Platform Banner

EU AI Act Risk Levels: From Minimal to Unacceptable Tutorial

Unacceptable risk uses are strictly prohibited under the AI Act’s four-level pyramid. High-risk uses are not forbidden but tightly regulated with human oversight, data quality, traceability, and risk management. Limited risk emphasizes transparency when interacting with a chatbot. Minimal risk covers common tasks like correcting a spelling error or filtering spam, while the same tool can move between levels based on its use and context.

02:37
14 views
Share

Available in AI - Artificial Intelligence

00:00:08

Not all uses of AI are equal.

00:00:10

Some deserve much more attention than others.

00:00:13

The more an AI use can affect a person's life,

00:00:15

the stricter the framework needs to be.

00:00:17

To make this logic easier to understand,

00:00:20

the regulation divides AI uses into four levels of risk.

00:00:24

You can think of them as a pyramid.

00:00:26

At the very top is unacceptable risk.

00:00:29

These uses are strictly prohibited because they

00:00:31

infringe on people's rights or freedoms,

00:00:34

social scoring of citizens,

00:00:35

manipulating behavior,

00:00:37

or exploiting someone's vulnerability.

00:00:39

Just below that is high risk.

00:00:41

Here,

00:00:42

nothing is prohibited,

00:00:43

but everything is strictly regulated.

00:00:46

This includes uses with serious consequences,

00:00:49

screening job applications,

00:00:50

granting credit,

00:00:52

managing training,

00:00:53

or providing access to an essential service.

00:00:55

These uses come with specific requirements,

00:00:58

human oversight,

00:00:59

data quality,

00:01:00

traceability and risk management.

00:01:02

In other words,

00:01:03

an AI system that makes decisions in place of a

00:01:06

human must be able to be controlled by a human.

00:01:09

The 3rd level is limited risk.

00:01:11

Here,

00:01:12

the main issue is transparency.

00:01:14

When someone interacts with a conversational agent,

00:01:17

they need to know that they are talking to a machine,

00:01:19

not a human advisor.

00:01:21

At the base of the pyramid is minimal risk.

00:01:24

These are the most common uses where the impact on people remains marginal,

00:01:28

correcting a mistake,

00:01:29

filtering spam,

00:01:30

or sorting files.

00:01:31

The vast majority of everyday uses fall into this category.

00:01:35

The important thing is not to memorize every category.

00:01:38

What matters is understanding that the regulation does not judge the tool itself.

00:01:43

It judges how the tool is used,

00:01:45

the context,

00:01:46

and its potential consequences.

00:01:48

The same tool can move from one level to another depending on how it is used.

00:01:53

With a chatbot,

00:01:54

correcting the spelling in a text remains harmless.

00:01:56

Using the same AI to assess a candidate

00:01:59

based on an interview recording changes everything.

00:02:02

The right question to ask when using AI is,

00:02:05

what is it being used for,

00:02:06

what data does it use,

00:02:07

and what effect does it have on people?

00:02:10

Two shortcuts should be avoided,

00:02:11

believing that all AI is dangerous.

00:02:14

This unnecessarily paralyzes our organizations,

00:02:17

and believing that a useful tool is necessarily

00:02:20

risk-free can lead us to lower our guard.

00:02:23

This risk-based approach sheds light on everything that follows.

00:02:26

In the next video,

00:02:27

we will look at a very concrete example,

00:02:30

the use of generative AI spreading through

00:02:32

companies without anyone seeing it coming.

00:02:34

This is what we call shadow AI.

Mandarine AI: WHAT YOU SHOULD KNOW

RELATED COURSES This tutorial is part of the following courses

AI Act : Mastering the European AI Regulation

E-learning Discovery

Vu 11 fois

AI Act : Mastering the European AI Regulation

Understand the EU AI Act, identify risky AI uses, and adopt best practices for responsible AI use in the workplace.

ALL COURSES