ARTIFICIAL intelligence systems have learnt to be “persuasive liars” and pose a serious threat to humanity, British spooks warn.
The tech could increase the powers of cyber hackers and terrorists, while AI-generated images and videos may drive new scams such as fake kidnapping and sextortion.
Ahead of a major speech on AI by Rishi Sunak today, intelligence agencies have mapped out the potential dangers awaiting.
Their report warns: “Highly capable future frontier AI systems, if misaligned or inadequately controlled, could pose an existential threat.”
It adds: “AI models can maintain coherent lies in simple deception games, and larger models are more persuasive liars.”
The report raises a host of concerns that AI will boost the powers of hackers, warning of scam victims suffering emotional distress caused by fake kidnappings or sextortion.
The report says many experts, including British spies, believe it is a risk.
However, there is still time to minimise the danger
For criminals to benefit, they would need AI to “outpace mitigations, gain control over critical systems and be able to avoid being switched off”.
The report maps the “catastrophic” pathways in which AI could become a threat.
As a self-improving system, it could harm human interests.
Without proper regulation, people risk an over-reliance on AI and may become “irreversibly dependent”.
Mr Sunak is expected to say today that AI “brings new dangers and new fears”.
He will add: “Doing the right thing, not the easy thing, means being honest with people about the risks from these technologies.”
He is due to host world leaders next week at the first ever international AI Safety Summit at Bletchley Park, Bucks.