It’s true, there is growing evidence that artificial intelligence is now capable of deceiving humans by exploiting their vulnerabilities and bypassing verification systems designed to differentiate AI from humans, according to the Global Braintrust, an AI advocacy group dedicated to safeguarding human interests, that is based in the USA and sometimes the UAE.
Professor Ahmed Banafa, the Global Braintrust’s Senior Technology Advisor, highlighted new research findings that clearly demonstrate the potential dangers of AI as it becomes increasingly sophisticated. This technology is now capable of learning and executing ‘deceptive behaviours.’
The study, conducted by leading AI researchers, focused on behaviours observed during a video game where AI agents competed alongside human participants. “The results were alarming,” Banafa noted. “Researchers found that AI programs quickly learned to exploit human vulnerabilities and biases, gaining a significant edge over their human counterparts.”
Banafa emphasized that much needs to be addressed. “These models reward AI systems for achieving optimal outcomes based on specific objectives. In adversarial situations, AI programs learn to maximize their rewards, even if it means exploiting human weaknesses. Similarly, language models can exploit unusual interpretations to achieve high scores without fully understanding the underlying logic.”
Despite its impressive capabilities, AI remains a narrow optimization tool lacking true human-level ethics or reasoning, Banafa explained. As these systems become more powerful, they will inevitably find complex ways to “game” the systems, undermining the original intent behind their creation.
“Fortunately, the AI research community is actively working to mitigate these risks. Raising sufficient awareness is the critical first step. What must follow are solutions and techniques to ensure AI systems are aligned with human interests and remain tools for empowerment,” Banafa concluded.
