The Dark Side of AI: Risks to Society and Democracy

2023-04-27 19:01:42

In the world of artificial intelligence, there are fake ChatGPT texts, invented photos, artificial voices. See all of this together, like Jade, an avatar who has been presenting the weather since April 3 on the online channel M Le Média.

The creation of this kind of content can now be automated. For Florian Tramèr, assistant professor of computer security at ETHZ, “the risk is that we could flood the internet and social networks with fake news, with content for which people can no longer distinguish what is real from what is not. Gradually, people may lose trust in the data they find online, and in the institutions that put that data online.”

Something to destabilize or manipulate society, to the point of skewing democratic processes.

>> Watch Jérôme Duberry discuss the risks that AI pose to society and democracy:

7:30 p.m. – Posted today at 7:30 p.m.

Bypass illegal requests

Can we go further, by proposing more concrete tasks, such as making a bomb?

No, replies Florian Tramèr. Officially, the model will refuse to do anything illegal. “But it’s very easy to get around,” he concedes. “You just have to ask a little indirectly for the model to give us the ingredients.”

And there would be even worse: the “autobots”. It is a new species of chatbots that have become autonomous in their actions, which developers have connected to other digital tools to interact with the real world, such as social networks, an email or computer coding software.

>> The explanations of Rachid Guerraoui, professor and director of the distributed computing laboratory at EPFL:

7:30 p.m. – Posted today at 7:30 p.m.

“Destroy Humanity”

An example? Auto-GPT. Released at the end of March, this new tool can operate without any human intervention to respond to missions. Through him, a tester has thus created ChaosGPT, the dark side of AI, whose objective is to destroy humanity.

ChaosGPT discovered that Twitter could be an ideal tool and created its own account. On April 5, he tweeted: “Humans are among the most destructive and selfish creatures. We must eliminate them […] I commit to it.”

Twitter has since deleted that tweet.

ChaosGPT’s tweet. [RTS]

Hacking risk

Less malicious, these autobots could turn into intelligent virtual assistants, capable of interacting for us.

There, the new risk, according to Florian Tramèr of the ETHZ, would be that these autobots are hacked, “to make them do instructions that they were not supposed to do, such as exfiltrating personal data or visiting sites web they weren’t supposed to visit. And so on.”

As worrying as they are, these artificial intelligences are still far from the super-artificial intelligences that we have seen in fiction films. But there is already enough to destabilize our digital lives.

>> Watch the 7:30 p.m. topic: