I’m currently reading a pair of books about artificial intelligence. The Emperor’s New Mind was written by Roger Penrose in 1989, and is an exploration of how computers work, and why they will never be able to mimic the intelligence of the human mind. How to Create a Mind is a book by Ray Kurzweil written in 2012 about how the human mind works, and how computers will be able to mimic those structures and reach a level of true artificial intelligence. I’m deliberately reading them semi-simultaneously to maximize the juxtaposition of time and technological advancement, which is stark. But I don’t want this blog post to be a comparison of the two books or approaches. Rather, I want to talk about the question that has been plaguing me since I started reading them.
What is intelligence?
More specifically, how do we identify intelligence in others? If we’re thinking about what is artificial intelligence, what are our standards? How do we know when we’ve made it there?
I am intelligent. No one, and nothing can convince me of otherwise. My logic for this pronouncement is trite and cliche. I think, therefore I am. I am intelligent because I can recognize my own intelligence.
So I know that I am intelligent. But how do I know that you are intelligent?
Well, since you are reading these words, that provides a measure of intelligence. But a computer can read this words as well, technically. If you leave a comment, you demonstrate your ability to interact, to communicate based on these words. To create something new, based on what I’m talking about. You’ve probably noticed in your blog though, lots of spam comments that are computers doing exactly that. But those computers won’t do it as well as you can. I can recognize a spam comment versus one from one of you. So that commenting ability shows your intelligence. If we were in person, we could have a conversation about what I wrote about. That conversation would demonstrate to me that you are intelligent.
In fact, conversation has long been one of the hallmarks of artificial intelligence, in something called a Turning Test. The test is this. In a blind experiment, can a human judge have a conversation with a computer and a real person, and identify which one is the computer reliably? Essentially, can a computer mimic human conversation, and pass as a human? And indeed, a computer named Eugene passed a limited Turing Test a few years ago(well, kinda sorta, ish, not really, but hey good effort). Does that make Eugene intelligent? Is mimicry enough of a requirement for intelligence?
I think not (as would most of the world). Part of the problem that we face with artificial intelligence is that the most convenient way to test intelligence is by performing tasks. How smart are rats? Put them in a maze and find out. How smart are monkeys? See if we can teach them sign language. But intelligence is not defined by task completion. I am not intelligent because I can hold a conversation. I can hold a conversation because I am intelligent. I am not intelligent because I can write these words, and drive a car, and play chess, and write music. I can do those things because I am intelligent. A computer can do all of those tasks as well. But can a computer do them because it is following a preset series of commands, or an algorithm created by a human to complete those tasks. So in a world where more and more of the tasks that we might use to show our intelligence can be completed as well or better by a computer or machine, how do we identify intelligence? How do we know when we truly have an artificial intelligence, and not just some combination of algorithms (which is of course assuming that we are more than simply a combination of algorithms, which is hardly a sure thing).
Perhaps the task that best shows our intelligence is the task of questioning. My intelligence is not demonstrated by my ability to converse adequately with you. My intelligence is demonstrated by my ability to question the world around me. And out of those questions, manipulate the world around me. We are intelligent because we can ask the question “What is intelligence” and then design experiments to try to figure it out. In other words, we are able to not simply follow a series of commands to solve a problem. We are able to identify the problem, and then create the series of commands to use to solve the problem. Perhaps we’ll know when we have created artificial intelligence because the computer will start trying to determine whether are not we are intelligent. Maybe God knew that He got mankind right when we started to question his existence?
Question for another day: Why are we thinking about intelligence in terms of human intelligence, based on human tasks? Why wouldn’t artificial intelligence look completely different than human, with its own set of values and goals?
Only tangentially related, but just because I wanted to add a video: https://www.youtube.com/watch?v=9TRv0cXUVQw