Question: What is AI really caperable of?

Carl Peter Robinson answered on 21 Jan 2024: last edited 21 Jan 2024 8:35 pm

I like this question! It’s one we should constantly be asking ourselves, both in terms of what can AI do for us and what do we want AI to do for us.

I honestly don’t know what an AI system would be capable of in the future – we may find a brand new architecture that cracks a whole lot of scientific and philosophical conundrums regarding this topic. But until then, I’d say anything that involves analysing, interpreting, categorising, identifying and predicting specific types of data is up for grabs.

However, here we should go back to my earlier statement and ask whether it is worthwhile creating an AI system for a specific task. For example, ChatGPT is really good at generating text. You can ask it to write a poem and it will do an okay job. But now imagine, what if you wrote a poem as part of a poetry competition. Would you want ChatGPT to be the judge of your poem or would you prefer a human judge? After you answer that question, ask yourself why you gave that answer.

AI systems are good at finding patterns in the data they are given. These patterns could be, for example, grouping similar data together, finding features of particular data, identifying specific data points, and separating data into categories. From that, these AI systems can be used to do things like make predictions about future data points, that either do or do not exist yet, or they can classify new data based on existing data they were trained with.

From what we’ve seen so far, AI systems are good, mostly, at categorising existing data and making predictions about what new data should be or look like. They’ve also proven themselves to be good at replication and some form of interpretation (which some people see as them showing signs of creativity). But their output has still been found to contain errors at times so we should certainly not take what they output as 100% accurate.