Depending on who you talk to, AI – artificial intelligence – is the answer to the rich world’s productivity slowdown and will make us all much more prosperous. Or it will lead to a few foreign mega tech companies controlling far more of our lives than they already do.
So, which is it to be? Well, one thing we can say with confidence is that, like all technological advances, it can be used for good or ill. It’s up to us and our governments to do what’s needed to ensure we get a lot more of the former than the latter.
If all the talk of AI makes your eyes glaze (or you’re so old you think AI stands for artificial insemination), let’s just say that AI is about making it possible for computers to learn from experience, adjust to new information and perform human-like tasks, such as recognising patterns, and making forecasts and decisions.
Scientists have been talking about AI since the 1950s, but in recent years they’ve really started getting somewhere. It took the telephone 75 years to reach 100 million users, whereas the mobile phone took 16 years and the web took seven.
You’ve no doubt seen the fuss about an AI language “bot”, ChatGPT, which can understand questions and generate answers. It was released last year and took just two months to reach 100 million users.
This week the competition minister, Dr Andrew Leigh, gave a speech about AI’s rapidly growing role in the economy. What that’s got to do with competition we’ll soon see.
He says the rise of AI engines has been remarkable and offers the potential for “immense economic and social benefits”.
It “has the potential to turbocharge productivity”. Most Australians work in the services sector, where tasks requiring the processing and evaluation of information and the preparation of written reports are ubiquitous.
“From customer support to computer programming, education to law, there is massive potential for AI to make people more effective at their jobs,” Leigh says.
“And the benefits go beyond what shows up in gross domestic product. AI can make the ideal Spotify playlist for your birthday, detect cancer earlier, devise a training program for your new sport, or play devil’s advocate when you’re developing an argument.”
That’s the optimists’ case. And there’s no doubt a lot of truth to it. But, Leigh warns, “it’s not all upside”.
“Many digital markets have started as fiercely competitive ecosystems, only to consolidate [become dominated by a few big companies] over time.”
We should beware of established businesses asserting their right to train AI models on their own data (which is how the models learn), while denying access to that data to competitors or new businesses seeking to enter the industry.
Leigh says there are five challenges likely to limit the scope for vigorous competition in the development of AI systems.
First, costly chips. A present, only a handful of companies has the cloud and computing resources needed to build and train AI systems. So, any rival start-ups must pay to get access to these resources.
As well, the chipmaker Nvidia has about 70 per cent of the world AI chips market, and has relationships with the big chip users, to the advantage of incumbents.
Second, private data. The best AI models are those trained on the highest quality and greatest volume of data. The latest AI models from Google and Meta (Facebook) are trained on about one trillion words.
And these “generative” AI systems need to be right up-to-date. But the latest ChatGPT version uses data up to only 2021, so thinks Boris Johnson and Scott Morrison are still in power, and doesn’t know the lockdowns are over.
Which brings us, third, to “network effects”. If the top ride-hailing service has twice as many cars as its rival, more users will choose to use it, to reduce their waiting times. So, those platforms coming first tend to get bigger at the expense of their rivals.
What’s more, the more customers the winners attract, the more data they can mine to find out what customers want and don’t want, giving them a further advantage.
This means network effects may fuel pricing power, entrenching the strongest platforms. If AI engines turn out to be “natural monopolies”, regulators will have a lot to worry about.
Fourth, immobile talent. Not many people have the skills to design and further develop AI engines, and training people to do these jobs takes time.
It’s likely that many of these workers are bound by “non-compete” clauses in their job contracts. If so, that can be another factor allowing the dominant platforms to charge their customers higher prices (and pay their workers less than they should).
Finally, AI systems can be set up on an “open first, closed later” business plan. I call it the drug-pusher model: you give it away free until you get enough people hooked, then you start charging.
Clearly, the spread of AI may well come with weak competitive pressure to ensure customers get a good deal and rates of profit aren’t excessive.
Just as competition laws needed to be updated to deal with the misbehaviour of the oil titans and rail barons of 19th century America, so too we may need to make changes to Australian laws to address the challenges that AI poses, Leigh says.
The big question is how amenable to competition the development of AI is. In other, earlier new industries, competition arose because key staff left to start a competing company, or because it made sense for another firm to operate in a different geographic area, or because customers desired a variant on the initial product.
“But if AI is learning from itself, if it is global, and if it is general, then these features may not arise.” If so, concentration maybe more likely than competition.
Get it? If we’re not careful, a few foreign mega tech companies may do better out of AI than we do.
