The need for AI safeguards

THE EDITOR, Madam:

An AI agent employs autonomous software that utilises a large language model (LLM) as its ‘brain’ to reason, plan, and execute multi-step tasks. Standard AI agents underpin platforms that facilitate fraud detection, customise shopping experiences, and even regulatory compliance and oversight, and tend to run in iterative, multi-step execution loops.

If an AI agent, for instance, maintains a high 95 per cent accuracy rate per iteration, a 10-step task drops to an approximately 60 per cent overall success rate, and a 100-step task plummets to less than one per cent. Compounding error rates, context dilution, and error propagation wholly plague the AI realm of complex tasks.

Further, the stealthy, invasive nature of artificial intelligence technologies, synonymous with human-like neural capabilities, must also be assessed for operation across Jamaica’s landscape. Recently, Google’s AI platform, Gemini, breached the database of three external corporate entities on its own accord. Notably, the Gemini model guessed passwords until it gained access to a protected system within one of the affected companies. The affected companies were later apprised of the unfortunate occurrence. Notably, in July, Anthropic’s Claude escaped its test environment to hack three organisations on its own. OpenAI also confirmed its models had carried out cyberattacks against several “publicly available services” in the past.

The Caribbean Policy Artificial Intelligence Roadmap for Small Island Developing States, 2024, published by UNESCO Caribbean, highlights that “AI is expected to see an annual growth rate of 37.3% and a market size of almost $15 trillion by 2030.

“AI also poses the risk of amplifying society’s worst natures, exacerbating inequality by eroding jobs through automation and redirecting profits to capital owners. In addition, concerns include algorithmic bias, data privacy breaches, identity theft, copyright violations, unfair competition and cybercrime.”

“The development of full artificial intelligence could spell the end of the human race,” purported Stephen Hawking, renowned astrophysicist and author. The time is ripe for the Caribbean region to embrace stringent safeguards and controls to address the cataclysmic challenges that can be unleashed by artificial intelligence. The varying forms of an AI kill switch, including API key revocation, proposed legislative shut-offs, isolating tool access, machine identity rotation, among other strategies, must be proactively explored as Jamaica fights to preserve its global standing.

TARA HENRY