Aon’s David Andrews explains how machine learning and generative AI can reduce legal costs and accurately predict patent case outcomes.
Of the many areas set to be transformed by AI, intellectual property (IP) is one that could be impacted in both positive and negative ways.
The rise of generative AI systems such as text-to-image generators have raised questions around copyright, such as who truly owns the created images and if these systems are copying the style of certain artists.
Advanced AI models like ChatGPT have been shown to be able to write its own books and plays from simple prompts. But while the chatbot doesn’t own these works under EU law, the question of who truly owns AI-generated works will likely remain a hot topic for the future.
While there are examples of AI creating confusion around IP law, people like David Andrews believe these systems have the potential to transform the landscape for IP law firms.
David Andrews is Aon’s chief data and analytics officer and has previously represented clients in “high-stakes patent litigation”. He leads an Aon team of data scientists and engineers to answer IP valuation and quality questions.
Speaking to SiliconRepublic.com, Andrews said machine learning – where a machine learns and replicates human actions – and natural language processing (NLP) has the potential to cut costs and save time in the field of IP law.
“We use [deep learning] techniques to value IP assets,” Andrews said. “By combining IP data and historical valuation data, we develop a deep understanding of an IP holding’s quality and breadth.”
“Now, many patent holders can benefit from large-scale patent analysis originally too costly with human effort.”
Using data troves to predict outcomes
Andrews explained that patent law is an area that has “hundreds of years of human data recorded in a semi-regular format”, in the form of patent examinations, litigation cases and assignment records.
He said that by examining these data troves, behaviour patterns can be learned and applied to AI systems.
“Machine learning helps us understand how complete the initial examination is and how likely a patent is to survive litigation, and by using NLP, we can start to understand the text of the patent claims,” Andrews said.
By combining these systems together, Andrews said predictions can be made on which patents will survive an “invalidity challenge”. These techniques can also be expanded to other forms of invalidity to determine “the likelihood of infringement” and help legal decision-making.
“Together, machine learning and NLP can help identify valuable or ineffective patents or help inform which court cases should be settled early and which should be fought,” Andrews said.
The same techniques used in patent law could also be applied to other IP examples, such as copyright, trademark, and trade secret law. While these IP areas have different characteristics, Andrews said the “fundamental technique remains the same”.
Andrews expects firms to gain “competitive advantages” through the adoption of machine learning, which will eventually lead to “rapid adoption across the industry”.
“As the ability to find relevant IP increases and quality measures improve, buy-versus-build IP decisions become more precise,” Andrews said. “This should further increase liquidity in the IP marketplace, making IP-backed lending more desirable because IP assets can be sold more quickly in case of default.”
Cutting legal costs
Andrews said generative AI systems present both a “threat and an asset” to IP, with the benefit being the ability to speeding up the legal process.
“One of the barriers to protecting legal rights is the cost associated with the legal process and as IP attorneys adopt generative AI tools, the cost of IP portfolio maintenance can be reduced,” Andrews said.
Generative AI systems have been adopted further into the legal sector in recent months. OpenAI – the creator company of ChatGPT and GPT-4 – has been looking at this market through its investment into Harvey, a start-up developing AI for law firms.
This start-up is using GPT-4 to develop law-focused AI models and has attracted some large companies with its technology, including a partnership with PwC in March.
Andrews said that over time, the large context these generative AI models bring will allow people to automatically search for infringing works, determine an IP’s uniqueness and “more accurately predict the outcome of IP lawsuits”.
“As IP matters become more transparent, valuing them becomes more precise and repeatable, and as values become more agreed upon, market liquidity increases,” Andrews said. “Large learning models such as GPT-4 will be key to unlocking even more value in IP by making the assets more liquid.”
10 things you need to know direct to your inbox every weekday. Sign up for the Daily Brief, Silicon Republic’s digest of essential sci-tech news.