The Truth about AI

There really is no formal, universally accepted definition of Artificial Intelligence (AI), but there are a few things that most definitions can agree on.

1. AI is intelligence of a machine and not that of an organism.
2. AI systems can learn to adapt their intelligence.
3. Intelligence here is a spectrum. This means that at one end a fully functioning human is capable of many things but at the other end, a calculator can only do a few tasks. The measure of intelligence is not of kind, but of scale, speed, degree of autonomy, and generality ^[1].

The flexibility of the AI definition leads many companies to claim they have developed an AI solution. Given that intelligence is actually a spectrum these claims may not necessarily be false, but the real question is where on that spectrum does a solution fall?

Here are a few helpful things to consider when evaluating an “AI” system.

• Artificial General Intelligence, or the ability for a machine to fully emulate the ability of a human to understand and learn any task, is probably not achievable (at least in the near future).
• Any solution claiming to utilize AI is not general but falls somewhere between that and a calculator.
• Generally (but not always), AI systems learn via some machine learning algorithm like neural networks, gradient boosting, random forests, or the like.
• Deep learning is nothing more than a neural network with many layers.
• Most AI systems perform a very narrow set of things. Those that are not narrow are easily recognizable, such as a self-driving car.

So… where on that spectrum does a solution fall? Of perhaps greater practical importance than the strict definition of AI is will a given solution deliver the value I expect?

The real problem is that AI is an over-used buzzword. Many can claim to have it in their solutions. Not all that claim to use AI, actually do.