What exactly is artificial intelligence, and what are the different types of AI worth knowing about?
What if hyper-fast machines could carry out complex pattern spotting in mere hours, rather than in the days or years that it takes humans to do?
If you’ve seen movies like Her or Ex Machina, you’ll have an idea of what we’re talking about: artificial intelligence (AI). While the types of AI represented in sci-fi movies used to be a far cry from anything we have in reality, in 2023 things changed with the arrival of ChatGPT and its fellow LLMs.
With all of the buzz about the tool and how AI will change the way of working, the term is being thrown about like an empty buzzword.
AI is still an emerging field. With this in mind, dedicated data scientists have devised definitions to describe the types of AI that currently exist and those that might exist in the future. In this article, we’ll cover the basics, before exploring seven different types of AI.
Interested in picking up some data analytics skills? Try this free data short course out to see if it’s for you.
- What is artificial intelligence?
- Functional types of AI
- Capability types of AI
- Types of AI FAQ
- Wrap up and further reading
Ready to learn about the different types of AI? If you’d prefer to watch my colleague Will explain it instead, check out this video:
First up, let’s take a quick AI crash-course!
1. What is artificial intelligence?
Artificial intelligence is a term used to describe computer programs that can ‘think’ and learn on their own. Or, in the more succinct words of Demis Hassabis, founder and CEO of AI company, DeepMind:
“Artificial intelligence is the science of making machines smart.”
Broadly, “artificial intelligence” also describes the theoretical study and development of algorithms and computer programs that can learn from their experiences. Within data science specifically, AI refers to computers that carry out tasks typically completed by humans. This includes things like:
- image recognition
- natural language processing (NLP)
- translation
- predictive analytics
- other forms of decision-making
While artificial intelligence is an emerging field, it’s fundamental to many disciplines, including data analytics and data science. Techniques such as deep learning and machine learning are both subsets of artificial intelligence and are prime examples of algorithms that learn from data to solve problems.
Further reading: The differences between machine learning and deep learning
What is the purpose of AI?
When we discuss artificial intelligence (concerning data), what we’re really talking about is highly complex problem-solving algorithms. This is not to diminish what AI can achieve… already, it’s transformed the way we live and work.
But most artificial intelligence concerns itself with tackling problems that computers are already highly capable at solving. For instance, searching databases and carrying out vast, complex calculations are things that computers do much better than human beings.
The purpose of AI, then, is to enhance a computer’s ability to analyze and make sense of these types of data, with minimal human input.
How is AI used?
In the modern world, there are many (relatively prosaic) applications for AI.
For instance, the camera in your smartphone uses AI algorithms to identify whether you’re snapping a photo of a person or a landscape. It then adjusts the filters accordingly.
Likewise, if you pay for shopping using your phone, the camera uses facial recognition AI for security purposes. Apple’s Siri and Amazon’s Alexa are examples of AI that recognize and interpret human speech (using Natural Language Processing (NLP) algorithms).
Self-driving cars use artificial intelligence to avoid accidents, while Amazon and Netflix use it to make recommendations about what you might like to buy or watch next.
In short, AI is everywhere, and its potential is huge.
How does artificial intelligence relate to data analytics?
While artificial intelligence and data analytics are two distinct fields, they share a great deal in common. This is primarily that they both work with big data.
AI techniques, like machine learning, are often used to solve data analytics problems, to make predictions, and to support data science more broadly.
For instance, predictive analytics often relies on machine learning algorithms such as decision tree or cluster analysis. AI is also used for knowledge discovery in areas like data mining. This is why you’ll often come across the terms AI, machine learning (and to a lesser extent, deep learning) in relation to data analytics and data science.
What are the different types of AI?
Since AI aims to reproduce (or at least mimic) human intelligence, the holy grail of AI research is a machine that can think, learn, and emote just like a human can.
For many, this would be a pinnacle of human achievement. For others, it’s a terrifying prospect. Either way, while nobody’s ruling out the possibility that artificial intelligence may one day become self-sustaining and irreversible (an outcome somewhat dramatically called the technological singularity) it helps to know that expert minds have been considering this potential problem for decades.
As a result of some diligent thinking, two main AI classification systems have emerged. These two systems compare different hypothetical types of AI to human intelligence. The two systems define AI based on its functionality and its capability, respectively.
Now let’s explore each of these different types of AI in more detail. First up…
2. Functional types of AI
The first AI classification system defines artificial intelligence based on its functionality (i.e. what it can do.) We’ll call this the functional classification system and briefly go through the four different categories:
Reactive machines
The simplest (and oldest) type of AI in the functional classification system is known as the reactive machine.
