What’s Wrong With the Typical Explanation of AI
When looking for a simple explanation of artificial intelligence (AI), you will most likely find something like the following:
Artificial intelligence is the ability of machines to perform tasks that typically require human intelligence, such as learning, reasoning, problem-solving, and understanding natural language.
This explanation is simple in the words it uses, but it’s unsatisfying because it describes artificial intelligence in terms of human intelligence. Human intelligence is extremely complex and poorly understood; consequently, basing an explanation of AI on human intelligence does little to help you understand it.
Specifically, using human capabilities to explain artificial intelligence suffers from the following problems:
- Humans can do things that machines can also do but are not considered AI, such as simple arithmetic.
- AI can do things that the humans can’t, such as generating fake, photo-realistic images and video.
- AI systems don’t work like the human brains do, and they don’t need to in order to be considered AI.
- Human intelligence is poorly understood.
The Best Simple Explanation of AI
Given that we don’t want to describe AI in terms of human intelligence, here is a great, more general alternative:
Artificial intelligence is a system that figures out how to solve a problem on its own and then provides the solution.
Let’s see how this definition applies to the familiar problem of reading handwriting. The difficulty with recognizing handwriting is that it’s impossible to explicitly specify rules that can be followed step-by-step to distinguish one character from another, especially in terms of pixels, which is what computers see. The variation is just too great, there are lots of borderline cases, and it’s just too complicated.
For humans, knowing how to read doesn’t help much in coming up with that set of rules. That’s because there are a lot of subconscious things happening in the brain we don’t know of and can’t articulate.
So how do we solve this problem? We let computers look at a lot of handwriting examples and perform a lot of fast calculations to come up with the rules for us.
A Simple Explanation of What Machine Learning Is
The above is what all machine learning (ML) algorithms do, including neural networks:
Machine learning algorithms come up with a set of rules to solve a problem by looking at a lot of examples, i.e. data.
Processing a lot of data to feed or “train” machine learning algorithms takes considerable computing power. For many years, it was unclear whether machine learning methods could perform well with the toughest problems in AI, such as driving or holding conversations. But in recent years, computers have become powerful enough, and now we know they can. For its success, machine learning is becoming synonymous with AI. However, ML is not all of what AI is.
Non-Machine Learning AI
There are algorithms that don’t work by coming up with a set of rules on their own but are still considered AI. Instead, they “fake” the process of coming up with a solution on their own, and therefore it’s considered fair to include them under the AI umbrella. (And before the recent successes of machine learning techniques, they were the bedrock of AI.)
I’ll make it clear what I mean in a moment, but first I’d like to say that by using the word “fake,” I don’t mean to diminish the usefulness and effectiveness of other non-ML based methods. It’s just how I describe them in order to better understand and explain their relationship with the definition of artificial intelligence above.
How can we “fake” coming up with a way to solve a problem? The “fake” process is simply trying out many different solutions and picking the one that works best. This will not work with all problems, but it’s effective for some. The problem has to be one for which checking a solution is easy, even if coming up with one is not.
A computer is faster than humans in many ways. That’s why this method can be applied by a computer where a human would have to do something else like strategize, do pattern recognition, and all that magic that the brain does. But a computer can figuratively use “brute force” to solve the problem. And often, this method, i.e. searching for a solution among many possibilities, is a more efficient way for a computer to solve a problem.
For example, let’s consider the problem of finding the shortest road path from point A to B, which is what GPS navigators need to do. A person would look at a map and do this by sort of… “eye balling” it. It’s hard to describe how we, humans, do it, right? We could try to come up with a set of rules that do what humans do in this case using ML, but we can also try all paths and pick the shortest one we find.
Sometimes there would be a very large number of paths to try, but we can also quickly eliminate many obviously bad choices, using some simple rules, such as, “if you’ve already found a path of length x, don’t go to an intermediate destination farther than x (in an easy-to-calculate, straight-line distance) in future paths you try.” Coming up with clever ways to cut down on the number of possibilities is the tricky and clever part of inventing these algorithms.
Conclusion
Now you know what AI is… at least for now, because what is commonly considered AI might change. Perhaps in the future, only general artificial intelligence that thinks like humans do will be called AI. But, for now, you are all set.