What is Modern AI? – From Pac-Man Ghosts to ChatGPT

The rise of tools like ChatGPT, Google Bard, and DALL-E 2 have brought AI into most people’s lives very quickly, accessibly, and cheaply. AI isn’t new term to a lot of people; most people know it means artificial intelligence. However, for quite a few years now AI has been used in many contexts, perhaps the most common one is in games.

Whether it is Luigi, Toad, and others as you race in Mario Kart, the ghosts in Pac-Man, or the opposing players on FIFA; all of these are known as AIs in the gaming world, but are they truly artificially intelligent? To many people’s disappointment, they are not. That might seem a bold statement to have made but let’s look at how the understanding of AI has been laid, compared to what it is these days.

What has AI been used for in the past?

As previously stated, games use the term ‘AI opponent’ very loosely and frequently. The gaming industry have just been a little lazy here, what they should really be saying is,

“Opponent controlled by a series of rules with set outcomes based on a players input”.

Obviously, that is far too long winded and boring to listen to.

This is all that they are doing though, the ghosts in Pac Man are told where the player has moved to, and they recalculate their directions based on this and their current movement. So, they are just following a list of instructions. Even with a game like FIFA, 11 players moving around to stop you scoring, and the 10 players on your team moving to help you score. If you watch them carefully you can spot patterns in their movements, especially on the easier difficulties, but to make it look more intelligent the players on the same team will ‘talk’ with each other so there is never a situation where all the players are chasing the ball. That would be chaos.

Away from games, early website chat bots were claimed to be AI as well. Now all these were doing was taking your question, for example, “What are your terms and conditions?” and conducting a search of the web pages available to it, or list of defined responses, and returning these back to you but in a little more human way. Such as saying, “Here is a link to our terms and conditions”, rather than just returning the link like Google would. But again, this is still not being intelligent it is just clever programming.

So, what is AI in the modern sense?

AI is now actually intelligent, well relatively. ChatGPT for example, was fed information from the world up until 2021. So, if you were to ask it who the monarch of the UK was, it would reply ‘Queen Elizabeth II’ but swiftly followed by a disclaimer stating that it is only trained on data up until 2021 and so this could have changed, basically it is doing like Johnny 5 and asking for more input.

Ask Google Bard the same question, this will give you King Charles III because it is linked to the internet as its knowledge base. But how is this different to those chat bots fetching web pages?

Well, if you ask ChatGPT for information, and then ask it to format it into a table, it will know what you mean and execute your request. It also doesn’t get told how to respond; it teaches itself. For example, ask it to give you a list of motivational quotes and it will but without their authors. Ask it to give you the authors and it will. But if you keep the same conversation thread going, after just a few “give me quotes” followed by “with authors” you will no longer need to say with authors. It will just serve you the quotes with their authors.

Surely it is just another program with another set of rules? Aren’t they just set responses? These are valid questions; however, the truth is the model works by using a neural network with thousands of nodes. As the name suggests this works similarly to a brain. At a very high-level view, it takes information, stores it in a node and then makes links to other, similar nodes. Then when queried it follows the nodes and links that they have made to provide you with your answer. It can understand your questions and requests by using natural language processing (this is how Alexa, Google home, and Siri can understand our voices as well).

Therefore, the answer to our question, ‘what is modern AI’ can simply be it trains itself to give responses to input, based on the information that is given to it by its human creators. Not to be confused with the delight of Skynet, which is artificial general intelligence (the ability for a machine to teach itself to do things without being trained by a human on how to do them), we are a long way off from this.