How to democratise artificial intelligence

The rapid advance of artificial intelligence evokes both wonder and dread. Many regard AI as an object of marvel and awe (a Stupor Mundi, to borrow a Latin phrase), while others believe it can be a benevolent savior (a Salvator Mundi). Regardless of whether AI is seen as miraculous or merely helpful, the question remains: How can we ensure that its benefits are available to everyone?

To answer this question, we need a nuanced understanding of AI. That means rejecting several simplistic narratives: functionalism, which says humans should adapt and augment themselves to keep up with technological progress; sensationalism, which depicts AI as an existential threat; cynicism, which seeks to exploit AI for profit; and fatalism, which implies a resigned acceptance of AI’s inevitable rise.

What these scenarios overlook is that the future is still ours to shape. Adopting the verum-factum principle – knowing through making – is crucial to developing a more profound understanding of AI’s capabilities and implications.

To prevent a minority from co-opting AI’s transformative potential, it must be democratized. Equitable access is the key to ensuring that the benefits of technological progress are broadly shared, and that AI serves as a unifying force, rather than exacerbating the divisions within our fragile societies.

The potential benefits are enormous. In the 1990s, Joseph Stiglitz observed that “a child anywhere in the world who has access to the internet has access to more knowledge than a child in the best schools of the industrial countries did a quarter-century ago.” By democratizing access to AI, we can empower today’s children to engage with humanity’s brightest minds in a way that caters to their individual needs.

But achieving this depends on how we shape the narrative surrounding AI’s adoption and future impact. Instead of making lofty promises like “AI will solve world hunger,” we should focus on its ability to bring about incremental yet meaningful improvements in people’s daily lives.

In this regard, the technology’s rapidly expanding capabilities and declining costs create new opportunities for smaller-scale models and enable individual users to personalize AI solutions, mirroring the internet’s freewheeling, creative early days. Just two years ago, for example, the leading open-source AI model was Meta’s OPT-175B. Today, a leading open-source model, Mistral 7B, is 40 times smaller, at least 40 times cheaper to operate, and outperforms its predecessor. Remarkably, it was developed by a company of just 18 people.

And this is just the beginning. AI is currently experiencing its own version of Moore’s law, setting the stage for rapid uptake, akin to the diffusion of telephones and televisions. This accelerated process calls for a shift in focus toward developing practical applications and mitigating risks, rather than fixating on reducing costs.

The rise of AI is a double-edged sword. The technology could be a powerful equalizer or a source of division, depending on how it is deployed and who controls it. Like previous technological revolutions, it promises to create new employment opportunities while simultaneously threatening to displace existing jobs. A recent report by the International Monetary Fund underscores this point, warning that AI could lead to a growing divide between tech-savvy individuals, well-positioned to reap the economic benefits of innovation, and those who risk falling behind.

But our understanding of these technologies must reflect their complexities and the power of human ingenuity. By developing and promoting AI systems that significantly improve essential services, especially in underserved regions, we could ensure that their benefits are broadly shared. To achieve this, AI technologies must be deployed with the explicit objective of reducing existing inequalities.