Take a look at this:

It’s my hand. Pretty nice hand, isn’t it?
Now check out the rear:

Pretty cool, huh?
It’s a very simple instrument: it consists of five fingers, one of which is a thumb. Yet it’s a remarkably useful tool. I can do all kinds of things with it!
I can play guitar with my hand.
I can hold a glass of beer with my hand.
I can write articles with my hand.

I can cook food with my hand.
I can drive a car with my hand.
I can play video games with my hand.

My hand can be utilized for a vast range of tasks, as long as it is powered by my brain. Therefore, my hand is an example of a general-purpose algorithm, or, as people in the buzzword biz like to call it, "general AI".

There are countless of definitions for the term "Artificial Intelligence". One of the many ways to categorize AI is through the terms "general AI" and "modular AI". A general AI is a tool that is able to execute a very wide range of tasks, whereas modular AI is made for one very specific purpose. For instance, most of the AI that you may see in Hollywood movies are examples of general AI: be it The Terminator, the AI in Her, or the robots of I, Robot. They often look or act like humans, and are capable of doing most things that humans are able to do, often stronger, better, and faster.
Okay. So general AI is quite neat.
We should put it in a computer.

Well no one really knows how.

There are countless of books on the subject of general AI. Many speakers, many podcasts, and many publications have discussed and continue to discuss this form of AI, and the implications it will have on not only businesses but on society at large, yet man-made general AI doesn’t actually exist.
Hmm.
Then how was my __ hand created?
Evolution > My Hand > Human Algorithms
My hand, and incidentally also the rest of me, was created by Evolution. You may not have thought about it before, but evolution is actually an algorithm. It is the longest running known algorithm in the world, and it is running on the most powerful system: our planet. In the same way that software runs on a computer, and my hand runs on my brain, evolution runs on planet Earth.
Evolution is also one of the five main schools of machine learning (a subcategory of AI), the other four being symbolists (logicians), connectionists (neuroscientists), Bayesians (statisticians), and analogizers (similarity theorists).

We already know how to emulate evolution. For instance, we can apply evolution theory to the classic video game Super Mario Bros. In it, the player, controlling Mario, tries to make it as far to the right of the screen as possible, overcoming various obstacles in the way. If you were to train an AI to beat Super Mario Bros. as efficiently as possible, evolution theory would be a perfect algorithm to adopt. Give birth to 100 different Marios’, each with somewhat different variables (stats), and see who can make it the furthest to the right. The winner gets to give birth to 100 more Marios’ — a second generation — each inheriting slightly altered DNA from the winner, until you have created the perfect Mario. Another, more practical example of using evolution in computers, is the creation of a spam filter. Many e-mail spam filters use algorithms based on evolution theory, where words that indicate spam are being paired to give birth to children terms.

Regardless, going back to the example of Super Mario Bros., that perfect Mario that we created is nonetheless an example of modular AI. All he is good at is playing Super Mario Bros. His only skill is being able to go from left to right without falling to an obstacle.
Yet evolution could be used to create a general-purpose algorithm, as could any of the other four schools of machine learning, at least in theory. But it’s incredibly complicated. It requires an immense amount of training, along with clearly defined routines and processes. What more: it requires an actual business case.
The Business Case of My Hand
AI is a buzzword. It could mean absolutely anything. But when you are looking at actual applications that exist today, that claim to use "AI technologies", they are always using some form of modular AI. They have been built for a very specific purpose. It’s like a hand that has been made exclusively for playing piano.
It isn’t only that General Ai is hard to build, but it is also hard to prove its business value. When creating AI solutions, it is typically better to build a skateboard first, then a scooter, next a bicycle, after that a motorbike, and then a car, rather than making the various components of a complex car straight away. Similarly, creating a hand that excels at doing one thing has a stronger business case. You can see a clearer return on investment, and you have a stronger case when selling the solution. Clients don’t want a massive, comprehensive, hard-to-digest solution, they want a solution to their immediate specific issue.

Yet many great thinkers and engineers are debating and building general AI that works very similar to my Hand. Some argue that a general-purpose algorithm would be very complicated, featuring potentially millions of lines of code. Others argue the opposite, that general-purpose algorithms could actually turn out to feature but a few hundred lines of code.
My hand is a perfect example of why. Provided the data model – and the data – is strong enough, powerful algorithms could turn out to be surprisingly simple.
Like my cool hand.

Interested in artificial intelligence? Here are a few more articles you might enjoy.