There will be several overarching and coupled themes in our study of AI.
One focus will be quite technical and pragmatic — a tool-based perspective. We will study the declarative encoding of knowledge in a computer, and the procedural reasoning that the computer performs with that knowledge. Students will be learning techniques, representations, methods, strategies, and tools for expanding the behaviors that computers typically exhibit. These behaviors may seem intelligent, and perhaps are intelligent, but they are typically stove pipes in support of very specific tasks.
A theoretical perspective is that AI is the study of non-deterministic processes. There is an old adage that computers can only do what they are programmed to do, which is true enough, but computers can be programmed to explore, evaluate, and choose among alternatives, and in these explorations, it can learn what the best choices are. Our study of machine learning will be limited, because AI is not synonymous with machine learning or with deep learning. But we will get to machine learning nonetheless.
The perspectives above fit a broader perspective of AIs as assistive technology, designed to augment human intelligence, but where the AI itself won’t approach the capabilities of human intelligence. Arguably, the pairing of assistive AIs with humans, will surpass the intelligence of humans alone.
Computer science generally may have caused you to become more reflective of your own thought processes, but I think in studying AI, you’ll think about your own thinking even more. Most obviously, you’ll reflect on how you do something so that you can design software to do the same thing. After this first step, you’ll design different, “better” ways to do the same thing, and this will feed back into how you think in the future. Thus, another perspective of the course is as critical thinking lab, and I want us to reflect on how humans reason — from evidence and instinct, for short-term emergencies and long-term planning — and how this reasoning can be represented for, and improved by, an AI.
There is plenty to be excited about in the assistive AI perspectives above, but some-to-many students first walk into an AI course, as I did many years ago, thinking that they will grapple with holistic and artificial general intelligences (AGIs) — the kind that they see in science fiction and what they hear alluded to by Elon Musk, Bill Gates, and others. Commentators believe that when AGIs become smart enough, they will be capable of “recursive self-improvement” that will result in super-intelligences through an “intelligence explosion.” Some are concerned that these SIs will usurp humankind, and others believe that they will be a windfall for humankind — maybe both.
Frankly, I think that we are at a precipice of falsly-perceived general artificial intelligences, which come with their own dangers, but that we are rather far from authentic artificial intelligences of a general nature. Some students probably feel let down by the assistive perspectives of the first paragraphs, but they are important and interesting. Nonetheless, I have built time into the Schedule so that we can talk about general, holistic AIs as well.
The assistive perspectives and general-AI perspective come together in the perspective of collective intelligence. This term has traditionally connoted groups of humans, where the group is (ideally) smarter than any one individual in the group. In the future, however, the smartest collectives are likely to be of AIs and humans. Presumably, these hybrid collectives will work best if the AIs among them aren’t human wannabe’s in mind or body. Rather, collective smarts will be increased by cognitive, conative, affective, and physical diversity. So, while human-like AIs are interesting and important, I get most excited about AIs that are very different from humans — alien intelligences — but with an appreciation for and understanding of humans. And of course, students in this course will get a glimpse of these alien intelligences of the future too, because the appreciation and understanding should be bidirectional.
Doug Fisher