What is Artificial Intelligence

Useful software utilities we can't understand or reproduce

In the quest to understand artificial intelligence (AI), many treatments of the subject tend to veer into philosophical territory. Some focus on whether AI is conscious or whether it is equivalent to human intelligence. While these questions may be intriguing, they are ultimately unhelpful, partly because they lack a clear definition of what intelligence is. This lack of clarity has led to past efforts at creating AI being considered merely programming today. When people look at these creations, they simply think of them as cleverly designed software, and no claims about them being artificial intelligence survive.

A witty way to skirt around the problem is the "you know it when you see it" approach. However, this too is unhelpful, as people from different generations have developed different intuitions about what constitutes AI. It may not even be that important to know what artificial intelligence is, so long as we can use it effectively.

One helpful way to distinguish AI from other types of software is to define it as a software utility that we can't explain or reproduce. The technique used to produce current AI software, machine learning, has exactly that property. It does useful things, but we can't fully explain how it does them, and even the most accomplished programmer couldn't hope to reproduce what it does in code.

That doesn't mean that we don't understand in principle how AI works - the architecture of deep neural networks and the process of training models using data are well-known concepts. However, once a model is trained, it becomes difficult to discern exactly how it encodes the knowledge and capabilities it possesses.

Instead of focusing on having a deep philosophical understanding of what artificial intelligence is, we should treat complex machine learning models as black boxes. These models have an input and an output, and the capability to transform from one to the other. We can use this transformation without understanding exactly how it works – and that's good enough for us to harness the technology to our advantage.

While it may be tempting to delve into philosophical debates about the nature of artificial intelligence and consciousness, it is more practical to focus on how we can use AI as a tool. By treating complex machine learning models as black boxes with inputs and outputs that we can harness without fully understanding their inner workings, we can continue to make progress in applying AI to solve real-world problems.