“That is a real key piece of (AI) sovereignty, which is building up and buying from your own companies. If you need the companies to build more, then countries should invest in them. This should not be something where you need to parachute in experts from somewhere else,” he said.
Surman was in New Delhi recently to attend the AI Impact Summit.
Though several frontier technology companies have open-weight AI models that can be used to build products, they are not entirely useful for companies in countries such as India, as the firms building these AI models will rarely share details, such as those about pre-training, the data sets used, and others, Surman said. In such a scenario, countries such as India should build truly open-source AI models, so that everyone benefits from them, he said, adding that public money should be spent towards building products that benefit the public, not companies.
The development of a frontier AI model, however, does not mean starting from scratch, Surman said: “The really amazing part about the last five years is that the models are not really the difficult part anymore. That has become a commodity. You can borrow all of the innovations that have happened. That is what open-source allows us to do.”
It is also critical for countries to demand that the AI used by their citizens reflects and can be shaped by their respective cultures and societal norms, he said. Projects such as the Common Voice, being shepherded by the Mozilla Foundation aims to fix the problem of ‘language tax’ that AI users in the Global South, including India, have to face since most of the frontier AI models have been trained in English and are suited to Western culture as well as societal preferences, Surman said.
“The government or businesses in India can pick up what is already made under Common Voice and build it according to their own preferences,” he said.
 "Mark Surman, the president of the Mozilla Foundation")
