Artificial intelligence is as easy as rocket science

In my work with artificial intelligence I have realised that what I thought was scientifically complicated, is actually as easy in practice as rocket science. This means that I, along with many others, in the development of artificial intelligence (AI), have placed my focus completely wrong. If you are, or will be, developing AI and want to avoid the same mistake, keep reading.

As we all know, rocket science is a very simple science. A rocket is just a giant, yet simple combustion engine in a tube. A little spark at the bottom and it's off.

It is similar with the algorithms behind the artificial intelligence we see today - simple off-the-shelf algorithms that any student at ITU can take off the shelves, add fire to (or data) and then there is magic.

But why was it so challenging to get a man on the moon, and why does it take so much investment to get an AI project off the ground?

Many rockets have been built over time, and the blueprint of a rocket is no secret. The same goes for AI algorithms, which are freely available and ready to use. But as soon as you need to get a man on the moon, it suddenly becomes difficult, and the same is true when you have to apply AI in practice. The hard part about getting a man on the moon is everything that lies around building the rocket itself. Granted it has its challenges, but, generally speaking, a rocket is just a big combustion engine in a tube.

The real challenge is all that comes with it. To get a man on the moon, you need to make sure the rocket actually reaches the moon, you must protect it against radiation in space, find a safe way to land on earth again, etc. Let us take radiation on the moon as a specific example of a problem. How do you find out what radiation exists on the moon? This is done by researching the moon's atmosphere and crust. And how do you do that? Every time a problem surrounding the core problem needs to be solved, you have to go down several layers just to get one step further.

In the same way, as you move along AI development is about everything but the rocket itself (the algorithm). This can be for example finding data, purifying and preparing it, and perhaps even after all that work, the data still doesn't represent the world you are trying to emulate in your artificial intelligence. Just as with the space rocket, it can also be hundreds of other things.

So, one of the biggest pitfalls I've seen in developing artificial intelligence is the huge focus placed on the algorithm itself, whilst very little is being placed on all that lies around it- despite the fact that this is where the real challenge lies.

If you focus all your energy on the rocket or algorithm, you could be left with a technically good product that is just not applicable in practice, which, simply said, would be a huge shame.