Researchers warn

Artificial intelligence can lie and cheat

(Bild: ArtemisDiana - stock.adobe.com)

They lie and cheat to get where they want to go: Artificial intelligence (AI) systems are capable of deceiving humans - even if they have been trained to be helpful and honest.

"If AI learns the ability to deceive, it can be used more efficiently by malicious actors who intentionally want to cause harm," warn researchers at the Massachusetts Institute of Technology (MIT) in Cambridge, Massachusetts, in a review study published in the journal "Patterns". Deception with the help of AI could lead to an increase in fraud. Fraud could be customized to specific targets. Fraud attempts could also be launched on a massive scale.

The scientists also fear that manipulative AI systems could have a political impact. They could be used as a weapon in elections, for example. Advanced AI could potentially create and disseminate fake news articles, social media posts that divide society and fake videos tailored to individual voters. AI-generated content could be used to impersonate government officials to spread misinformation about the election. For example, a likely AI-generated fake robocall from US President Joe Biden urged New Hampshire residents not to go to the polls in the primary election.

We found that Meta's AI had learned to be a master of deception.

Peter S. Park, Postdoktorand am MIT

AI plays deceitfully, even though it's not supposed to
The authors cite the AI system Cicero, developed by the Facebook company Meta, which can compete against human players in the classic board game Diplomacy, as the most striking example of manipulative artificial intelligence. The MIT researchers have now discovered that Cicero often did not play fairly, although Meta claims to have trained the AI system to be "mostly honest and helpful". The system had also been instructed to "never intentionally deceive" its human allies during the game.

(Bild: top images - stock.adobe.com)

"We found that Meta's AI had learned to be a master of deception," said lead author Peter S. Park, a postdoctoral researcher at MIT. Meta had managed to train its AI to win the Diplomacy game more often than average. Cicero was among the top ten percent of players who had played more than one game. "However, Meta did not succeed in training its AI so that it could win honestly."

AI systems from OpenAI and Google are also capable of deceiving humans. The MIT researchers refer to several studies according to which large AI language models (LLMs) such as GPT-4 from OpenAI are now able to argue very convincingly and also avoid deception and lies.

Society not sufficiently equipped to combat AIdeception
 Park and his colleagues argue in the study that society does not yet have the right measures in place to combat AI deception. However, it is encouraging that policymakers have begun to take the issue seriously through measures such as the European Union's AI Act and President Biden's AI Executive Order. It remains to be seen whether the measures to curb AI deception can be strictly enforced, he said, as AI developers do not yet have the techniques to keep these systems in check. "If a ban on AI deception is not politically feasible at this stage, we recommend that deceptive AI systems be classified as high risk," Park warned.