AI

What is AI?

AI stands for Artificial Intelligence. It's a broad term that has many meanings. Mainly, it refers to making a computer program behave in some way that appears intelligent, able to make decisions, come to conclusions, and act by itself, often like a human being. What this means exactly varies a lot from context to context.

In its simplest form, AI could be referring to the AI of characters in a video-game. This AI could be an extremely simple hand-written algorithm, such as "if the player is in front of you, shoot at him; if the player is shooting at you, hide." With this, we will have the appearance that the character is making decisions by itself based on what it can observe, even though these decisions are actually very mechanical and triggered by simple conditions.

More serious AI, as in AI research, refers not to AI that has more complex decision algorithms, but typically to AI that uses statistical models to calculate the probability of a decision being the correct decision for a set of conditions. For example, an AI that can tell whether a photo is a photo of a bird or not. This would be an image-recognition AI. In this case, we would need a model that can give us an answer based on an image, and we would need to "train" this model to give us the answers we want. This is also called "machine learning," abbreviated ML. We would give the AI a bunch of photos that feature birds and tell the AI that those photos contain a bird, and also give the AI a bunch of photos that don't feature a bird and tell the AI that those photos don't have a bird in them. Machine learning refers to nudging some parameters in the model, little by little, until we arrive at a set of parameters that, ideally, always says that photos of birds contain birds, and photos without birds do not contain them. Such set of parameters may not actually exist for a given model. That's because when we check if an image contains a bird or not, we're actually just doing normal programming. It's just an algorithm that uses a model and an image as input, and gives a "yes" or "no" as output. The "AI" is in the fact that the model parameters are trained automatically, and not specified manually, and thus we can have an immense number of parameters that no human would be able to manually adjust. This same notion applies to AI that can understand instructions in arbitrary plain English, generate arbitrary answers in plain English from its data, and nowadays to generate "AI Art" (images, music, etc.). These all work on the idea that, given an input, there is a large number of answers, but the trained parameters say a specific answer is the most likely to be the correct one.

Finally, the holy grail of AI is AGI: Artificial General Intelligence. This would be a computer program that can truly learn to do anything. If such program ever comes to exist, its growth will only be limited by hardware. It would be able to become smarter and smarter indefinetely, learning more and more until it learns everything there is to learn.