The arrival center stage of Chat GPT and generative AI has generated a great deal of excitement in many areas of economic activity.
The world of intellectual property is particularly feeling this tension on one hand there is a strong sense that generative AI has a vast potential to streamline many tasks in the IP world, including prior art searching, infringement detection, preparing responses to official actions and even supporting patent drafting. On the other hand, the sensitive and confidential nature of the information being handled, and the inherently opaque nature of how machine learning systems use data gives rise to a great deal of caution. During my participation in the American Intellectual Property Law Association (AIPLA) annual conference this week, I have attempted to gauge the overall mood with respect to these tools in IP, and offer the following observations.
End clients are concerned about the possibility of direct exposure of their data, the possible injection of third party material into their applications, and the more subtle issue of confidential information being used in training models, which might in some way orient or improve the results obtained by competitors from the same tool.
The major AI solution providers are aware of these issues, as reported during the conference. Two classes of solutions are advancing in parallel. Firstly, the negotiations by large IT players with the AI solution providers over the last eight months have focused on providing a contractual response to these concerns. Clauses relating to the confidentiality of information, limits on the service providers freedom to retain data, the permissibility of the use of submitted information in model training, have all been hammered out. On this basis service providers have developed standard license clauses addressing each of these issues which may now be available to other commercial users (that is, providers of commercial software solutions building on a LLM for back end functionality on a paid basis) as a matter of course. Mechanisms for the verification of information to rule out “hallucinations” for example in case law citations are available. Commercial users are also able to offer indemnity for any breach of these requirements.
Commercial offerings incorporating some or all of these measures are already on the market, although there is little evidence of whole-hearted adoption of these offerings so far. IP professionals are well aware of the cautionary tale of the MATA v AVIANCA affidavit, and wary of getting their fingers burned in a similar way. The general feeling in the IP profession seems to be that while many agree that it is important not to miss the AI tools bus, so that firms are keen to evaluate current solutions, they are still far from feeling able to integrate them into their everyday workflow.
Although in principle these measures should go a very long way in addressing end client’s concerns, IP practitioners remain very circumspect. It seems that while there is a consensus that such tools are the way of the future, there remains a substantial barrier to be cleared before practitioners feel comfortable to fully embrace them. It remains clear that whatever solutions might be envisaged, any possibility of integration in a manner transparent to the end client is a long way off for the foreseeable future, even assuming an optimal contractual situation, use of such tools may generally require informed consent from the client.
Meanwhile, technical solutions are also advancing. While the main solution providers are generally built on the model of an AI service operating in the cloud, the perspective of a private, local AI implementation offering comparable performance is approaching fast. GPT4ALL or Meta’s Llama technology in principle support such implementations, although the technical barrier to entry remains potentially dissuasive in comparison to the extraordinary accessibility of ChatGPT in particular. Nevertheless, such solutions will address many of the major concerns of end clients, and off the shelf solutions meeting the needs of IP professionals are no doubt under development.
There is no doubt that the field of ML based professional solutions is advancing at an eye watering pace, and that we are living through a period of transition. The stumbling blocks that have slowed adaption are known to the major players and solutions, both legal and technical, are maturing. It seems very likely that within the next 12 months such tools will become a key tool in the toolchest of many IP professionals, but there is no suggestion at this stage of a critical watershed in adoption.