Question: I’ve always wanted to know how Artificial Intelligence works. I strongly agree that it can be a big help one day but don’t understand how far it can really go. What are your views?

Carl Peter Robinson answered on 2 Feb 2024: last edited 2 Feb 2024 12:59 pm

Great question, on several levels. In the past, AI consisted of numerous theoretical and practical areas, dependant on the capability of the computation hardware at that time. These areas included things like computational logic, expert systems, knowledge based systems, etc. These days, the most common field in use under the umbrella of AI is something called machine learning (ML). If you want to know how AI works, start reading about machine learning.

ML is heavily based around several mathematical areas including, but not limited to: probability, statistics, linear algebra, and calculus; depending on which ML algorithm you look at. In fact, when you hear someone talking about an “ML model”, underneath that fancy term is a whole bunch of mathematical methods and operations, put in place most often by humans writing software (e.g, in Python).

In a nutshell, an AI system works by being given some data, or lots of data. The system performs those mathematical operations I mentioned earlier on that data, to try to discover any patterns in it. The system then outputs its results based on any patterns it has found. This output could consist of several statistics according to the instructions the system has been programmed with. In terms of ML, I like how Francois Chollet puts it in one of his books, “A machine-learning system is trained rather than explicitly programmed.”

To be honest, I don’t know how far developments in AI can really go. I don’t think anyone does for sure, but we can all speculate and make predictions. Personally, the way I see the near-future developments going are that the AI systems and tools that will be released will be more powerful and more capable for the tasks they are created to do. But, like I said, they will still be tools we use to help us. I don’t see anything yet that points to an AI system in charge of humans or in full autonomous control of a decision-making sector of society. We’re also in a great era of AI discovery, so I think there is potential for new ML model architectures to be created that will perform as well as the massive models we have now, but on fewer resources, and on edge devices, like your smartphone. Looking in these new directions may also finally help us crack artificial general intelligence (AGI), whereby a system is built that can perform all kinds of tasks for us, including applying very human traits like reasoning, and even a kind of intuition.

If you’re really interested, here are some books I can highly recommend:

– The first is Max Tegmark and his book Life 3.0. He describes possible AI futures really well in here (https://en.wikipedia.org/wiki/Life_3.0).

– Another is a book by Ray Kurzweil, a proponent of Futurism who believes we will reach true AGI by 2045 (https://en.wikipedia.org/wiki/The_Singularity_Is_Near).

– And thirdly, look up Stuart Russell, a British professor who has done a lot of research into AI, and his book Human Compatible (https://en.wikipedia.org/wiki/Human_Compatible).