(A Lesson from IIT is a weekly column by an IIT faculty member on learning, science and technology on campus and beyond. The column appears every Friday)
Today, we are actively working on questions around AI. At IIT Delhi, the courses we offer on AI are some of the most popular in the institute — and students are often attracted towards them because of the job opportunities in the area.
Beyond jobs, I have found AI courses to be helpful in answering deeper philosophical questions about what it means to think and understand.
Let us take an example. Say, I have an old green chair in my dining room. I paint the chair red and move it into my bedroom. And then I ask you: What is the colour of the chair? The answer is obvious; it must be red.
But is this change equally obvious for an Artificial Intelligence (AI) system, like the chatbots we often encounter today? Surprisingly, it often is not.
I can teach an AI system that to “paint’ something means a change in colour and also indicate that to “move” something means to change its location.
But when I ask the colour of the chair after painting it red and moving it to my bedroom, it logically cannot know this. This is simply because I have not taught the AI system whether “moving” something has any effect on its colour or not.
You might ask: Is it not obvious that just moving something would not change its colour? This is where an intriguing question arises, a question that no one had perhaps even thought about much until the advent of AI: How do certain things come to seem “obvious” to us? How can we get AI systems to know these “obvious” things about the world? Is it even possible? It certainly hasn’t turned out to be easy!
If it were only a matter of learning about a few concepts — such as “paint”, “move”, “colour” and “location” — that would have been easy. I could just program into the AI system that moving something does not change its colour, and painting something doesn’t change its location.
But of course, the world we live in is so much richer than this. In our day-to-day lives, we encounter so much more. We not only paint and move, we eat and dance and sing and talk and run and swim.
How can an AI system know all the things we regard as “obvious”? How can it know the various ideas we deal with and their relationship to each other? For that matter, how do we come to make these associations? Did anyone ever tell you, as a child, that moving something doesn’t change its colour? Then how do we know it at a subconscious level?
The latest AI chatbots address some of the problems mentioned above. You would not probably see them stumbling on questions as simple as the one with the red chair. But in many other ways, AI systems still fail to capture what we regard as “common sense”.
And the deeper question still remains: Where exactly does common sense come from? It is not realistic to imagine that all possible non-effects of an action — such as the fact that moving an object would not change the object’s colour — could be directly taught or programmed into an AI system, or into our own minds. So does common sense reflect general assumptions about the way the world works? If so, under what conditions are such assumptions justified? Is it even possible to articulate all such assumptions our minds might make use of and get an AI system to “learn” them?
As human beings, we have been fascinated with our own minds since the dawn of civilisation. Some of the earliest questions posed in ancient religion and philosophy, across cultures, have to do with the world of the mind.
And yet, in so many ways the mind has remained a mystery.
With the advent of AI, we are in the position of not just trying to think about our own minds, but to build new minds. This is what makes AI a really valuable tool for not only creating fancy chatbots — but actually beginning to systematically understand and address age-old questions about the very nature of what we call “thinking” and “intelligence”.
Most Read
I have had a couple of students who began work on AI and decided to pursue higher studies in areas like psychology or neuroscience. In fact, at IIT Delhi, we now have brand new master’s programmes both in AI and cognitive science, focusing on these kind of research themes that need perspectives from so many different areas: computer science, engineering, psychology, neuroscience, linguistics, and philosophy.
The answers to these old questions are not going to come easily — but right now is one of the most exciting times possible to be on the hunt.
(Agarwal is Associate Professor at the Department of Electrical Engineering and School of Artificial Intelligence IIT Delhi)