AI Terminology Explained

Know What You're Talking About

People (especially those outside of tech) have a hard time sticking to the right terminology, when it comes to anything related to artificial intelligence (AI). Terms like AI, machine learning, deep learning and LLMs are oftentimes used interchangeably. And while that is not really a big issue when it comes to business people and marketers, you definitely don’t want to belong to that group of people as an engineer. Unless you want to signal to everybody that you don’t know what you are talking about.

The goal of this blog post today is to clarify what each of the “buzzwords” means, so that you can always use the proper terminology.

Artificial Intelligence (AI)

Let’s get started with the most commonly used term – artificial intelligence (AI). Interestingly, this one is oftentimes used correctly. While one could think that people “overuse” it, since everything seems to be called AI nowadays, the definition of this term is actually quite broad.

According to Wikipedia it refers to: “…the capability of computational systems to perform tasks typically associated with human intelligence, such as learning, reasoning, problem-solving, perception, and decision-making” (Source)

Furthermore, the page lists (among others) search, logic and probabilistic methods as techniques of AI. Not just neural networks and GPT models.

In a nutshell, artificial intelligence can be considered to be the general broad term covering all techniques, which lead to computers showing intelligent behavior. It comprises search algorithms, linear regression, support vector machines (SVMs), basic neural networks and generative pre-trained transformer architectures (GPTs).

So yes, when you see a computer doing something intelligent, you can almost certainly call it AI. Practically speaking however, when people talk about AI, they talk about machine learning (ML).

Machine Learning (ML)

Machine learning (ML) is one sub-category of techniques for producing intelligent behavior in computers. It’s main characteristic is that the intelligence is the result of learning from data, rather than following hard-coded algorithms and formulas.

A classic example here would be teaching a computer to correctly classify images of cats and dogs. Manually implementing an algorithm that does that in a deterministic way doesn’t seem very feasible. You would have to hard-code every feature (e.g., pointy ears, whiskers etc.) and also consider all sorts of transformations (e.g., roations, varying contrasts etc.).

What we do in machine learning is what we do with humans in real life. When you teach your child to recognize different animals, do you explain to them every single feature that makes up a cat or a dog? Or do you just show them a bunch of examples while telling them “This is a cat.” and “This is a dog.”?

The latter is exactly what we do in machine learning. We prepare a dataset full of examples and then we choose an architecture or model (e.g., SVM, CNN etc.) and train it to generalize based on the dataset.

Summarized simply: Machine learning is when we achieve artificial intelligence by learning from data, instead of hard-coding the problem solving.

Deep Learning

Now, machine learning comprises every algorithm and model that learns from data. This includes linear regression, k-nearest neighbor classifiers, support vector machines, principal component analysis etc.

When we talk about AI or ML nowadays however, we are almost always talking about neural networks. These are specific architectures, inspired by the human brain, which learn by adjusting their parameters (weights and biases) using something called backpropagation.

Whenever you hear of some impressive achievement or progress in the field of AI, it’s almost certainly some time of neural network that is being talked about. AlphaGo beating humans at Go? A CNN-based architecture. LLMs solving complex problems in math and coding? Transformer-based language models. State-of-the art models generating photo-realistic images and videos? Likely diffusion models, also based on neural networks.

So put simply: Deep learning is machine learning done with neural networks.

Data Science

This is now the first term that is not just a sub-category of one of the other terms. Data science is a term that is even broader than artificial intelligence. It may make use of AI, machine learning etc. but in addition to all that it also requires domain-specific knowledge.

It is an interdisciplinary field where the goal is to extract insights from data, by using all sorts of tools (including AI). Another focus here is also visualizing and presenting the insights in an effective way.

Artificial General Intelligence (AGI)

Now we get to the real buzzwords. Some people seem to prematurely call every new LLM on the market AGI, when it outperforms all previous ones (some companies do it even without performance gains).

We can take a look at the Wikipedia definition once again, which states that AGI is: “…a type of artificial intelligence capable of performing the full spectrum of cognitively demanding tasks with proficiency comparable to, or surpassing, that of humans.” (Source)

I think this makes it pretty clear. Using that definition, not a single model that we know of right now qualifies as AGI. Why? Because of the phrase “full spectrum”. If there is any cognitively demanding task where humans are better than our current AIs, we don’t have AGI.

Besides that, even in fields like coding and mathematics, LLMs fail miserably when things get a bit more complex and when there is no documentation or tutorial on the internet.

To summarize simply: An AI can be considered an AGI when it outperforms humans at any cognitively demanding task.

Artificial Superintelligence (ASI)

Finally, we have artificial superintelligence (ASI). This would be an artificial intellect that vastly outperforms humans in the cognitive domain. If you are interested in reading about possible paths to and dangers of AGI, I can recommend the (quite technical) book Superintelligence by Nick Bostrom.

Summary

Let’s summarize each of the terms as one-liners:

  • Artificial Intelligence (AI): Computers exhibiting intelligent behavior.
  • Machine Learning (ML): AI that learns from data.
  • Deep Learning: Machine learning done with neural networks.
  • Data Science: Interdisciplinary field focusing on extracting insights from data.
  • Artificial General Intelligence (AGI): AI that outperforms humans at any cognitively demanding task.
  • Artificial Superintelligence (ASI): AI that vastly outperforms humans in the cognitive domain.