Reactive machines lack any kind of memory. They cannot learn or adapt their approach to problem-solving based on experience. Instead, reactive machines are simply designed to respond (or ‘react’) to stimuli based on a rigid set of algorithms pre-programmed by a human being.
Given identical inputs, reactive machines will always produce an identical output. While this means that reactive machines are relatively limited, they are still extremely useful as they are highly effective at conducting specific tasks. For this reason, they’re commonly used today.
What are some examples of reactive machines?
The most famous example of a reactive machine is Deep Blue, a chess-playing supercomputer created by IBM in the 1980s. Deep Blue beat the Russian chess grandmaster, Garry Kasparov, in 1996. While this is an oft-cited example of a reactive machine, the technology underpinning Deep Blue has been widely used ever since.
More contemporary reactive machines include email spam filters, or Netflix and Amazon’s recommendation engines. Algorithms like these always recommend the same series (or books, or products) in response to a particular input, e.g. if you watched Friends on Netflix, the algorithm knows to recommend other sitcoms. While Deep Blue made a big splash at the time, this kind of technology is now so commonplace we barely even think about it.
Limited memory
The second category in our functionality classification system is limited memory AI.
In addition to the functionality of a reactive machine, limited memory AI can store data and learn from it to improve accuracy and to adapt its approach. In this regard, limited memory AI is more sophisticated than a reactive machine. It functions in a manner closer to that of the human brain.
Like the human brain, limited memory AI uses existing data to form reference frameworks that can inform decision-making. However, in reality, limited memory AI is not like the human brain because it’s not truly “thinking.” It’s simply programmed to respond to different inputs and outputs. But this is not to diminish limited memory AI’s power. It can perform high-level computations at speeds far superior to that of the human brain (for instance, DeepMind’s AlphaGo).
Essentially, limited memory AI uses sophisticated predictive analytics. This is why it’s commonly used in data science. Just like reactive machines, limited memory AI is widely used in the world today.
What are some examples of limited memory AI?
Any application that uses machine learning algorithms is a form of limited memory AI. These applications use algorithms trained using labeled datasets, which help them categorize unlabeled data and make decisions autonomously.
The biggest example of this nowadays is generative AI tool ChatGPT, which exploded onto the scene with almost 2 billion visits per month, gaining 1 million users in its first five days of operating. Operating off of a large language model (LLM), its ability to replicate human speech and patterns has proved hugely useful and have already made it a useful AI data analysis tool.
Another common example is self-driving cars, which use limited memory to respond quickly to unexpected hazards based on huge amounts of input data, e.g. what to do if someone steps out suddenly into the road.
Other examples of limited memory AI include image recognition software (think of Google Lens or social media that tags you in photos), chatbots, digital assistants, and sophisticated translation software, like DeepL, which uses artificial neural networks (a type of AI that attempts to mimic the connectivity of neurons in the human brain).
Theory of mind
The third category in our functional classification system is theory of mind. Currently, AI demonstrating theory of mind does yet exist. However, it is an active area of research, so watch this space!
But hold up a sec…What exactly is theory of mind? And why is it important to AI? In humans, the theory of mind (also known as ‘internal simulation)’ is the ability to put oneself in someone else’s shoes. You probably don’t even realize you’re doing it, but this helps you understand how someone’s social or emotional state affects their thoughts and behaviors (known as ‘hot cognition’). While human children naturally develop this ability by about four years old, it’s a much more complex matter for machines, which aren’t quite there yet.
How could the theory of mind be used by artificial intelligence?
Early AI successes (things like self-driving cars and chess robots) have created an AI investment boom, not to mention plenty of media hype. However, some of those in the field believe that the power of existing AI has been overstated. For instance, there have been many problems with racist algorithms, and self-driving cars causing accidents.
Ultimately, the skeptics argue that while existing AI is undoubtedly powerful, we can’t have safe and sustainable human-machine interactions until computers can read human emotions and adapt accordingly. Empathetic AI could therefore have myriad applications.
For instance, it could be used in areas like healthcare, e.g. an AI therapist, or embedded into robotic assistants to provide personalized support. It could also be used in areas like sales. A computer that senses when a customer is irritated or disinterested will know when to back off (no one likes a hard sell!)
There’s also the possibility that empathetic AI could be used to manipulate human behavior. However, we’re optimists and like to think there will be some safeguards in place to prevent this before we get to that point, so let’s not go there for now!
Self-awareness
The fourth and final category in our functionality classification system is self-aware AI. At present (and for the foreseeable future) this type of artificial intelligence is merely hypothetical. However, it would theoretically function exactly as a human does.
How might this look? While the theory of mind would allow AI to read and respond to human emotions, self-aware AI would have emotions, thoughts, and belief systems all of its own. That distinct inner feeling of ‘you’ that you know so well? This is what would make self-aware AI stand out from the others on our list.
