The Fundamental Theories Behind Artificial Intelligence

Original article was published on Artificial Intelligence on Medium

AI Today

As we saw earlier, much of what was discussed in that initial proposal is reminiscent of the current dialogue surrounding AI. In fact, it was advancements in some of these original ideas that allowed companies like Google, Amazon, and Apple to build their AI technologies.

Neural networks

At its most basic level, a neural network is a set of algorithms — but this is sort of like saying that a human brain is a collection of atoms. Algorithms are the building blocks, though, and these groups of algorithms are specifically modeled (loosely) on the human brain.

A neural net can include thousands or even millions of individual, complexly interconnected processing nodes. These nodes are designed to recognize patterns, much like the human brain does. The advancement of neural networks has made possible advancements in the field of deep learning — a branch of machine learning, and an integral part of modern AI.

Machine learning and deep learning

If I had to point to one field and say, “That was it. That is what made modern AI possible,” it would be machine learning. Machine learning is the idea that computers can learn from experience. These algorithms use statistics and massive amounts of data to identify and utilize patterns.

You’ve likely heard about the many, many privacy issues plaguing users on the internet today. We all know why companies want to harvest our data — money. But how does my data translate to dollars and cents? Some companies sell data for profit. Others, though, feed all of that data — every mouse click, every video watched, the length of time the video was watched, where I went next after the video was over, etc. etc. — into these algorithms. These highly sophisticated, perfectly tuned algorithms digest all of these patterns to create startlingly accurate predictions of user behavior. The ingestion and analysis of this data is how we get suggestions on what to watch next, and end up watching for hours and hours and hours.

If this wasn’t complex enough already, deep learning is an even more complex, more evolved version of machine learning. It involves the most sophisticated neural networks and is capable of picking up on the most subtle behavioral patterns.

Supercomputers

Image source: Wikipedia

If you’ve ever heard your MacBook Air start making odd sounds and heating up after one of these long Netflix binges, you may be asking yourself how in the world this type of computing is possible. For a very long time, it simply wasn’t.

Above is a picture of a MacBook from 1995. Now consider whatever device you are reading this on right now, and think how far we have come in even just my lifetime. This feat is nothing short of awe-inspiring. Now think about how far we have come since McCarthy, Minsky, Rochester, and Shannon spent their summer at Dartmouth. Wild.

These advancements in computing power have been crucial to the success of AI. It takes incredible computing power to be able to run the machine learning algorithms needed for artificial intelligence. Some people don’t realize this, but computing power is a significant source of the limits in technology. For instance, modern password guidelines and encryption are only secure because we don’t have computers that are capable of brute force hacking them (yet). Sure, it would take billions of years to do it, but that is based on today’s computing forces. As computers become more powerful, that timeline shrinks.

Big data

Everything above really comes down to one thing: data. The unimaginably massive volume of user data that exists is the fuel in the rocket ship that is artificial intelligence. Yes, without neural networks, machine learning, and computers capable of running both, the data would be relatively useless.

But the data is really what machines learn from. The more that a machine gets, the smarter it becomes — the more refined its knowledge is and the more accurate its predictions are. Everything that we think of when we think of AI, whether it’s talking to our Google Home or getting creeped out upon seeing an advertisement for that thing you were thinking about this morning, is born out of analyzing data sets that are so massive, they can guess how to successfully interact with you. Maybe you’ll be a little nicer to Siri next time she doesn’t know how to respond.