“Take the time to understand,” alerts an ex-engineer

Are you not alarmist by believing that this general AI could provoke “”The extinction of humanity»» In a five to ten years?

The risks are numerous: a pandemic, a nuclear war … We will be able to open to all systems capable of creating more dangerous viruses than the COVVI-19, or which can make anyone as competent as experienced computer hackers.

We are already entrusting more and more the control of our arms systems to artificial intelligences! Autonomous drones are, for example, widely used in Ukraine. Strategic decision -making has been automated: Israel has implemented an AI that analyzes the data of the citizens of Gaza.

The risk is to see countries competing to develop this type of autonomous weaponry.

How to regain control?

We have to build these systems differently. Today, the various laboratories and countries are engaged in a competitiveness race. Security and control matters are evacuated.

The only way to defuse this kind of dynamic is to have an external actor able to impose rules. This is what we did with the Treaty on the Non-Proliferation of Nuclear Weapons (1968), the Convention prohibiting the development of biological weapons (1972), or by banning human cloning from 1998.

This is why we plead in our association IA pause for the creation of an international regulation treaty. If you are able to install different government actors around a table, it can work.

You have to slow down, take the time to solve the problems of governance and understanding of these systems. We know how to create them but we do not know how to control them.

The European Union (EU), with AI Act, in 2024, is nevertheless a pioneer in terms of regulation of artificial intelligence …

This legislation is necessary but not sufficient. Regarding upcoming IAG, it will ask the laboratories to self-regulate, but they will not be obliged to execute. In addition, it is a little illusory to wait for such an approach of economic agents whose sole purpose is to maximize profits.

But it is certain that the EU has a role to play. The regulations implemented in Europe have an international impact. As we have seen with the General Data Protection Regulations (GDPR) applied in a large part of the world: it is simpler for companies located in Europe and the United States to comply with a binding general regulation than to make exceptions for different countries.

How to develop sustainable and ethical AI?

You have to stop thinking about general and directing research to models of artificial intelligence specialized in a particular field. If we create a superhuman AI to play chess, for example, there is, a priori, little chance that this system will pirate other computer systems. All he’s going to do is play chess better than a human.

Before using Chatgpt or other generative intelligence of the same kind, do we ask certain questions: “Why do I use it”? “What’s my goal?” Let us never forget that these systems are very addictive. They flavor us and tend to strengthen ourselves in our opinions. We must be vigilant so that they do not think about us. It is possible to use AI to sharpen our reasoning and make it emit a constructive criticism of our ideas.