AI: Living The Dream

If only we could make a program that could learn. If only we could package the world’s knowledge in a format which a computer could understand.

An MIT lecture from 1981 addresses these exact subjects while doing a better job of explaining AI than most presentations. This isn’t an information overload about neural networks etc. but rather a deep dive into the conceptual goals of machine learning and model training.

Furthermore the topics of iterative software development, gathering & refining requirements, and communication between domain experts and programmers are discussed in detail. The impacts of these on both traditional and AI software development cycles are covered while mostly avoiding implementation details, making this decades old video lecture still highly valuable today.

Are explainable AI’s easier to train? How can we debug errors in the output of a model? How can we debug errors in the gathering of requirements? Have we implemented AI today in a way that can be explained or even debugged? Looking at today’s leading models, how close are we to achieving the “dream” of artificial intelligence? What steps lie ahead to improve the status quo? Answering these with honesty, what are the most appropriate and useful ways to apply today’s technology?

Watch: Artificial Intelligence - Learning and the TIERESIAS Program :: 59m