some scholars suggesting it may surmount the limits of human cognition, as captured by the assumption of bounded rationality. We argue that such asymptotic rationality considerations of AI need not be limited to known task environments in which they are usually invoked. We highlight how human agents often draw on codifiable techniques – which we summarize as meta-cognition, micro-experimentation, and enforcing preferences – to make unknown task environments amenable to finding breakthrough innovation. Yet, codifiable techniques may also be taught to machines, so that those should not only surpass human agents in identifying potential breakthroughs in known task environments, but play an increasingly important role beyond those. Based on these arguments, we discuss implications for the role of human agents in breakthrough innovation, for theories of search, and for policy and practice.

This event will take place in room T09-67, Mandeville Building. Alternatively, click HERE to join this seminar online.

Meeting ID: 981 6215 9528