The EU AI Act came into force on 1 August 2024, a new dawn for AI regulation. For those using or developing cutting edge analytics solutions, it is easy to assume that they are “low risk” and therefore free of regulation. That would be shortsighted. The responsibility to ensure ethical deployment sits at the epicenter of the responsibilities of analytics providers. This is not some lofty and philosophical nirvana, but the rigorous adherence to guidelines that build trust and confidence with responsible AI.
RELX, the parent company of LexisNexis, has had strict guidelines around responsible AI and illustrates where the bar needs to be set:
Source: RELX Responsible Artificial Intelligence Principles, June 2022
Let’s take each one in turn:
Real world impact: the primary purpose of analytics is to support decision making, with curated and trusted data. Analytics developers need to go beyond asking what is possible and how much AI can be included to the heart of WHY. Does the solution solve a real-world problem – and that means listening and understanding the nature of the customers’ problems.
Bias: the mainstream trigger for bias is analytics that deliver less favorable outcomes for individuals or groups based on gender, ethnicity, socio-economic status, and other personal attributes. But that’s too limited. Much of today’s AI in patent analytics is powered by supervised machine learning or generative AI – both of which are enabled with training data (and for genAI vast quantities of it). Consider a classification algorithm with a really high level of precision and recall. Now bring along a human who trains it by reference to patents from only a small set of owners or a limited number of geographies. Expect unwanted bias in those directions. Imagine a patent drafting tool whose training set predated Covid. Expect misleading results hampered by limited vision. These are the issues that keep the analytics providers up at night. Avoid all imposters who don’t have a coherent explanation of their strategy to avoid bias.
Explainability: in the days where Boolean algebra was the dominant approach to patent search, life was easy. Results included patents that included the word (or combination of words) or codes e.g. CPC/IPC, entered by the user. The results were often woefully inadequate but, on the plus side, they were explainable.
In the early days of AI, critics complained that AI was a “black box” and to be dismissed on that basis alone. In the era of genAI, this level of fear, uncertainty and doubt has largely gone away (and not because genAI is explainable!). But that does not lower the need for transparency. Providers should be willing to disclose how their systems operate and where in the stack AI is doing the work.
LexisNexis Classification, formerly known as Cipher, one of the earliest of the AI classification platforms, undertook independent testing and submitted a peer reviewed paper for evidence. At the very least, customers should not have to struggle to understand what’s going on. Perhaps most importantly, transparent, and explainable tests must be available to validate accuracy. Customers care about knowing how a system operates when the solution is better, faster, cheaper than the next best alternative.
Oversight and accountability: there is no shortage of media coverage around the satnav user who drives into the sea having failed to observe the sign for the ferry crossing. Or the hapless lawyer who submits raw ChatGPT output with phantom citations in a court brief. What’s critical is getting the balance right between those who provide analytics and the users. It is early days for genAI, and this means that providers must go the extra mile to communicate and train customers on capabilities. Remember, genAI is trained to deliver plausible outcomes. In anthropomorphic terms, a genAI powered system is trained to give an answer even when it doesn’t have one. It doesn’t like to disappoint. Once again, this ability to “hallucinate” should not deter engagement. Think of AI as assisted or augmented intelligence – or perhaps not intelligence at all.
Privacy and data governance: many users of analytics in the field of intellectual property are lawyers. This means that towering above the principles under discussion is an obligation not to infringe the rights of others. This needs recognition by vendors who serve this market. This is difficult when the foundations on which large language models are based are being challenged on copyright and other fronts. This legitimate concern can create an impediment to adoption. It has to be countered with legal frameworks developed to support the delivery of these tools.
Over the last decade, AI has had an enormous impact on patent analytics. Patent owners and their advisers report measurable impact across a range of business decisions. This includes strategic portfolio optimization, licensing and litigation. Responsible AI is a guiding light for all providers of analytics. Here are the learnings gleaned from customer stories:
Teamwork: positioning responsible AI as purely a supplier responsibility diminishes the responsibility that sits on the shoulders of the user. There is a pressure for those handling patents to be innovative and to achieve more with less. Delivering on these goals is not risk free – and that’s OK. What is required is multi-stakeholder engagement, including others across the legal, R&D and finance groups. This enables IP to learn from the experiences in other areas of the business. Business intelligence is a near universal fuel for almost all evidence-based decisions.
The Iron Triangle: when looking for new approaches, it is common to seek better, faster and cheaper solutions, and compromising with two of the three. The skill is having the ability to compromise at the right time and in the right place. Take the simplest example of a patent search that cost $5k and takes a week using offshore resource achieving 90% accuracy. Now compare that with AI landscaping that delivers the 80% accuracy in under 5 minutes. And this is at a fraction of the cost). The decision on whether to adopt this solution with depend on the specific circumstances – and not on some misguided quest for perfection.
Take the inside track: the market is chaotic, with new solutions for every problem. By focusing on the solutions that will deliver the most value (an entirely subjective assessment), this will typically suggest a finite number of promising solutions. Some will be ripe for adoption, others will be developing nicely. In both cases, secure a position inside the tent. Analytics providers working at the cutting edge of AI welcome industry support. Providers who think they have all the answers typically don’t.
There is significant concern over the widespread adoption of AI, and this has led to the rise of regulation around the globe. However, in times of rapid technological progress, self-regulation is often the best path and to be encouraged. This means that organizations must develop, maintain and communicate ethics and responsibility principles that have the clarity to create confidence while at the same time being flexible enough to accommodate the constantly evolving world of AI.
Nigel Swycher is co-founder and CEO of LexisNexis Cipher – His background is in law, where he led the IP practice at leading law firm Slaughter and May.
Nigel has been recognized for many years by the IAM 300 as a leader in the field of IP strategy.
Members of the PatentSight+ AI-Insider program will lead the advancement of artificial intelligence in intellectual property and help revolutionize the way high-value strategic insights can be gained from AI-powered IP search and analytics.
