Hammer or screwdriver?
“Why choose hammer if you can hit the nail with a screwdriver.”
That sentence came out of the famous GPT-2 text-generation neural network, the predecessor GPT-3, when I gave it the seed “Why choose hammer…”. AIs like GPT-2 might not yet pass the Turing test, but the sentence it generated reveals several important aspects of AI and AI tools.
First, well designed and trained neural networks can do incredible things. The sentence above is perfect English, it is internally coherent and even has some humour in it. That is what GPT-2 was trained to do: to produce text that resembles what a human being would produce.
Second, focusing on the semantic substance of the sentence, the problem defines the tool of choice. If the purpose is to get the nail banged in, it really matters which tool you are carrying.
Third, every now and then it is useful to question your current tools and methods and think of the alternatives. Would the screwdriver indeed work for this purpose? Or, a revolutionary thought: should I consider replacing the nail with a screw to make less noise, improve adherence and even be able to use an electric drill to get it done faster?
Nailing it or screwing it up?
Many patent related tasks are data- and labour-intensive, making them seemingly difficult to automate. However, when you dig deeper into them, you often find some repetition either on the procedural level or substance level. Both levels can in many cases be facilitated by modern AI. The following questions may help you to identify the spots where automation can help either today or in the future:
A general trend seems to be that IP professionals need to achieve more with less resources. A proper tool can definitely help in that. No tool does it all, but for many areas there are solutions available and they often have more comprehensive capabilities than you might think. A rule of thumb is that if a process at least occasionally feels repetitive and tedious, it might be something that a machine can learn too. When you decide to invest in an AI solution, the transition will typically happen gradually as the users learn together with the machine.
At IPRally we are dealing with one of the most data-intensive, time-consuming and intellectually challenging tasks: patent searching and analysis for prior art and FTO purposes. Our method is so-called Graph AI, designed the solve the patent search problem in the most efficient way. That applies to practically all AI solutions: they are designed to solve a single problem and you need to know their strengths and limits. We’ll discuss this aspect in the sequels of this blog series.
However, although we have a pretty strict focus, we have seen that there are consequences that go further. For example, as a “side product”, since the AI understands the technical content of patent well, we have built an efficient patent classifier. Also, even though we can bring immediate time and cost saving only to the search stage, the indirect financial and business impacts over the years to follow may be huge. Dozens of saved office action rounds, a killer reference against a competitor’s patent, smoother and more enjoyable every-day workflows.
For most of us, the above is definitely something to strive for. However, there are pitfalls and too many examples of failed tooling projects. In the next articles in this series, we will focus on some of the aspects that make a tool evaluation and acquisition process more likely to succeed.
Is your organization willing to be in the IPR frontline? Get in touch to get a demo or take a sneak peek of the future of patent AI as we see it.