Race to deploy LLM tools must be matched by AI safeguards: Nicolas Miailhe

Rapid diffusion of AI apps raises systemic safety and security risks, warn AI Safety Connect cofounders

The pace at which artificial intelligence (AI) tools built on large language models (LLMs) are diffused and distributed to the general public must match the safety and security required for the technology to flourish and reliably deliver gains, said Nicolas Miailhe, cofounder of AI Safety Connect (AISC).

 

“One problem we collectively face is that capabilities are accelerating and delivering amazing results, but they are difficult to stabilise industrially to give us the gains and controls. This creates systemic safety and security risks,” Miailhe said.

 

AISC is a Paris-based AI advocacy body that organises dialogues on AI safety with stakeholders across the globe.

 

Technologies such as AI, which raise systemic safety questions, must also be viewed through the lens of speed and the direction in which they are progressing, he said.

 

In India, AI progress is coupled with democratic institutions that keep a check on the technology’s direction. The discussion on the safety and security of AI systems is not pessimistic, however, as control over such general-purpose technologies is not feasible, said Cyrus Hodes, also a cofounder of AISC. 

 

“Mathematically, we do not have the knowledge to achieve that control over AI. However, there is hope in enlightenment. India has a massive role to play there,” Hodes said.

 

Investments in safety and security, especially for countries like India and others in the Global South, are paramount to ensure that AI delivers gains in a way that protects users’ rights and builds trust. If customers do not trust the technology, they will walk away from it, he added.

 

Most forecasts project that the total economic output from the market built on the AI stack will likely exceed $1.5 trillion by 2030. By building tools that ensure AI users’ safety and security, India can carve out a $300 billion opportunity from the larger pie, Hodes said. 

 

India can replicate its successes in the pharmaceutical sector as a generic drug manufacturing and testing ground for AI by investing in the testing, evaluation, validation, and verification of frontier models used in advanced AI tools, Miailhe said.

 

“India can create an influential position for itself on the global stage by harnessing the power of evaluating these AI safety and security tools. Testing and evaluation remain an unsolved scientific and engineering problem. Historically, India has always been good at solving these,” Hodes said.

 

The country, he said, can be among the leaders of the middle-power nations that can help establish the boundaries of AI safety and security for everyone, including leading countries such as the US and China.