A large number of startups are pursuing business ideas that use artificial intelligence (AI) such as machine learning. For example, systems for image data processing, speech recognition or for the "Internet of Things" are being developed that contain an AI component. Under certain conditions, an AI-based business idea is patentable.
General advantages and disadvantages of a patent for startups
Patents are monopoly rights with a term of 20 years. They can hinder competitors in the market, exclude them from the market or generate licensing income. Patents protect against imitation and are therefore popular with investors. Patents are thus an additional raw material for start-ups to successfully position a business idea on the market. Disadvantages are, of course, the costs and the fact that the AI invention is publicly available after 18 months after the patent application is filed with the patent office. Disclosure of the invention is the inventor's consideration for his 20-year monopoly right.
A higher patenting hurdle applies to AI inventions in Europe
AI inventions are treated by the German and European Patent Offices like software inventions or like software patents. Compared to inventions from traditional technical fields such as electrical engineering or mechanical engineering, the patenting hurdle for software inventions is higher. Compared to traditional inventions, software inventions have to fulfill additional requirements in order to be patentable. The higher patenting hurdle is ultimately a political requirement. In other markets, e.g. Japan, the patent hurdle for software inventions is significantly lower. A look at other non-European markets can therefore be well worthwhile for a startup in order to gain a competitive advantage with a patent.
Requirements for the patentability of AI inventions
First, the idea must not yet exist. If the AI invention has been described in an online article, for example, or if a patent application has already been filed by a competitor, it cannot be subsequently protected by a patent. The AI invention should have been filed for patent before publication or before the competitor. Incidentally, this applies to all inventions, not just AI inventions.
Another requirement is that the AI is used in a technical application or is specifically technically implemented.
How does AI satisfy the technical application requirement?
In connection with the invention, the AI must produce a technical effect that solves a technical problem. This is the case, for example, if
- the AI's output data is used to control a particular technical system (e.g., an X-ray machine) and produces a technical effect in the system that is beneficial (e.g., that the X-ray machine measures more accurately);
- Audio, image, or video data is enhanced or analyzed by AI (e.g., person recognition in a digital image). Such data are typically viewed as physical entities, and AI is then viewed as a technical process that processes these physical entities;
- in a computer network, load balancing is optimized by AI.
How does AI satisfy the requirement of specific technical implementation?
As an alternative to technical application, it may be sufficient for patentability that the AI is specifically technically implemented. This is often the case if the AI is not implemented by a programmer but by a "technical programmer". The technical programmer makes technical considerations that go beyond simply finding a computer algorithm to execute a procedure. For example, a patent was issued in the past that delegated the execution of data-intensive training steps of a machine learning algorithm to a graphics processing unit (GPU) and preparatory steps to a standard processing unit (CPU) to take advantage of the parallel architecture of the computing platform. This idea, implemented by a technical programmer, did not exist at the time.
Last hurdle of the AI invention
If the above requirements are met, the door to a patent is open for the AI invention, just like for an invention in a classical technical field. In practice, there is basically only one hurdle left to overcome: The AI invention must not arise in an obvious way from the prior art, as already indicated above. The prior art is usually determined at least rudimentarily before a patent application is drafted and filed with the patent office, e.g. by a patent attorney or patent searcher. Thus, even before the cost-intensive drafting of the patent application, it can be estimated whether a patent application based on the prior art is reasonable. After the invention has been filed with the patent office, a patent examiner again searches for prior art ex officio and then conducts an official patent examination.
Patenting an AI invention can lead to a competitive advantage and therefore make a lot of sense for a startup. For AI inventions, additional requirements have to be fulfilled in Europe and Germany compared to inventions in classical fields of technology.
If you have any questions about AI inventions, the author Patent Attorney Michael Schüller will be happy to help.