What is actually “AI” and how is it related to other fancy stuff like “machine learning” and “deep learning”? Let’s face the beast!

AI means “Artificial Intelligence”, so far so good. But what distinguishes it from e.g. the intelligence of the blue wolf in Commander Keen 1?

“intelligent” blue wolf of 1990’s Commander Keen 1: It follows Keen across obstacles in the artificial world. Don’t get caught!

Or reformulated for those readers who were not into video games in the 80s’ and 90s’: What is actually the difference between AI and any other software programs or electronic circuits, e.g. of an automatic door?

“Intelligent” door of TCDC’s entry in Bangkok: It opens when you approach the entry

The answer is: NOTHING!

According to Wikipedia, Artificial intelligence (AI) is intelligence exhibited by machines, particularly computer systems. So they translate AI to “machine intelligence”. And translated for you: Software or electronics, it’s all AI, the blue wolf and the door opener included.

This also means that the fancy buzz word “AI” is actually outdated and almost meaningless. Sorry for all who use it so often nowadays (me and my blog posts included).

The best definition of AI (computers) I have ever read is a quote of Pedro Domingos:

“People worry that computers will get too smart and take over the world, but the real problem is that they’re too stupid and they’ve already taken over the world.”

(you can actually stop reading here)

OK, back to serious business, we wanted to learn something about AI, and how it is related to other terms like “machine learning” or “deep learning”.

A nice overview is illustrated in Goodfellow’s et.al. book Deep Learning:

Venn diagram showing the relationship between AI, machine learning and deep learning; (Deep Learning, Ian J. Goodfellow, Yoshua Bengio and Aaron Courville, MIT Press, 2016, p. 9; indication of neural networks is added)

According to Goodfellow, AI is the umbrella term containing more specific fields. Note that AI can also include simple knowledge bases. Hence, even Wikipedia might be considered as part of an AI technology.

When we refer nowadays to “AI”, we typically mean a software program that is obtained by machine learning, more in particular deep learning. All relevant AI technologies use deep learning, regardless whether it drives your car (computer vision) or it discusses with you (Chat-GPT). So, I recommend the terms “deep learning” or “machine learning” instead of “AI”, in order to pretend some more expertise in this field of technology.

Now we all know software programs since at least the early 90s. However, they have been implemented by conventional programming, i.e. manually coding explicit instructions and rules to solve problems. You can imagine a program for an automatic door containing the command:

If:

• the door is closed, and

• the sensor’s light barrier is blocked,

Then: Open the door!

In contrast, machine learning utilizes a data-driven approach: It trains a mathematical model having many tunable parameters (e.g. GPT-3: 175B, GPT-4: 1 Trillion) with a large set of data samples. In other words, instead of manually defining and coding any rules, the mathematical model learns itself from the data. Imagine you want to define all rules of a proper German Grammar. That’s practically not possible, unless you use a data-driven approach. this is also the reason why machine translations were not useable until DeepL and other companies started to train neural networks on the translation task.

You have certainly heard that AI uses neural networks as a “core”. But how does this core of an AI model (neural network) look like? Have a look to the next Article: AI Basics Part 2 !

author: Christoph Hewel
email: hewel@paustian.de

(photo: kudos to the cats of IPKat. But we look different, don’t we?)