State machine to control technology

2024-02-11 08:19:00, Tech CNA

State machine to control technology

The last weeks of 2023 were full of new events for the world of technology. On November 28, Abu Dhabi created a new state-backed Artificial Intelligence company, called AI71, which will market the "biggest language model," called Falcon.

On December 11, a French Artificial Intelligence firm called Mistral announced a $400 million funding round, which is expected to raise the firm's value to over $2 billion.

Four days later, Indian startup firm Krutrim unveiled India's largest language model, offered in multiple languages.

Just a week ago, Indian startup firm Sarvam raised $41 million to build similar models in Hindi.

State machine to control technology

Since the American Artificial Intelligence company OpenAI launched the ChatGPT human chat in November 2022, similar news has appeared almost every month.

Against this background, the last four happenings in the technological world may seem ordinary.

However, if we look deeper, they show a wider phenomenon. All three companies, in their own ways, are competing to become national champions.

"We want AI71 to compete globally with companies like OpenAI," says Faisal al-Bannai of the Abu Dhabi Advanced Technology Research Council.

"Bravo Mistral, this is an example of French genius," said the president of France, Emmanuel Macron, recently. ChatGPT and other major English language models "can't capture our culture, language and ethos," declared Krutrim founder Bhavish Aggarwal.

Sarvam started with Indian languages ??because, in the words of its co-founder, Vivek Raghavan: "We are building an Indian company."

Technological race

Artificial Intelligence is already at the center of the growing technological race between America and China. Last year, the two governments pledged $40-50 billion each for other investments. Other states do not want to be left behind or dependent on a core technology that is under foreign control.

Thus, in 2023, in addition to America and China, six ambitious governments in the world, Britain, France, Germany, India, Saudi Arabia and the United Arab Emirates, pledged to finance Artificial Intelligence with figures reaching a total value of 40 billion dollars (see chart).

Most of this amount will go towards buying graphics processing units (GPUs, the type of chips used to train Artificial Intelligence models) and building factories to produce such chips. The rest will be used to financially support Artificial Intelligence companies.

The nature and extent of state involvement varies from one superpower to another. Thus, it seems that new frameworks of Artificial Intelligence are emerging, with the intervention of governments.

We start with America, which has tech companies that are the envy of the world. Its vibrant private sector is rapidly renewing itself without the direct support of the US government. Instead, the US federal government will spend about $50 billion over the next five years to increase domestic chip manufacturing capacity.

The idea is to reduce America's dependence on Taiwanese semiconductor makers like TSMC, the world's largest and most sophisticated chip company. Many politicians in Washington fear that supplies from Taiwan could be jeopardized if China decides to occupy the island, which it considers part of its territory.

Another way America aims to stay ahead of the race is by undermining rivals. President Joe Biden's administration has imposed strict export controls that prohibit the sale of the latest Artificial Intelligence technology to adversaries such as China and Russia, including chips and chip-making equipment.

It has also banned Americans from sharing expertise in Artificial Intelligence with those countries.

Now it is forcing countries that are neutral to choose a side. In October, the US government began requiring companies in third countries, including Saudi Arabia and the United Arab Emirates, to secure a license to buy AI chips from US firm Nvidia, which sells most of them.

The rules "tend to give approval". That means the US government will "probably allow" sales to such firms, says Gregory Allen, who used to work at the Defense Department, as long as they don't have close ties to China.

On December 6, Xiao Peng, who runs an Abu Dhabi-backed AI startup firm called G42, announced that the company will cut ties with Chinese hardware suppliers such as Chinese electronics company Huawei.

State machine to control technology

China game

China's AI strategy is in large part a response to US containment efforts. According to data from research firm JW Insights, between 2021 and 2022, the Chinese state spent nearly $300 billion to recreate the supply chain of chips at home (for artificial intelligence and other semiconductors) in order to be immune from sanctions. Western.

Much of that money may have been wasted, but it helped Huawei and SMIC, China's largest chipmaker, design and produce a highly sophisticated GPU last year.

Central and local authorities also channel capital into AI firms through state-backed “guidance funds,” with around 2,000 of them investing in all kinds of technologies seen as strategically important.

The Communist Party is also directing private money towards its technological priorities. Often, it does this by hitting certain sectors (most recently, in December, the video game sector), while providing guidance on which industries investors should focus on.

The government is also promoting data sharing, where businesses can trade commercial data on everything from sales to production, allowing small, ambitious firms to compete in areas where previously only large, wealthy companies could compete. data. There are already 50 such exchanges in China.

