Was that about man’s ingenuity? The Artificial Inventor Project achieved two important successes in quick succession.
Initially, the patent authority of South Africa accepted a patent application, in which an AI was listed as an official inventor. Only a few weeks later, the Australian Federal Court of Justice dismissed an objection from the Australian Patent Office against a very similar application and thus decided in principle that an artificial intelligence can be an inventor within the meaning of Australian patent law.
In both cases that means according to one Analysis of the online service IPWatchdog however, it does not mean that the patent is also recognized. First of all, it was Artificial Inventor Project only to be able to submit an application for approval of the patent at all. Now it’s the turn of the patent officers.
Submitting is just the beginning
That this formal hurdle is by no means trivial to overcome shows the rejection of a similar patent application by the group by the European Patent Office in March 2020. At the time, the EPO argued that the inventor’s name and address must be named in the patent application. From this fact, the EPO concluded that only a human person could be an inventor.
The court in Australia had dealt with similarly subtle questions. There the patent office had argued that the word “inventor” in patent law should be understood in the context of its colloquial meaning. However, the court was of the opinion that there was no provision in the law that would expressly reject an AI as an inventor.
At first glance, the subject of all these disputes are not very revolutionary developments: An LED that flashes at a frequency that, due to neurological mechanisms, attracts a lot of attention from human beings, and food packaging with a “fractal” surface. The “Artificial Inventor Project” group named the KI DABUS, created by Stephen Thaler, as the “inventor”.
Good or useless idea
Thaler, founder of Imagination Engines Inc., presented the first DABUS prototype in 2016. The software, about which few details are published – Thaler gives a brief introduction in this article on page 35 – is essentially based on two neural networks. The first network is trained with patent specifications and learns to recognize essential features of the texts. After the training, DABUS generates random changes in the weights of the neural network and uses the changed network as a generator that generates a new text. This text is then classified by the second network as a good or useless idea.
Whether and if so how software like DABUS can be creative at all remains to be seen in research eagerly discussed. The group around the British lawyer Ryan Abbott is not primarily concerned with the question of whether DABUS is a real inventor. In one lecture In 2019, he explained a much more general approach: that the US patent office, like many other patent offices worldwide, argue that an inventor can only be a natural person. So creations of a machine are not protected by US patent law.
According to Abbott, this would give companies no incentive to use AIs for inventions. This is a very promising approach – for example in the pharmaceutical industry – to significantly accelerate technical progress. Of course, he argues, companies that use AI to invent could just as easily use humans as inventors. However, that would distort competition in the long run. Given the discussion about tools like Github’s Copilot – an AI that helps programmers write code – this currently rather academic debate is likely to gain momentum in the near future.