There is the popular understanding of artificial intelligence, which involves sentient robots acting like humans — and sometimes trying to take over the world. Then, there is the practical understanding of artificial intelligence, which concerns more mundane computer programs crunching data for business success. How can both of these concepts fall under the umbrella of AI?
Frustratingly, AI is an exceedingly broad category of tech, which can make talking about AI confusing and difficult in different circles. To help you better understand the promises and limitations of AI, here is a rundown of the four major types of AI in development today:
The vast majority of AI solutions fall under the category of reactive machines. These tools don’t form memories or learn from mistakes; rather, they are built only to react to the data put immediately before them. Machines built with this type of AI make direct perceptions and come to decisions without developing any kind of internal concept of the world.
A perfect example of a reactive machine is Deep Blue, the AI supercomputer created by IBM to play chess in the late 1990s. Deep Blue was not programmed to remember past chess moves it or its opponent completed; instead, it made predictions about future moves based on the current layout of the board. In doing so, Deep Blue was able to beat international grandmaster Garry Kasparov — but after winning the game, the AI didn’t remember how it reached its victory, let alone celebrate the success.
Reactive machines are among the largest and most important category of AI, especially in the practical world. Business leaders are most likely to learn about this type of AI in an online AI course on business strategy, and enrolling in such a course is a good idea if you struggle to understand how AI tools could contribute to your business goals.
Some AI tools have a limited memory, which is what sets them a step above reactive machines. By being able to remember a small amount of the past, these machines are capable of making slightly more informed decisions with a slightly larger data set. Though the simple memories stored by limited memory AIs tend to be transient — that is, they aren’t saved into a library of experience to inform future decisions — they can be vital for creating certain types of AI functionality.
For example, self-driving cars tend to be limited memory machines. Unlike Deep Blue, which can look at a static chessboard to gain all the data it needs for its next decision, automated vehicles are moving through a dynamic world. Thus, they need to be able to monitor objects around them over time to determine speed, direction, and other important factors.
Limited memory AIs are more difficult to build than reactive machines, and their temporary memory capabilities are rarely applicable. As a result, you are unlikely to encounter many limited memory machines in the real world.
Theory of Mind
In psychology, “theory of mind” is a cognitive skill that entails understanding your own mental and emotional state as well as those of the people around you. In tech, this third classification of AIs includes machines that can do just that — form representations about the world and other entities within the world. This is an incredibly important leap in artificial intelligence, and it is one that AI developers have yet to achieve in truth.
Developing a digital theory of mind is a critical step toward creating the type of thinking, feeling AIs imagined in science fiction. It is also key for creating machines that can work hand-in-glove with a human workforce. Until the theory of mind emerges for AIs, AI solutions will remain mere tools.
Finally, the last type of AI — and another we have yet to conquer — is those AI programs that have gained self-awareness. This will not be an easy feat; consciousness is not something we fully understand in ourselves, so recreating it in machines is going to take time and effort. Still, it is possible that many will not accept the existence of AI until we have machines that are truly sentient.
Artificial intelligence is all around us in the 21st century, but not all AI is created equal. By knowing a bit more about the types of AI, we can use different AI systems appropriately and accept new AI tools as they come.