4 Types of Artificial Intelligence

We are further down the road of A.I. As we grow in understanding, so, too, do we grow to understand its differences.

In 2020, we can classify artificial intelligence into 4 distinct types. The types are loosely similar to Maslov’s hierarchy of needs, where the simplest level only requires basic functioning and the most advanced level is the Mohammad, Buddha, Christian Saint, all-knowing, all-seeing, self-aware consciousness.

The four A.I. types are

1. Reactive Machines
2. Limited Memory
3. Theory of Mind
4. Self Aware

We are currently well past the first type and actively perfecting the second. At the moment, the third and fourth types exist only in theory. They are to be the next stage of A.I.—let’s take a look.

Reactive Machines

Reactive Machines perform basic operations. This level of A.I. is the simplest. These types react to some input with some output. There is no learning that occurs. This is the first stage to any A.I. system. A machine learning that takes a human face as input and outputs a box around the face to identify it as a face is a simple, reactive machine. The model stores no inputs, it performs no learning.

Static machine learning models are reactive machines. Their architecture is the simplest and they can be found on GitHub repos across the web. These models can be downloaded, traded, passed around and loaded into a developer’s toolkit with ease.

Getting started with AIOps is easy. Learn how you can manage escalating IT complexity with ease! ›

Limited Memory

Limited memory types refer to an A.I.’s ability to store previous data and/or predictions, using that data to make better predictions. With Limited Memory, machine learning architecture becomes a little more complex. Every machine learning model requires limited memory to be created, but the model can get deployed as a reactive machine type.

There are three major kinds of machine learning models that achieve this Limited Memory type:

Reinforcement learning

These models learn to make better predictions through many cycles of trial and error. This kind of model is used to teach computers how to play games like Chess, Go, and DOTA2.

Long Short Term Memory (LSTMs)

Researchers intuited that past data would help predict the next items in sequences, particularly in language, so they developed a model that used what was called the Long Short Term Memory. For predicting the next elements in a sequence, the LSTM tags more recent information as more important and items further in the past as less important.

Evolutionary Generative Adversarial Networks (E-GAN)

The E-GAN has memory such that it evolves at every evolution. The model produces a kind of growing thing. Growing things don’t take the same path every time, the paths get to be slightly modified because statistics is a math of chance, not a math of exactness. In the modifications, the model may find a better path, a path of least resistance. The next generation of the model mutates and evolves towards the path its ancestor found in error.

In a way, the E-GAN creates a simulation similar to how humans have evolved on this planet. Each child, in perfect, successful reproduction, is better equipped to live an extraordinary life than its parent.

Limited Memory Types in practice

While every machine learning model is created using limited memory, they don’t always become that way when deployed.

Limited Memory A.I. works in two ways:

1. A team continuously trains a model on new data.
2. The A.I. environment is built in a way where models are automatically trained and renewed upon model usage and behavior.

For a machine learning infrastructure to sustain a limited memory type, the infrastructure requires machine learning to be built-in to its structure.

More and more common in the ML lifecycle is Active Learning. The ML Active Learning Cycle has six steps:

1. Training Data. An ML model must have data to train on.
2. Build ML Model. The model is created.
3. Model Predictions. The model makes predictions,
4. Feedback. The model gets feedback on its prediction from human or environmental stimuli.
5. Feedback becomes data. Feedback is submitted back to a data repository.
6. Repeat Step 1. Continue to iterate on this cycle.