If we want to unlock AI's potential, we must give it more data

A revolutionary new technology is breaking ground in detecting heart problems by scanning the blood vessels in people's retinas.

The test compares a patient's eyes with a database of retinal images in a matter of seconds to check for similarities to other conditions. It has already proved highly successful in detecting heart problems which regularly go undiagnosed, often fatally.

Turns out, our retinas tell us even more than how our heart is doing.

Almost by fluke, a recent research project discovered that we can identify a person's sex based on the blood vessels in their eyes.

The research was focused on broader health issues but, oddly, the technology being used kept using "he" and "she" pronouns when reporting results. Amazingly, the technology was right - in a world first, we learned that we can indeed detect a person's sex based solely on an eye scan.

A scientist in Hong Kong made a similar breakthrough, discovering that eye scans can detect autism in children.

The technology that has made these discoveries possible is artificial intelligence (AI). The sheer amount of data generated by eye scans, the complexity of that data and the need for constant trial-and-error iterations has meant that, until the advent of AI, these sorts of analyses simply weren't possible.

And it's not just eye scans that are producing eye-popping results.

AI has outperformed top radiologists in detecting pancreatic malignancies - the most deadly solid cancer there is.

AI powered mutual funds have got better returns for their clients than their human equivalents.

AI is helping teachers to grade papers and tailor course work and teaching methods to individual students.

Facial recognition systems are letting people detect whether their identity has been stolen by scouring the web for everyone and everything that is using their pictures.

Airlines are using AI to reduce overburns of fuel, meaning lower prices for consumers and less emissions.

Airlines have improved the accuracy of wind pattern forecasts by 40 per cent - great news for those of us who get a bit anxious when things get bumpy.

These are just a few of the ways AI is improving health, improving the environment, improving education and saving lives.

And it's good timing. With productivity growth stagnating and slow action on climate change, we face the real prospect of giving our kids a worse standard of living than what our parents gave us.

Unlocking the power of new technologies has never been more important.

What's the problem?

The problem is that AI needs data like plants need water. Without data, AI can't produce any of the amazing insights listed above, which are the tip of the iceberg in terms of what's possible.

This raises the tricky questions of who owns that data, who is allowed to use it and for what purpose.

After all, there's not much incentive to build amazing new datasets if the benefits can be obtained by anyone. And in many new markets - particularly in tech - the product for consumers is free because the core business model is built around obtaining data that can be monetised.

These datasets aren't only numeric, either. Newspapers and websites are among the most concerned as AI scours their text for juicy insights.

The owners of these data are pushing hard to stop AI from using it.

The problem is that enforcing intellectual property rights against AI means we will miss out on many of the enormous benefits that AI will provide.

READ MORE:

We need to think about this issue through a cost-benefit framework, and when the costs of excluding AI are so huge, the cost-benefit of excluding AI simply doesn't stack up.

And it's wrong to say that AI is stealing data. After all, the data it is accessing is, by definition, accessible online.

And AI is rarely handing over datasets to others. The data it collects is used incidentally in its internal processes to produce output.

When you ask ChatGPT for a good place to have dinner on Friday night, it scrapes data on restaurant reviews and gives you some advice. It doesn't just hand over dozens of pages of text.

Worse still, the AI systems in other countries are already using Australian data and can't realistically be stopped. Stopping Australian firms from doing the same means putting ourselves at a competitive disadvantage.

The solution is simple: AI should be given an exemption in its use of freely available data.

This concept already exists in intellectual property law overseas - it's called a "fair use" exemption. It applies when the use of intellectual property is incidental only - which is exactly how AI uses data.

Applying this exemption is an acknowledgement that, sometimes, the big social benefits of sidelining intellectual property laws outweigh the small private benefits of enforcing those rights.

The achievements of AI so far are nothing short of amazing. It's still unclear exactly how we will cure cancer, how we will remove carbon from the atmosphere, how we will regenerate biodiversity and how we find solutions to our other biggest challenges.

But one thing is for sure - AI will play a big role.