HPS340H1

The Limits of Machine Intelligence

With the recent headline-making breakthroughs in deep learning neural networks (DNNs), it might seem that we are on the cusp of living with artificial systems that match or exceed human intelligence. But there remain longstanding philosophical challenges around the definition of intelligence that AI researchers use, how they measure the performance of their systems, and what DNNS could really be capable of, that still need addressing. For example, how close are DNNs to passing the Turing test? How close are we to building general intelligence and what do we need to get us there? How can we draw fair and meaningful comparisons between artificial and biological systems? We will draw on material from the history and philosophy of science to evaluate and inform current debates around the limits of AI. For example, we’ll consider what kinds of explanations DNNs can provide. We’ll also look at how debates between the rationalists and empiricists (e.g. Locke, Hume, and Kant) inform current debates between AI nativists and empiricists. This course explores these questions through contemporary texts across the fields of philosophy of science, artificial intelligence, comparative psychology, and cognitive science, among others.