An explanation of the different types of AI

Sabrina Polin, Managing Editor

In this video, TechTarget editor Jamison Cush will talk about the four main types of AI.

Do you really know what AI is? Well, it might just depend on the context.

AI is a spectrum, with experts considering some AI narrow or weak and some closer to true artificial general intelligence, or strong AI. Here, we'll talk about the four main categories of AI.

You can sort AI into four categories: Reactive, limited memory, theory of mind, and self-aware. Let's talk about each.

One, reactive AI -- most machine learning models fall under reactive AI. These models use statistical math to consider huge chunks of data, then produce a seemingly intelligent output.

Reactive AI is suitable for simple classification and pattern recognition tasks. For instance, it's responsible for famously defeating chess Grandmaster Garry Kasparov in 1997; chess is a numbers game and AI models can compute all possible moves faster and more accurately than any human could.

Also, consider Netflix -- while it seems like Netflix is thinking about what content you'd want to watch, it's all just math, using other customers' purchase history to base a recommendation algorithm off of.

But reactive AI is incapable of dealing with imperfect information that humans, in comparison, can easily navigate; we are masters of anticipation and work with imperfect information all the time.

Two, limited memory machines -- we are currently here with AI development; conversational AI chatbots and self-driving cars fall under this category. Limited memory machines use deep learning, which imitates the way humans gain knowledge. This type of AI can handle complex classification and use historical data to make predictions.

This graphic states that this video won a national silver award from the American Society of Business Publication Editors.

And while limited memory machines can far outperform humans at certain tasks and autonomously improve these tasks over time, this is still considered narrow AI because they require huge amounts of training data to learn a task that a human can learn with just a few examples. For instance, give a toddler two or three photos of a cat and they'll generally know what a cat is. But it would take thousands of photos for a limited memory machine to learn the same thing.

Three, theory of mind AI -- theory of mind AI is -- hypothetically -- capable of understanding human motives and reasoning. Also referred to as artificial general intelligence, it can learn with fewer examples than limited memory machines, can contextualize information and extrapolate that knowledge to a broad set of problems.

We're still far from achieving theory of mind because of understanding, which is a main limitation of AI. Currently, an AI model could write an essay or book accurately, but it won't understand what it just produced; theory of mind AI would.

Four, self-aware AI -- while theory of mind AI can understand others' emotions, self-aware AI refers to systems that are aware of their own internal state as well as others. In other words, a machine that is on par with human intelligence and emotions.

This is sometimes referred to as artificial superintelligence. But this level of AI is not even comprehensible, as we ironically just don't know enough about the human brain to build an artificial one.

Do you think artificial general intelligence is achievable? Should it be pursued? Share your thoughts in the comments and remember to like and subscribe.

Sabrina Polin is a managing editor of video content for the Learning Content team. She plans and develops video content for TechTarget's editorial YouTube channel, Eye on Tech. Previously, Sabrina was a reporter for the Products Content team.