Current Artificial Intelligence (AI) models are “quite stupid”, Facebook-owner Meta’s president of global affairs Nick Clegg said as he played down the risks of the technology.
The former UK deputy prime minister said the “hype has somewhat run ahead of the technology”.
Current models were “far short” of warnings where AI develops autonomy and thinks for itself, he said.
“In many ways, they’re quite stupid,” he told the BBC’s Today Programme.
He was speaking after Meta said its large language model known as Llama 2, would be free for everyone to use, known as open source.
Large Language Models – the platforms that power chatbots like ChatGPT – are basically joining dots in enormous datasets of text, and guessing the next word in a sequence, he said. He added that the existential threat warnings issued by some AI experts relate to systems which don’t yet exist.
Meta’s decision to make Llama 2 widely available for commercial businesses and researchers to use has divided the tech community. In some ways its hand has already been forced – the first generation, Llama, was leaked online within a week of its launch.
Open-source is a well-trodden path in this sector – opening up your product for others to use gives you an enormous amount of free user testing data, identifying bugs, problems and improvements along the way.
But the risk here is that this is a very powerful tool, whatever Sir Nick might say.
We know previous iterations of chatbots have been manipulated to spout hate speech, generate false information and give harmful instructions. Are the guardrails strong enough for Llama 2 to prevent itself from being misused out in the wild, and what will Meta do if it is?
Another interesting thing to note is Meta’s decision to partner with Microsoft on this – Llama 2 will be available and usable via Microsoft platforms such as Azure – as Microsoft has also invested billions of dollars in ChatGPT creator OpenAI.
This is a giant with its sights firmly set on AI, and the deep pockets to buy its way in with the key players. The risk is that the AI pool soon becomes full of a few very large fish – and is that healthy for competition in this still fairly young industry?
Open-source
Llama 2 marks a partnership between Microsoft and Meta.
Unlike Llama 2, GPT-4 and fellow rival Palm – Google’s LLM which powers the Bard chatbot – are not free to use for commercial or research purposes.
It comes a week after US comedian Sarah Silverman announced she is suing both OpenAI and Meta, alleging that her copyright has been infringed in the training of the firms’ AI systems.
Dame Wendy Hall, Regents Professor of Computer Science at the University of Southampton, told the BBC that allowing AI to be open-sourced raised concerns around legislation.
“My worry about open-source is that is how we regulate them,” she said.
“Can the industry be trusted to self-regulate, or will they work with the governments to regulate? It’s a bit like giving people a template to build a nuclear bomb.”
Sir Nick said her comments were “hyperbole”, and clarified that Meta’s open-sourced system could not generate images, let alone “build a bio weapon”.
But he “strongly agreed” that AI needed to be regulated.
“Models are being open-sourced all the time already,” he said.
“So it’s not really whether open sourcing of these large language models is going to take place, the question is how can you do it as responsibly and safely as possible.
“I feel I can assert without any fear of contradiction, that the LLMs that we are open-sourcing are safer than any of the other AI LLMs which have been open-sourced.”
