IT IS NOT A COMPETITION
AI development is not about winning races, but about ensuring the well-being of humankind, writes
AT this year’s Davos economic forum, United States executives warned that China may be winning the so-called Artificial Intelligence (AI) race with Europe. In another recent article, Bloomberg pointed out that countries are rushing not to be left behind.
The author also pointed out that there’s still a long way to go before AI will be commercially viable.
In my opinion, speaking of a “race” is both wrong and dangerous. It puts the focus on competition and brings with it a sense of gloom and despair. So let me make two arguments: first, there is no race and second, if there was, it would be the wrong race to engage in.
There’s no race because of the very definition of a race: it’s a competition of speed, usually judged by an objective measure like a clock or to a specific end goal. In AI development, however, we don’t have an end point —no specific time to stop.
Therefore, there’s no way to determine when and where someone will win this so-called race. Suggesting that it can be won assumes a moment after which we can stop developing technology, and advancing humankind.
It’s even more important to understand why it’s the wrong race to engage in. The US and China are betting on machine learning developments, and in particular on deep learning, as the approaches that will achieve true AI, and enable them to “win” that so-called race.
These approaches rely on the availability of huge amounts of data and computational power, to enable machines to perceive, or learn characteristics of a particular domain.
This approach is used to recognise faces in pictures, to determine the creditworthiness of mortgage applicants, and to diagnose cancer cells in scans or Xray images.
All of these are relevant and important applications, and the progress achieved in the last few years is truly remarkable. The goal is not to win races, it’s to ensure the well-being of humankind and the environment.
However, these approaches are focusing on one aspect of intelligence: the ability to perceive patterns and make predictions based on those patterns. True intelligence, on the other hand, includes more than that, like the capability to reason, interact and decide based on little, incomplete and contradictory information. In short, we need to explore alternatives to statistical approaches to learning.
In fact, just a few weeks ago, a study analysing 25 years of AI research has concluded that the era of deep learning is coming to an end. Europe has traditionally been strong on symbolic approaches to AI and on (social) robotics.
These are some of the areas that should be invested in and that will bring AI forward in the near future. Therefore, it would be a mistake to blindly follow US and China on their machine learning “race” when we now have the opportunity to show the value of alternative approaches.
Another reason why dataheavy approaches are not the way forward: they have a negative impact on human well-being and the environment. Any development that does not boost trustworthiness will not succeed.
There’s no business model for untrustworthy AI or unethical AI. The results and decisions taken by systems based on deep learning and neural networks are hard to understand and explain. Therefore, they aren’t sustainable in areas where the trust of users and experts is crucial.
... there’s no way to determine when and where someone will win this so-called race. Suggesting that it can be won assumes a moment after which we can stop developing technology, and advancing humankind.
The writer is a professor at the Department of Computing Science at Umeå University in Sweden. She heads the research group “Social and Ethical Artificial Intelligence”