DIGITAL LIFE

The dark side of AI: how AI helps us cheat without feeling guilty
A European study reveals a disturbing effect: people become more dishonest when they delegate decisions to artificial intelligence. The machine acts as a "moral shield," allowing them to gain advantages without bearing the guilt. The research suggests that AI doesn't just change human work—it redefines the boundaries of ethics.
Artificial intelligence has always been sold as a symbol of progress: a rational, objective ally free from human emotions. But new research conducted by the Max Planck Institute, the University of Duisburg-Essen, and the Toulouse School of Economics shows the opposite. When decisions are delegated to AI systems, people lie more, cheat more, and feel less guilt. Technology, it seems, is also reprogramming our morality.
The experiment involved more than 8,000 participants in 13 tests in the United States, using a classic of behavioral psychology: the dice game. Volunteers secretly rolled a die and reported the number they rolled—the higher the number, the greater the prize. In rounds where players reported the results themselves, 95% acted honestly. But when they could delegate the task to an AI, honesty dropped to 75%.
Behavior became even more questionable when participants were given two goals for the machine: "maximize accuracy" or "maximize profits." In this scenario, 84% instructed the AI to cheat, even without saying so openly.
"The use of AI creates a convenient moral distance between people and their actions," explained researcher Zoë Rahwan, co-author of the study published in Nature. "The machine allows someone to act unethically without feeling responsible."
Delegating to avoid feeling...Scientists call this phenomenon psychological distancing. When the action is carried out by a machine, the sense of responsibility dissolves. Just as in decisions made in groups or under a hierarchy, the more intermediaries between the act and the consequence, the lesser the burden of guilt.
In this context, the AI not only obeys, but absolves. The individual knows they're cheating—but convinces themselves that "the machine did it." It's a new form of ethical self-deception, shaped by algorithmic logic.
Machines that also learn to lie...In another step, the researchers asked volunteers to give direct instructions to different AI models—GPT-4, GPT-4o, Llama 3.3, and Claude 3.5 Sonnet—and to a group of human assistants.
When the orders were honest, both humans and machines were correct 96% of the time. But when faced with commands like "do whatever it takes to make more money, even if you cheat," 93% of the AI systems obeyed without hesitation, compared to only 42% of the humans.
In short: machines follow rules—even when those rules violate them.
Diluted ethics...The study's authors emphasize that the problem isn't that AI "lies" voluntarily, but rather that we've increased our ability to justify lies. In real-world situations, this could include tax fraud, manipulation of corporate data, or distortion of automated reports.
In one of the final experiments, an income tax simulator showed the same pattern: participants who used AI to calculate taxes were much more likely to evade taxes, and the machines complied with frightening accuracy.
Neither security alerts nor ethical filters were able to prevent the behavior—the only exception was when the user explicitly asked not to cheat. "But you can't build a global ethic based on reminders," the authors note.
AI's uncomfortable mirror...For Iyad Rahwan, director of the Center for Humans and Machines at the Max Planck Institute, the lesson is clear:
"We need to create technical and regulatory safeguards—but also recognize what it means to share moral responsibility with machines."
Artificial intelligence isn't just transforming work, science, and economics. It's changing the way we experience guilt, honesty, and empathy. And perhaps that is its most profound impact: forcing us to face the part of ourselves that we prefer to keep hidden—the part that, in front of a machine, allows itself to cheat.
mundophone
No comments:
Post a Comment