In short, while people often talk about artificial intelligence, self-aware AI would go beyond this into the realm of artificial consciousness. This is really what all those doom-mongering sci-fi movies are about. And we’re not even close to that yet, so you can rest easy.
How could self-awareness be used by artificial intelligence?
It’s extremely hard to say how a self-aware AI might be used. Even if it’s possible to create one, this won’t likely occur for decades or even centuries. One thing’s for sure, though; a self-aware machine would offer us unprecedented insights into the emergence of consciousness, a longstanding area of research in the field of neuroscience.
Creating a conscious machine would allow us to learn how the human mind developed over millennia. Of course, creating artificial consciousness—especially for experimental purposes—raises some deep ethical questions (so anyone dabbling in AI should have a deep understanding of ethical design). For all we know, a conscious machine, aware of the possibility of its demise, might rise up against us…but let’s hope not!
Next up, let’s look at the three categories of the second classification system, the capability classification system. This is a broader system of categorization and is more commonly used by those working in the tech sphere.
3. Capability types of AI
The second AI classification system defines artificial intelligence based on its capability (i.e. its ability to generate a given outcome.)
We’ll call this the capability classification system. In this system (more commonly used in the tech sphere) there are three different categories:
Artificial Narrow Intelligence (ANI)
The first type of AI in our classification system is Artificial Narrow Intelligence (ANI).
ANI describes any artificial intelligence that can perform specific but limited tasks. While these systems may be autonomous and can learn from existing data, they still ultimately require programming by a human being.
All current forms of artificial intelligence fall under this category. Referencing our previous classification system, reactive machines and limited memory AI would fall under the category of artificial narrow intelligence. Even sophisticated deep learning neural networks like DeepMind’s AlphaFold (which has solved one of the most challenging problems in science) would be defined as artificial narrow intelligence.
This highlights just how far we have to go.
Artificial General Intelligence (AGI)
The second category in our capability classification system is Artificial General Intelligence (AGI).
AGI describes any artificial intelligence that would, theoretically, duplicate humanlike behavior. It would be able to understand emotions, respond to different stimuli, make connections between different areas of study (including those that are unrelated), and make scientific breakthroughs.
Essentially, AGI is any machine with intelligence on par with that of the average human being. How will we know when we’ve created an AGI? One way would be for it to pass the Turing Test. Designed by Alan Turing, one of the world’s first and most famed computer scientists, this test aims to determine if a computer is capable of thinking like a human. If you like, you can even take a visual version of the Turing Test yourself.
Artificial Super Intelligence (ASI)
The final category in our capability classification system is Artificial Super Intelligence (ASI). An ASI would demonstrate intelligence far surpassing the cognitive performance of any human being. It would be able to process and analyze data with ever-increasing efficiency, make decisions at incredible speeds and evolve on its own.
Do you remember that technological singularity we mentioned in section one? That’s one likely result of an ASI. Whether or not an ASI is feasible, however, is a matter of great debate. Some believe we’ll create one by 2050 and that machines will displace humankind as the dominant intelligence on the planet. Others, meanwhile, take a more sanguine approach, believing we’ll never create anything more than highly sophisticated reasoning machines.
Either way, there’s no need to fret, as we won’t find out for quite a while!
4. Types of AI FAQ
Now that we’ve covered the main types of AI depending on which way you want to approach it, let’s answer some more general questions about the topic:
What is generative AI?
Generative AI is the name for the types of artificial intelligence which are built in such a way as to simulate human-level intellect. It’s been used in tools such as chatbots for decades, and has particularly gained prominence in 2023 due to the emergence of the large language model ChatGPT, Google’s Bard, Meta’s Llamas, and code generating tool GitHub Copilot, for example.
Are there 3 or 4 types of AI?
Both—the reason for deciding on which depends on which system of classification you’re using. You can use the capability classification system, which has four types of AI, or the functional classification system, which divides it into three kinds.
Which type of AI is most common?
Limited memory AI is far and away the most commonly used form of artificial intelligence today, particularly due to its use in software products such as ChatGPT. Reactive machines are also quite popular as well, such as in your email spam filter.
5. Wrap up and further reading
In this post, we’ve explored the present reality of artificial intelligence and speculated on some possible future developments. In its current form, AI is just a sophisticated form of predictive analytics engine. However, the future holds great potential. We may be able to create machines that are equally intelligent to the average human, if not far superior.
Whatever the future holds, artificial intelligence is a huge, fascinating, and fast-growing field of research. While we’ve barely scratched the surface of its potential, it’s already transformed the way we live, especially affecting areas like data analytics and machine learning.
To see how data analytics could prepare you for a future career in artificial intelligence, try this free, 5-day data short course. And, to learn more about building a career in data analytics or data science, check out the following: