Modern AI technology, which has become increasingly popular in recent years, is based on deep learning and machine learning methodologies. These employ massive computing power to process thousands of sample input-output pairs in order to train adaptable data structure models. When provided with an nth + 1 input, they eventually generate their own correct outputs. These can be viewed as questions and answers. If an AI model is given 10,000 sample questions with correct answers, it will be able to answer the 10,001st question correctly on its own. After training, computing requirements are minimal.
AI is appropriate for situations involving repetitive decision-making processes due to the nature of the methodology. For one thing, during the training, numerous existing examples of correct decisions must be available. Following the training stage, a system is repeatedly applied to similar situations. As a result, the application space for AI is sometimes exaggerated. Once understood, however, this limitation prompts our attention to situations of decision-making that can be replaced by automation or made more efficient using AI.
Using AI for Patent Portfolio Management
If we think about patent portfolio management in contexts of constituent decision-making processes, we might be able to figure out which ones are suitable for AI application. Full portfolio assessments, forecasts of the potential value of specific patents, identification of weaker technical areas in a portfolio, identification of patents to cull, evaluation of external portfolios, correlations of all such portfolios to internal portfolios, the decision to enforce IP with respect to external organizations, acquisition of external portfolios, or sale of a portion of a portfolio are standard tasks in patent portfolio management. Which of these processes is most likely to benefit from artificial intelligence?
Understanding the applicability of AI to repetitive processes that require huge amounts of pre-existing sample inputs for training allows us to significantly narrow the field of action. Strategic decisions at the portfolio level, such as the decision to enforce IP with a specific competitor, are poor candidates because they occur infrequently and are highly context sensitive. Processes with a small number of practically achievable input samples are notoriously difficult to automate.
However, decisions operating at the level of individual patents, where tens of thousands of individual patents may be available as inputs, would be good candidates. Examples include determining which patents to cull, which patents have conceivably high value, and automatically determining a patent’s technology category. It should be noted that properly applied automation at the individual patent level can eventually help decision-making at higher strategic levels. But, by this time the decision-making is still made by humans, albeit centred on a richer data set offered by AI.
Training Samples and Experts in the Subject matter
Another practical limitation influences application selection: the generation of training samples. Vast numbers of appropriately formatted input samples are required for training the AI model. The mere presence of previously made decisions of a particular type is insufficient. They must be made available to the training department, properly vetted (i.e., thought to be correct), and configured for input. In the scenario of patent decisions, these are almost always made by highly trained technical staff. To feed samples into AI for the patent decision-making process, it’s necessary to have access to these technical experts. This is done during the sample-generation stage. In practice, this implies that subject matter experts are asked to participate in an evaluation activity. But, with targets chosen to produce the most effective training samples possible.
Once a sufficient number of decisions have been generated by subject matter experts, they can be used to train AI. Following the compute-intensive training phase, the AI is ready to answer additional samples quickly and without human intervention.
Decades of knowledge, as in the minds of the subject-matter experts, are distilled into a computerized process. This could significantly accelerate decision-making in patent portfolio management.