Solving conventional computing's problems

To solve computing problems, we use machine learning (ML) technology.

However, we need to remember one distinction between machine learning and artificial intelligence (AI).

By the very bare-bone definitions, AI is a term for replicating organic, or natural, intelligence (that is, the human mind) within a computer. Up until now, this has been an impossible feat due to numerous technical and physical limitations.

However, the term AI is usually confused with many other kinds of systems. Usually, the term is used for any computer system that displays the ability to do something that we thought required human intelligence.

For example, the IBM DeepBlue is the machine that played and won chess against the world champion, Garry Kasparov, in 1997.  This is not artificial intelligence as it doesn't understand how to play chess; nor does it learn how to play the game. Rather, humans hardcode the rules of chess, and the algorithm plays like this:

• For this current chess board, what are all the possible moves I could make?
• For all of those boards, what are all the moves my opponent could make?
• For all of those boards, what are all the possible moves that I could make?

It'll do that over and over, until it has a tree of almost every chess board possible in this game. Then, it chooses the move that, in the end, has the least likelihood of losing, and the highest likelihood of winning for the computer.

You can call this a rule-based system, and it's a stark contrast from what AI truly is.

On the other hand, a specific type of AI, ML, gets much closer to what we think of as AI. We like to define it as creating mathematical models that transform input data into predictions. Imagine being able to represent the method through how you can determine whether a set of pixels contains a cat or dog!

In essence, instead of us humans trying our best to quantify different concepts into mathematical algorithms, the machine can do it for us. The theory is that it's a set of math that can adapt to any other mathematical function, when given enough time, energy, and data.

A perfect example of machine learning in action is IBM's DeepQA algorithm which went behind Watson when it played and won Jeopardy!, Watson played on the game show against the two best human competitors on the game show, namely Ken Jennings and Brad Rutter. Jeopardy, is a game with puns, riddles, and wordplay in each clue—clues such as This trusted friend was the first non-dairy powdered creamer.

If we were to analyze this from a naive perspective, we'd realize that the word friend, which is usually associated with humans, simply cannot be related to a creamer, which has the attributes the first, powdered, and non-dairy. However, if you were to understand the wordplay behind it, you'd realize the answer is What is coffee mate?, since mate means trusted friend, and coffee mate was the first non-dairy powdered creamer.

Therefore, machine learning is essentially a set of algorithms which, when combined with even more systems, such as rule-based systems one could, theoretically, help us simulate the human mind within a computer. Whether or not we'll get there is another discussion altogether, considering the physical limitations around the hardware and architecture of the computers themselves. However, we believe that not only will we not reach this stage, but it's something we wouldn't want to do in the first place.