Rethinking Computer Science Through AI

Most of artificial intelligence (AI) in use today falls under the categories of the first two waves of AI research. First wave AI systems follow clear rules, written by programmer, aiming to cover every eventuality. Second wave AI systems are the kind that use statistical learning to arrive at an answer for a certain type of problem. Think of image classification system. The third wave of AI envisions a future in which AI systems are more than just tools that execute human programmed rules or generalize from human-curated data sets. The systems will function as partners rather than as tools. They can acquire human-like communication and reasoning capabilities, with the ability to recognize new situations and to adapt to them. For example, a third wave AI system might note that a speed limit of 120 km/h does not make sense when entering a small village by car.

In my opinion, it is time to usher in the third way of AI. Current second wave AI systems are highly specialized systems that are typically very good at specific, well-defined tasks. They are often not robust and without extensive re-training often fail even for modestly different circumstances. For instance, an object in a non-canonical orientation and context fools many second wave AIs for visual scene understanding. As Gary Marcus points out, they may fail to recognize a school bus tipped over on its side in the context of a snowy road. Understanding such limitations of second wave approaches is, to name only one of many instances, particularly important for self-driving cars. As impressive as they are already at the current stage, they utilize a composite of independent and narrow intelligent subsystems. Following Scott Jones, if you took the software from a self-driving care and put it in a golf cart, it is likely to be useless without considerable re-programming and -training. In contrast, any human who has learned to drive a car could get into a golf cart for the first time and would have no major problem navigating the fairways. This is because humans are very good at abstraction: we can easily generalize solutions and apply them to similar but different problems. And even if we were encountering problems driving the golf cart, we could articulate them and ask for help.

The third wave of AI poses many deep and fascinating scientific problems: How do we bring together different—and currently separated—AI regimes: low-level perception and high-level reasoning? Akin to Systems Biology, how should a systemic view on AI look like allows us to capture, understands and utilize individual AI algorithms as building blocks for a complex AI system in a mathematical and computationally sound way? How do we manage that non AI experts build, use and interact with AI systems? How do we make machine learning a co-adaptive process, in which a human is changing computer behavior, but the human also adapts to use machine learning more effectively and adapts his or her data and goals in response to what is learned by the machine? How do we increase the efficacy of the “teachers” given the learners, i.e., how to we move from machine learning to machine teaching? How to balance tasks that we would find useful to automate and tasks in which it might remain meaningful for us humans to participate. How do we incorporate computational models for human intelligence into AI systems so that machines can learn so much about the world, so rapidly and flexibly, as humans?

Meeting these challenges not only pushes AI. It provides a unique opportunity to rethink fundamental problems and methods of computer science itself, from hardware and software design, over databases and robotics, to human–computer interaction and software engineering. The third wave of AI is a team sport that holds the chance to grow even closer together in the CS departments and reaching out to other fields such as (computational) cognitive science.

The current special issue on Ontologies and Data Management, its second part, illustrates this very nicely. The contributions underscore the great value of combining reasoning and learning. Enjoy learning about the latest developments!

Stay healthy,

Kristian Kersting.