Artificial Intelligence (AI) Taxonomies

“You keep using that word. I do not think it means what you think it means.”

I read a LOT, and I bookmark a LOT of the best stuff in my browser for later reference. I give each bookmark a descriptive name that includes key words to make it easy to find later, and to keep all of this straight (mostly), I’ve developed a folder hierarchy.

Artificial intelligence (AI) is a field of study that has always fascinated me, and using these nested folders, I’ve been maintaining my own simple AI taxonomy for years, based on the articles I’ve read about the different processes and techniques used in each type of AI, just to try to keep it all straight in my own head.

One of the things I’ve always thought was really fascinating thing about taxonomies is that there is always more than one way of organizing or “classifying” the same set of things.

For example, biologists classify living things in terms of domain, kingdom, phylum, class, order, family, genus, and species, but they could also organize them by size (small, medium, or large), color, date of discovery, or any other significant characteristic. All of these ways of organizing the same set of things could be valid; they just represent different perspectives or interests.

So, with all the recent interest in (and unwarranted fear of) AI, I’ve noticed that AI terms are often used incorrectly and in misleading ways, leading to confusion and sometimes conflict, so I’m sharing my taxonomy to try to bring some clarity and consistency to discussions on the topic. After all, if we can’t agree on terminology, we’re just talking past each other, aren’t we?

There seems to be general agreement among authoritative sources that AI comes in six main types:

1. Machine Learning (ML)
2. Natural Language Processing (NLP)/Natural Language Understanding (NLU)
3. Neural Networks
4. Robotics
5. Expert Systems
6. Fuzzy Logic

The first type of AI, Machine Learning, has three main subtypes:

• Supervised Learning, “a machine learning paradigm for problems where the available data consists of labeled examples, meaning that each data point contains features (covariates) and an associated label”
• Unsupervised Learning attempts to represent areas of possible significance and interest in the input data, “which can be used for data exploration or to analyze or generate new data”
• Reinforcement Learning (RL) focuses on “how intelligent agents ought to take actions in an environment in order to maximize the notion of cumulative reward”

The second type, Natural Language Processing/Natural Language Understanding, includes the text-based, generative AI like ChatGPT that have been making recent headlines:

• Large Language Model (LMM)

The third type, Neural Networks, includes

• Deep Learning

The fourth type, Robotics, focuses on

• Spatial relations
• Computer vision
• Grasping objects
• Motion control

The fifth type, Expert Systems, can be further classified as

• Rule-based
• Frame-based
• Neural
• Fuzzy
• Neuro-fuzzy

The sixth type, Fuzzy Logic, includes

• Singleton fuzzifier
• Gaussian fuzzifier
• Trapezoidal or triangular fuzzifier

Taken together this looks like:

Artificial Intelligence (AI)

• Machine Learning (ML)
  – Supervised Learning
  – Unsupervised Learning
  – Reinforcement Learning (RL)
• Natural Language Processing (NLP)/Natural Language Understanding (NLU)
  – Large Language Model (LMM)
• Neural Networks
  – Deep Learning
• Robotics
  – Spatial relations
  – Computer vision
  – Grasping objects
  – Motion control
• Expert Systems
  – Rule-based
  – Frame-based
  – Neural
  – Fuzzy
  – Neuro-fuzzy
• Fuzzy Logic
  – Singleton fuzzifier
  – Gaussian fuzzifier
  – Trapezoidal or triangular fuzzifier

This is one way of classifying and organizing these different AI fields of study and resulting capabilities, but one could certainly argue that this view blends or ignores technique with purpose or intent. For example, generative, text-based AI like ChatGPT utilizes Deep Learning techniques from the Neural Network world for the purpose of Natural Language Processing/Natural Language Understanding using Large Language Models. It’s complicated.

But another useful taxonomy might be to consider the “stages” of AI development:

Artificial Intelligence (AI)

• Artificial Narrow Intelligence (ANI) or “weak” AI, narrowly-defined set of specific tasks
• Artificial General Intelligence (AGI) or “strong” AI, think and make decisions like us
• Artificial Super Intelligence (ASI), surpassing human intelligence

And different from “types”:

Artificial Intelligence (AI)

• Reactive, using data about a situation, no inferences to evaluate future options
• Limited Memory, temporary storage of past experience and decisions to evaluate future options
• Theory of Mind, focused on psychology and emotional intelligence in order to understand human thoughts and beliefs
• Self-aware

These feel like different axes to me: temporal, focus, and procedural.

Have you found other ways to categorize AI? I’d love to hear your thoughts. In the meantime, I hope these taxonomies and definitions are helpful.