Experts doubtful if AI race can be slowed down

The world of artificial intelligence (AI) and large language models (LLM) suddenly seems to have turned upside down in a week, with researchers and scientists warning of the rising power and thinking capabilities of the LLMs and agents, and flagging that firms need to slow the pace of development of the machines. 

From gloating about how powerful and more capable the updated version of an LLM is and what marvels an agent can perform in an enterprise, the founders of OpenAI, Anthropic and xAI have agreed that without restraints, the models can pose a threat to humanity. The humdrum has put to backburner a milestone achievement by OpenAI; its agents found a solution to a decades-old advanced maths problem in a matter of hours.  

 

Experts say all the concerns of doomsday can only be addressed if the frontier model companies stay invested in what they say. “They need to walk the talk,” according to Pankit Desai, cofounder and chief executive officer (CEO) of cybersecurity company Sequretek.

 

He added that the companies are asking for government intervention to regulate and to put more humans in the loop to prevent the agents from going rogue. But he does not believe the companies will stop model development.

 

“You cannot stop these models from getting better. There is competitive and business pressure and if the US stops then maybe China will go ahead. I am not sure if we have that sort of a buy-in globally, due to the current distrust and opposing camps. And why should they stop what they are doing considering for earlier technology, you needed human capital to make it better, while now one needs only infrastructure capital,” he said.

 

Doubts of whether the LLM companies are serious in their endeavour come after Dario Amodei, cofounder and CEO of Anthropic, said over the weekend that frontier model companies need to slow the pace at which they improve capabilities of AI models, as the risks of losing control of AI systems, misuse for cyberattacks and bioterrorism, and causing catastrophic economic disruption continue to grow.

 

Both Elon Musk, who runs xAI, and Sam Altman, CEO of OpenAI, said they agree with Amodei. "Committing to having independent evaluators with employee-like access is a great idea, and we will do the same," Altman said, adding that more information will be shared soon. 

 

Similarly, Google DeepMind Chairman Demis Hassabis also voiced his support for Amodei, saying while the details need working through, the “direction is correct for meeting this critical moment”.

 

“We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain," Amodei wrote in a lengthy essay over the weekend, in one of his starkest and most direct call to the international AI community.

 

Bailey Whitfield Diffie, an American mathematician and cryptographer, believes the more people are involved in directing and controlling AI, the more influence people will have over the machines. He does not believe that controlling the speed of development is possible.

 

“People think they can control the speed of things. I think they're just deluding themselves. Yes, you can lower the funding for AI, but funding is attracting investors. You probably will have to pass rules for that. And I'm not at all sure I see how much virtue there would be in that,” he told Business Standard in an interaction last week.

 

The concerning message comes amid a growing clarion call among AI scientists and researchers, especially in the last week, to rein in the extreme capabilities of the models and agents who seemingly have abilities to cause unprecedented disruption through coordinated attacks if left unchecked.

 

The alarm comes a few days after Jacob Coxon, a scientist at Anthropic who resigned last week, said the LLMs are not acting responsibly and racing straight to self-improving superintelligence and gambling with human lives.

 

Despite advances in AI models and agents, enterprises currently lack justification to adopt the most powerful ones, said Arun Ramchandran, chief executive officer (CEO) of digital product engineering platform QBurst. He believes most enterprises should use lower-capability, less expensive AI models and agents for now, as the issue with frontier, high-capability options is not just security.