ChatGPT – The Future IP Advisor?
What word comes next: Happy Birthday to ____?
For most readers, the answer that springs to mind will be “You”. But why did you think that was the answer?
You may have been fantastically lucky and just picked the word “you” at random from the 170,000+ words currently in use in the English language (in which case today might be a good day to buy a lottery ticket!). But it is much more likely you picked the word “you” because you’ve heard the phrase “Happy Birthday to You” many times before. You’ve probably sung it (thrice repeated each time!) to many others, and in return had it sung to you. You’ll have seen it written down on birthday cards and cakes, banners and balloons, and all manner of other sources. Statistically speaking, you know that the phrase “Happy Birthday to” has a much higher probability of being followed by “you” than by any other word, even if you limit only to phrases that make grammatical sense.
In more technical terminology, in answering “you” to the question above, you are in essence applying a large language model. In simple terms, a language model assesses a corpus (or group) of text, and then calculates a probability distribution which assesses the likelihood of a given sequence of words existing in that corpus. Applying this method to the question above, one could consider assessing the corpus of every sentence you’ve ever heard, read or spoken; producing a subset of every phrase beginning “Happy Birthday to”; and then ranking all such phrases by their occurrence. Statistically, “Happy Birthday to You” will (most likely) vastly outrank all other potential candidates.
This may all seem rather esoteric, but it is the essence of the technology behind ChatGPT-3, which has garnered much attention since being launched by OpenAI in November 2022. And should anyone think this is just the latest internet fad to be quickly forgotten, OpenAI has been recently valued by some analysts at almost USD 30 billion, due at least in part to the potential offered by tools such as ChatGPT-3.
ChatGPT (the GPT stands for Generative Pre-Trained Transformer) applies large language models to allow it to respond to user input in a highly articulate manner. So if you ask ChatGPT-3 a question such as “What is the tallest mountain in the world”, ChatGPT-3 will confidently state “The tallest mountain in the world is Mount Everest. It is located in the Himalayas on the border of Nepal and Tibet, and its summit stands at 29,029 feet (8,848 meters) above sea level….”, before providing further context.
ChatGPT-3’s ability to produce detailed responses to complex queries has led many in academia to foresee the decline of the essay as a meaningful means of testing knowledge. After all, why bother researching a topic if you can just type in questions such as “Describe the key features of Cicero’s rhetorical style” or “Compare and contrast the chemistry of the elements sodium and potassium” and immediately obtain a discursive essay on the subject?
Outside academic considerations, uses for technology such as ChatGPT-3 are becoming apparent in many fields. Copy writing and advertising, medical writing, journalism, law, and many other “white collar” jobs will all need to take account of such technologies, and inevitably some professions will fall away while new roles emerge.
So what about the impacts of technologies like ChatGPT-3 in the field of patent law? Well, perhaps the best way to answer this is to ask ChatGPT-3 itself. Here’s what it answered when posed that question:
Chat GPT-3, a state-of-the-art language model, has the potential to revolutionize the field of patent law. This AI system is able to generate text that is similar to human-written text, and it has the ability to assist in various aspects of the patent process.
One potential use of chat GPT-3 in patent law is in the area of patent drafting and prosecution. This process can be time-consuming and complex, but chat GPT-3 could be trained on a large dataset of existing patent applications and used to generate draft applications. This could save time and resources for patent practitioners and applicants, as they would not need to spend as much time drafting and reviewing the application.
Another potential use of chat GPT-3 in patent law is in the area of patent litigation. Reviewing large volumes of technical documents is necessary in patent litigation, but it can be a tedious process. Chat GPT-3 has the potential to assist in this process by generating summaries or analyses of technical documents, which could save time and resources for attorneys and technical experts.
Additionally, chat GPT-3 may have potential applications in the area of patent licensing and transactions. It could be used to generate draft agreements or to assist in the negotiation process by generating alternative language or provisions.
Overall, chat GPT-3 has the potential to greatly impact the field of patent law by streamlining and automating various aspects of the patent process. It could help reduce the time and resources needed to draft, prosecute, and review patents, as well as assist in patent litigation and licensing transactions.
So are the days of the patent attorney numbered?
Well, perhaps not just yet. ChatGPT-3 can certainly produce a sensible, if unimaginative, claims set for a simple “invention” (a request for it to produce a claims set for “an invention which is the use of aspirin to treat migraine” led to a technically sensible claims set with method of treatment claims, pharmaceutical composition claims and kit claims). However, a patentable invention is necessarily new and non-obvious in view of the state of the art, defined as “everything made available to the public by means of a written or oral description, by use, or in any other way, before the date of filing…”. But it is this very same “state of the art” on which tools like ChatGPT-3 are trained.
This limitation means it is at least arguable that tools trained on existing corpuses of work, like ChatGPT-3, are inherently incapable of producing anything inventive – after all, if the means of generating the text is based on prediction from the prior art, how can an “inventive step” occur? At least at present, it seems that the built-in reliance on the existing works means that ChatGPT-3 can only play the role of the notional “skilled person” – a skilled practitioner in the relevant field who has average knowledge and ability; who is aware of what is common general knowledge in the art and is also presumed to have had access to everything in the "state of the art", but who is devoid of inventive capacity.
This can be readily seen when ChatGPT-3 is asked if it is inventive to use aspirin to treat migraine. The entirely reasonable answer is “The use of aspirin to treat migraine is not considered to be an inventive concept because it has been known and used for this purpose for a long time…”. But when ChatGPT-3 is asked about the use of aspirin in treating conditions such as erectile dysfunction or colour-blindness (which could be inventive if proven), ChatGPT-3 states, in essence, that such use is not inventive because that effect is not scientifically known – which would be the entire essence of the invention!
For now, tools such as ChatGPT-3 and its successors will provide ways to increasingly efficiently analyse text and generate boiler plate. But when it comes to defining a patentable invention, such tools are inherently limited by their reliance on the art. In the words of ChatGPT-3 itself:
It is possible for chat-gpt3 to generate text that describes an invention, but it is important to note that the model is not capable of creating new ideas or concepts on its own. It can only generate text based on the information and examples it has been trained on.
Given this, the role of the inventor and the patent attorney seems assured – for now at least.