Patenting Artificial Intelligence and quantum technologies

Europe

IP and Life Science analysis: Caitlin Heard, IP Litigation Partner, and Rachel Free, Patent Attorney and Partner, at CMS UK, discuss the applicability of Artificial Intelligence (AI) and quantum technologies to the pharmaceutical industry and the challenges of patenting these new technologies.

The article was originally published on LexisPSL.

What is AI, machine learning and quantum computing?

AI is a branch of computer science in which scientists use computers to understand human intelligence. Many different computer science technologies have been created in the field of AI such as: planning, reasoning, constraint satisfaction, rule-based systems, natural language processing, computer vision, intelligent sensing and control, and machine learning. One of these technologies, machine learning, has become very successful in recent years.

Machine learning is where computers learn from examples in a manner such that the computer is able to generalize from the examples in order to handle a task it has not seen before. In machine learning, the computer has a model such as a neural network, support vector machine, random decision forest or other type of model with many parameters. During a learning process, values of the parameters are updated using update rules according to examples. When a new example which has not been seen by the model before is presented, the trained model is able to process the example and compute a useful prediction. Suppose the examples are medical images labelled as depicting cancer or not. The prediction will be a predicted label eg cancer yes/cancer no.

Quantum computing uses quantum bits (qubits) to process data. A qubit represents more than one state at the same time whereas a classical computer uses transistors which are either on or off. For some types of problems, a classical computer is unable to compute an answer in a realistic time scale because the number of possibilities to assess is too great. In contrast, quantum computers have potential to compute answers to certain types of problem in a realistic time because of the ability to represent multiple states at the same time. Shor’s quantum computing algorithm is an example which can find prime factors in realistic times. Quantum computers operate at extremely low temperatures and are extremely expensive and difficult to engineer.

What are the applications of AI, machine learning and quantum computing and why is this technology particularly useful for the pharmaceutical industry?

Machine learning involves lots of number crunching during the learning process when updates are computed. There is potential to use quantum computing to speed up that processing so that learning would be achieved more quickly. In the case where the training data is quantum data, the process of quantum machine learning seems to have particular potential. In the pharmaceutical industry where molecules, proteins and other chemical structures are to be designed and assessed, it is possible to use machine learning for various tasks such as predicting protein binding sites, interpreting x-ray crystallography results, and predicting molecule structures amongst other tasks.

What are the issues with patenting inventions relating to AI, machine learning and quantum computing?

There are a number of issues with patenting inventions related to these technologies, including:

  • Novelty: often novelty and inventive step is straightforward to demonstrate for quantum computing inventions as the pool of researchers in the field is small and innovation is rich. However, it may be necessary to include mathematical detail in the claim in order to achieve novelty and inventive step. In the case of machine learning, it can be harder to demonstrate novelty and inventive step since the field is more advanced, the amount of prior art is significant, and there are many ‘off the shelf’ solutions available.
  • Sufficiency: it is often difficult to demonstrate that quantum computing inventions meet the requirements for sufficiency, because often these are theoretical inventions found using mathematics and computer simulation and a physical quantum computer implementing the technology may not have been created.
  • Commercial considerations: deciding whether to proceed with a patent or with trade secret protection is a commercial decision influenced by many factors. At least in the field of quantum computing, competition for experts is fierce so that applicants have to think carefully about how to protect their technology in this nascent field.

Is it worth patenting these inventions?

A number of legal and commercial considerations come into play in assessing whether an invention should be patented. One must specifically consider the nature of the invention, and the nature of likely infringements. It is also essential to have an eye to enforcement as AI patents present some particularly unique challenges:

  • Duration of protection: this is likely to be a crucial factor in the field of AI/machine learning and quantum computing. Patents have a life of 20 years, but certain industry sectors are adopting AI applications quicker than others. With many commentators suggesting that quantum computers will not become widely commercially available for another decade, certain inventions may not reach commercial maturity during the life of the patent, so alternate protection should be considered.
  • Ease of detection: if infringements are likely to take place in a ‘black box’, detecting infringement will be difficult. Further, the nature of AI, and particularly machine learning, means that infringements may only be transient or temporary.
  • Practicality of enforcement: given that AI technology evolves over time, consideration should be given to where in the process or supply chain the claimed invention is directed. Consideration should also be given to whether one can patent the end commercial product. It is possible that any modification to the AI output could cause difficulty in claiming infringement.
  • Territorial nature of patents versus global nature of technology: Patents are territorial, and in a complex AI ecosystem it cannot be assumed that every step of the process will be carried out by the same entity, or in the same country – servers can be located anywhere. Therefore, one must assess whether the proposed invention is capable of being enforced on a contributory infringement basis (ie that the infringer provides the means essential to put the invention into effect).

Patents in this field should be drafted with an eye to these issues, in order that claims are prosecuted with the best prospect of being enforced.

Are there better or alternative defensive strategies?

Whether alternative defensive strategies are appropriate depends upon overall commercial strategy, and the nature of the business value derived from the technology. Possible alternatives include:

  • Non-monopoly IP rights: these will prevent a third party from copying – such as relying on copyright in source code, or database rights in machine learning data. The benefit of these rights is they arise automatically, with no obligation to disclose an invention.
  • Defensive publication: publishing an invention without patenting it is a cost-effective method of mitigating freedom to operate risk. Defensive publications can prevent third parties from later being able to patent the invention or be used defensively as prior art to invalidate a later patent.
  • Trade secrets: if the inventive technology is not capable of being easily reverse engineered, it may be viable to rely on breach of confidence, trade secrets and non-disclosure agreements to protect the invention. This might be worth considering where the commercial life of the invention will outlast patent protection.

Interviewed by Marie-Gabrielle Williams.