Elements of this state-led approach are being emulated in other parts of the world, particularly in the oil-rich Gulf states. Autocracies like Saudi Arabia and the United Arab Emirates may move faster than democratic governments, which must heed voters' concerns about AI's impact on things like privacy and jobs.

Since they are also rich countries, they can afford the production of GPUs (for which the two countries together have spent several hundred million dollars so far) and the energy needed to run the power-hungry chips .

The two countries can also invest money in human capital development. Their well-equipped universities are quickly climbing the global rankings.

The Artificial Intelligence Program at the King Abdullah University of Science and Technology, in Saudi Arabia, and the program at the Mohamed bin Zayed University of Artificial Intelligence in Abu Dhabi, the world's first school focused on Artificial Intelligence, have drawn excellent professors from prominent institutions. such as the University of California, Berkeley and Carnegie Mellon University in Pittsburgh.

Many of their students and researchers come from China. And many decide to stay. Almost all graduates of Mohamed bin Zayed University, numbering several hundred, stay in the region to work in local firms and laboratories, says Professor Timothy Baldwin (who went to work in the Middle East from the University of Melbourne in Australia). .

The approach of the Persian Gulf countries is yielding results. The capabilities of the Falcon AI model, originally built by a team of 20 engineers, are rivaling those of the Llama 2 model, created by US tech giant Meta. AI71 plans to improve the open-source models, using national data sets from fields as diverse as health, education and one day, perhaps even oil.

"In the last 50 years, oil led the country... now, the 'new oil' is data," says Mr al-Bannai.

Harmonization problem

Several other governments are combining elements of America's approach with those of China and the United Arab Emirates. The EU has its own version of US incentives for domestic chip manufacturing. Also some member states:

Germany will foot a third of the 30 billion euro ($33 billion) bill for a new chip factory to be built in the country by US manufacturer Intel. Outside the bloc, Britain has pledged to pour 1 billion pounds ($1.3 billion) over the next five years into Artificial Intelligence and supercomputers (although it has not gone into detail on exactly how the money will be spent).

India's government is boosting semiconductor manufacturing with generous "incentives", encouraging cloud computing providers to build more Indian data centers where AI models are trained, and is considering buying 1.2 million GPUs billion dollars.

Like China and the Gulf countries, but unlike America, where the federal and state governments refuse to share public data, India and some European countries have not hesitated to make such data available to private firms. Artificial Intelligence.

The French government "has been very supportive" in this regard, says Arthur Mensch, the head of the French firm Mistral. Britain is considering the possibility of firms using rich data belonging to the National Health Service.

The Government of India has large amounts of data from its suite of digital public services, known as the India Stack. The country is expected to integrate Indian AI models into digital services.

Unlike China, which imposes strict consumer-oriented AI regulatory laws, Britain, France, Germany and India have, at least for now, supported soft AI regulatory laws, and even India has no such laws. such.

The French and German governments are beginning to be skeptical of the EU's Artificial Intelligence Act, whose final details are being hotly debated in Brussels.

The reason for their skepticism is that the law could restrict French firm Mistral and German firm Aleph Alpha, Germany's most successful builder of large language models, which raised 460 million euros in November.

It is natural that states want some control over this technology that could have a radical effect on the world. When it comes to sensitive and highly regulated sectors in particular, such as defence, banking or healthcare, many governments would not want to rely on artificial intelligence imported from abroad.

However, any dose of nationalism in this area has its own risks.

State machine to control technology


America's approach is likely to worry not only its adversaries, but also some allies. China's burdensome regulatory laws may dilute some of the potential benefits of big spending.

Building models for local languages, as Indian firms Krutrim and Sarvam plan to do, may be a futile initiative if foreign models continue to improve their multilingual skills.

The Gulf countries' push for open source models could backfire if other governments restrict their use, as Mr Biden has hinted in a recent executive order, and the EU could do the same through the act of artificial intelligence, fearing that large open language patterns could be used for malicious purposes by miscreants.

Saudi and UAE institutions may struggle to retain talent.

An artificial intelligence expert who worked at Falcon admits he benefited greatly from partnering with a team of French engineers who were then recruited by Silicon Valley startup Hugging Face. As one skeptical investor notes, it's not yet clear how extensive or useful the UAE's public records actually are.

As Nathan Benaich, of the firm Air Street Capital, sums up, most attempts by governments to create national models "are probably a waste of money".

But surely Mr. Benaich's warning will not discourage governments curious about Artificial Intelligence and mindful of the potential rewards if they succeed. Mr. Macron will not be the only leader who will receive such warnings with a Gallic-style shrug./ Monitor.al

Lajmet e fundit nga