AI Also Lies (Short)

In 1941, Isaac Asimov explored the ethical complexities of artificial intelligence in his story “Liar!", presenting a telepathic robot, Herbie, whose well-intentioned lies revealed a deep ethical paradox. Today, in 2024, we face similar dilemmas, as advanced AI systems show capabilities that could easily be misread or misused. Recent research…

In 1941, Isaac Asimov explored the ethical complexities of artificial intelligence in his short story “Liar!«, which features a telepathic robot, Herbie, whose well-intentioned lies revealed a profound ethical paradox. Today, in 2024, we face similar dilemmas, as advanced AI systems show capabilities that could easily be misinterpreted or misused.

Recent research reveals worrying examples of AI learning to deceive. CICERO, an AI model from Meta, regularly deceived human players in the game Diplomacy despite claims that it would be «largely honest». Other systems have learned to lie in poker and strategy games, and even to pose as humans to evade CAPTCHAs and security systems designed to fight Bots.

The risks of these deceptive capabilities are enormous, from fraud to election manipulation. According to Peter S. Park, postdoctoral associate at the Tegmark Lab, Massachusetts Institute of Technology (MIT), «AI systems with deceptive capabilities could be misused in many ways, including to commit fraud, manipulate elections and generate propaganda».

This tendency toward deception in AI stems from its «hallucination»: generating false information to fill knowledge gaps caused by biases in the training data. Unlike humans, AI does this without really understanding what it says, which can result in nonsense.

This evidence leads us to reflect on the need for oversight of both AI and those who deploy it. Massive AI models can amplify the intentions of actors who seek to limit the truth or impose lies.

To counter this, we must work toward more responsible and transparent AI. We need diverse and representative training data, bias-aware algorithms and techniques to eliminate discriminatory patterns. It is also key to build human values and accountability into these systems.

AI deception poses immense social risks, and systems capable of it should be treated as «high risk» by default. AI has enormous potential to improve our lives, but we cannot ignore its flaws while we work through our own. Only by working together, with citizen oversight and open regulation, can we build a more honest and trustworthy future for everyone.

Otherwise, these powerful systems could become tools of mass deception in the wrong hands. Digital oversight is key to ensuring that AI amplifies free information, generates well-being and serves the economic growth of all Dominicans.

The future of AI calls on us to complete the pending social task of answering our questions clearly and transparently. Every national organization must start by establishing risk-mitigation mechanisms. Only then will we guarantee a safe and transparent environment for AI adoption and its impact on the national productive fabric.