The main types of artificial intelligence

In order to fully understand what AI is and how it works, one must take into account the current state of artificial intelligence and the potential scenarios towards which it can evolve as the technology is further developed.

According to the original outline of Arend Hintze, a professor at Michigan State University, there are four main types of AI. This categorization spans from the way we’re used to interacting with AI today, to the more “sci-fi” view of how AI might function in the future as sentient systems.

01. Reactive machines

In reactive machines, the AI’s main goal is to complete a task by reacting to the information presented to it. This type of artificial intelligence system isn’t able to store memory of previous data, therefore it can’t use data in order to fine-tune its responses to a present task. For this reason, reactive AI machines are generally used to perform specific tasks with set outcomes rather than learn from a multitude of different scenarios.

One of the most famous examples of reactive machines is IBM’s Deep Blue, a supercomputer built to play chess and ended up winning in a game against then-grandmaster, Garry Kasparov. While Deep Blue was able to look at a chessboard and identify chess pieces and potential moves, its intelligence was limited to making predictions on moves and taking the most logical next move. The machine wasn’t able to learn about its opponent by gathering data about his habits, game-play flaws, or signature chess moves.

02. Limited memory

Unlike Deep Blue and other reactive machines, a limited memory AI system is able to learn, to a limited extent, from the information it has already seen in order to inform its future actions. The opportunities with limited memory AI systems are a lot greater since they’re able to improve their behavior using the data they’re exposed to.

In order to create this limited memory, human teams need to train the AI system with a model so that it can learn to analyze new data. The machine needs to be consistently exposed to new data so that when it’s faced by a user, it has the existing memory necessary to predict what comes next. An example of limited memory technology is self-driving cars, which are exposed to enough data and models of different driving scenarios so that it can make its own decisions when on the road.

03. Theory of mind

Theory of mind AI systems have a much deeper psychological core, as they’re able to read and interpret human emotions and learn from social intelligence in addition to raw data. We have yet to achieve this level of artificial intelligence in our society, however, AI programs falling under the theory of mind category would be able to understand how humans make decisions based on emotions so that it could more accurately predict behavior. This would allow for more of a symbiotic relationship between man and AI-powered machines.

04. Self-awareness

The self-aware type of artificial intelligence also does not exist, but might conjure up images from films of robots taking over humanity as we know it. While that scenario is highly unlikely, the notion of AI developing into something with consciousness is the final type of artificial intelligence technology.

In addition to being able to understand the psychology and emotions of others as we saw in the theory of mind programs, this type of machine would also be aware of its own existence and place in the world. However, for now, this kind of AI remains the stuff of science fiction as it will take tons of advanced research into fully understanding and reproducing a human-like consciousness.

Weak vs strong AI

Another way that we use to divide the different types of artificial intelligence is by categorizing them as weak and strong, also known as narrow and general.

Weak (or “narrow”) AI

Weak AI refers to the kinds of artificial intelligence that we’re used to in our day-to-day lives. In other words, weak AI is the type of machine that’s meant to complete a set task very well. While these types of systems might seem highly intelligent, they’re functioning within boundaries that limit the level of intelligence they can achieve.

Examples of weak or narrow AI include any type of software that automates or analyzes data, virtual assistants like Siri or Alexa, and even weather apps. This type of artificial intelligence programs are more focused on making our lives more efficient, instead of simulating real human intelligence in all its capacity.

Strong (or “general”) AI

Strong AI, also sometimes called Artificial General Intelligence (AGI), refers to artificial intelligence systems that, at the moment, only exist in the movies. Robots from films such as I, Robot or in the series Westworld exemplify the extreme sides of AGI.

In reality, strong artificial intelligence in the future might look like AI systems that are able to completely mimic the scope of human intelligence, including emotion, creativity, and adaptability in order to fulfill tasks. However, unlike in dramatised versions of artificial intelligence machines in movies, general AI is likely to assist and expand our abilities as humans rather than replacing them entirely.

Generative AI

At its core, Generative AI  refers to a type of artificial intelligence that can generate new content, from images and music to text and code. In contrast to earlier versions of artificial intelligence, generative AI isn't just analyzing data and collecting insights, but is actually creating something new from what it has learned.