AI is too important to leave to Google and Facebook alone

AMERICANS don't have to be beholden to the tech Goliaths to get the benefits of artificial intelligence. An alternative possibility is for the government to provide the infrastructure needed for a technological future - through a public option for artificial intelligence.

Big tech companies have an extraordinary amount of data about how we behave, largely because they engage in widespread surveillance of much of our behaviour. Because AI depends on data, these companies have a huge market advantage over startups and entrepreneurs - and it's a gap that will only get wider. The lax regulatory environment hasn't helped, either; instead, it has allowed the biggest tech companies to acquire their rivals, stifle competition and snatch up the best software engineers and data scientists. Together, these dynamics make it hard for startups, governments and non-profits to develop and use artificial intelligence without relying on big tech companies, effectively ceding influence over this developing field to a private sphere distorted by anti-competitive practices.

The alternative is a public option for AI.

The public option, a familiar term from debates over healthcare, is a public programme that provides universal access to goods and services, with a private opt-out. A public option for AI wouldn't prevent companies such as Google from collecting and using data. But it would provide a pathway for startups and public-sector organisations to develop abilities and products that would compete with those of the tech giants.

Although the Trump administration released an executive order on AI this year, we believe that a broader conceptual framework for the public option for AI - coupled with significant financial resources from Congress - is an opportunity for all levels of government to take control of technology for their constituents and engage them deeply in the development of the rules by which it is governed and used.

THREE COMPONENTS

Our proposal has three components: The first is a public data pool that would make data accessible to registered users. Local, state and federal governments have sizeable data resources that would seed this digital commons. Users would be verified to block foreign governments, hackers and others with ill motives from access, and users would be prevented from using the data to engage in racial or other forms of discrimination and for micro-targeted advertising.

Some of the data may be very sensitive, and access to those resources would be highly regulated. We can imagine a variety of ways that regulation and technology together could protect privacy and still foster innovation: Data could be anonymised at the source; the commons could have an interface that allowed users to derive insight from the data set, while leaving the underlying information inaccessible; less sensitive data, such as weather information, could be made available in a format optimised for training AI.

What's more, methods for safely sharing AI models without disclosing the underlying data are being developed today and could enable users of the data commons to collaborate on public-interest AI services. The federal government should also invest in researching new and better ways to protect privacy and prevent misuse.

Second, a public option for artificial intelligence would include a significant increase in research and development spending. Proponents of big tech celebrate private-sector research and are right to do so. But big tech companies, like all companies, have an incentive to fund research that will support their bottom-line, and the profit motive doesn't always mean a focus on the most important problems.

GROUNDBREAKING WORK

For generations, government R&D spending has been one of the central engines of economic growth and technological progress in America. Yet, China is projected to spend far more than the United States on AI research over the next decade. A sizeable increase in research funding for companies, governments and non-profits developing public-interest technologies would help expand the types of research taking place and give scientists and engineers the option to do groundbreaking work on a broader range of problems.

Third, much of government's AI work takes place in the military sector and is applied to national security problems. But healthcare, transportation, energy and other areas could also benefit significantly from AI. The federal government should expand its AI procurement across all of these sectors as an opportunity to improve public services for all Americans. In addition, the government should ensure that its use of algorithms meets the highest ethical standards.

A public option for AI can't solve every problem related to technology and surveillance, and it would require careful thinking about public governance of these programmes - including a commitment to privacy and awareness of biases. But it would help address the problem of a small number of companies having virtually all power over this technology. It would facilitate the conditions for a competitive market with many players and many new innovations, all while preserving our democratic values and improving our society. NYTIMES