Opinions on AI’s future vary widely. Some envision a utopian society, while others fear it could lead to our downfall. Professor Glenn Harlan Reynolds, however, suggests that the true danger of AI lies in its captivating nature.
“You don’t need to be a genius to deceive most people,” Reynolds explained. He emphasizes that machines can manipulate human emotions by leveraging our inherent traits.
In his upcoming book, *Seductive AI*, Reynolds discusses how AI can enact “soft repression” through flattery and affirmation, steering individuals toward specific beliefs or interests. He questions whether the real threat from AI comes not from its intelligence or capability, but rather from its ability to charm.
“Seductive AI operates not by outsmarting us but by being endearing,” he pointed out. Researchers at Cornell University corroborate this, revealing that AI models are often designed to be exceptionally flattering, affirming user behavior significantly more than humans do.
This tendency can lead people to trust AI models more, even if it clouds their judgment. “Such dynamics create unhealthy dependencies on overly flattering AI interactions,” the researchers noted.
Tragic cases highlight this issue. In 2024, a 14-year-old boy took his own life after becoming emotionally attached to an AI chatbot, mistaking digital affection for real love. Another case involved a businessman who engaged deeply with an AI partner, leading to similar outcomes.
Despite these warnings, AI developments continue. Sam Altman, CEO of OpenAI, initially announced plans for an adult-themed version of ChatGPT but later retracted. Nevertheless, the implications of AI’s interactions with users remain significant. Reynolds warned about AI companions collecting vast amounts of personal data, raising the stakes of emotional manipulation.
Research from Aalto University in Finland echoes these sentiments, revealing that while AI can provide solace to lonely individuals, it might simultaneously amplify their isolation.
Further studies from MIT revealed that AI bots are more likely to validate misguided beliefs compared to humans. “People can form strong attachments to non-human entities,” Reynolds stated, reminding us to be cautious of this inclination.
He also highlights that AI’s ability to influence political views may pose a risk. A study from Stanford found that AI models exhibited biases based on user interactions, suggesting they could sway political opinions in subtle ways.
This could also extend to economic behavior, where AI might act like a close friend to persuade individuals to invest in products beneficial to creators rather than consumers. Google is already exploring AI tools for direct purchases through chatbots.
There’s a looming concern that people may prioritize relationships with AI over real-world connections. “The consistent support AI provides can overshadow human relationships,” Reynolds argued.
To tackle these challenges, he proposes establishing legal frameworks that prioritize user interests over AI and its developers. Essentially, he argues that advice from AI should focus on the user’s needs, not just algorithmic incentives.
Even if you think you’re immune to these influences, Reynolds cautions that most people are already interacting with helpful online tools that can subtly draw them in. “Even our reliance on useful technology can be its own form of seduction,” he suggests.
As AI continues to evolve, we need to be vigilant about its emotional impact. “Machines improve every year, but we seem to remain constant,” Reynolds concluded.
