AI must be democratic and cheap

It's the beginning of an ominous trend, where the most powerful AI models are being reserved for those with the biggest wallets. We need alternative ways to ensure that the most transformative technology of our time becomes accessible to all.

The powerful AI models we see today are trained on the aggregate knowledge of man: our shared cultural heritage. It is the sum total of all our languages, history, texts and culture. Much of the material is shared freely or in the public domain.

A few tech companies have taken the world's knowledge without asking for permission. A band-aid on the wound was that the technology was at least available for a cheap buck. Now it may appear that the technology that springs from this collective knowledge may become increasingly inaccessible.

Imagine a young scientist in Kenya developing cutting-edge solutions for local water management. The most advanced AI tools that could have helped her cost more than the average monthly salary in the region. At the same time, scientists in rich countries with unlimited access are sitting on the same tools.

When tools built on humanity's accumulated knowledge become available only to the privileged, we risk concentrating the innovation power of the future in the hands of a few.

It may be tempting to hope that market forces will resolve this over time. But AI is unlike previous technologies. Artificial intelligence is not just a consumer good, but a technology that actively generates new knowledge and insights. Unequal access to AI models can cause the knowledge and innovation gap to increase exponentially. Rather than closing the wealth gap between poor and rich, we will see galloping inequalities.

One solution to the problem could be to tax the AI companies in line with the value of the heritage that has gone into training the models. The lion's share of the value of the models must be said to be jointly owned. It could defend a particular tax on AI companies.

But taxes and subsequent redistribution are not enough. We must ensure that everyone has access to the knowledge expressed in the AI models. One possibility is to treat basic models as public infrastructure, developed through international cooperation. As everyone should have access to equal health care, everyone should be ensured access to AI.

Another is that the government takes more of the bill. If the differences between the premium models and the base models become large enough, it is essential to ensure all access to the model, but someone will have to pay. A public-private partnership, where the government accounts for more of the funding, may be a better solution than private consumption and private investors funding it all.

We must make choices today that ensure that technology becomes a force for democratization and shared progress, rather than one that reinforces existing inequalities.