What is Artificial Intelligence (AI)?
The term AI is being used a lot more frequently than machine learning for many reasons. One key reason is that it has used in general literature, movies etc while machine learning has used typically more by folks in the sciences. There are a lot of issues with the term AI though. To start with, there needs to be a clear definition of “Intelligence” itself. The reality is that there is none. Part of the problem is that intelligence is a spectrum all the way from the intelligence within a single cell organism to human intelligence (assuming we are focussing on intelligence only of a living system). If you just pick human intelligence, it is also spectrum all the way from the brain of a newborn to an adult brain. And even within the adult brain, there are different abilities, all the way from complex numerical computation to recognizing images to even more higher-order complex creative tasks like composing music or creating art and so on. So the context in which the term AI is used, matters. Two very broad ways AI is used are the following:
- The science-fiction version of the definition of AI would be a machine that can do every single thing that an adult human brain can do equally or better. The extreme end of that being human self-awareness or consciousness of course. One of the interesting things about the human brain is its ability to demonstrate “general” intelligence. This means that you can teach a human anything new in any domain (of course keeping its limitations like memory capacity etc in mind). In other words, the human brain automatically adapts and learns even with new “inputs” that it has never seen ever.
- If you go beyond science fiction, the term AI is used to refer to a machine that does a specific task that is equal or better than a human brain doing the same task. For example, numerical computation is one of those tasks. So you could (and Ray Kurzweil does) say that today’s computer and cell phone are already an AI because they can do specific tasks much better and faster than any human brain can. For a long time (from 1950’s until very recently), machines were not able to do certain higher-order tasks that a child’s brain does quite easily including things like recognizing images, understanding natural language etc. This has however dramatically changed in the last decade or so due to a variety of software/algorithmic breakthroughs, advancements in hardware (GPUs) and availability of a large amount data, all coming together to create a perfect storm! And so now AI is quite often used to refer to a machine that is capable of higher-order tasks like image recognition, natural language processing, playing very complex games like Chess and GO etc.
Based on these two broad distinctions, you can see AI from the lens of a machine that is capable of special intelligence vs one that is capable of General Intelligence. Demis Hassabis of Deep Mind looks at it this way. In his view, machines like IBM’s Deep Blue that was able to beat the best human chess player, IBM’s Watson that was able to beat humans in Jeopardy and Google’s Alpha Go that was able to beat the best human GO player are all of the type ‘artificial special intelligence’. Meaning that they are able to beat humans on one specific task at a time. His goal is to build artificial general intelligence (or AGI) that is capable of General Intelligence¹ meaning a single machine that is capable of doing everything equal or better than an adult human brain. This is when the real world will meet the science fiction of today!
Next….Why are we obsessed about comparing machines with the human brain?
References:
