AI for Patent Attorneys Q&A

00:00:00: In this video, I'm going to answer the three questions every patent attorney is asking about AI right now.

00:00:06: So today is Tuesday.

00:00:07: Yesterday, I was speaking at the search and examination matters, which is one of the flagship conferences organized by the European Patent Office.

00:00:16: We had over eight hundred participants watching live, which was really amazing, but we didn't have time to take questions.

00:00:23: So After the conference day yesterday, I looked through the chat logs and now I'm going to answer the three most common questions that came up.

00:00:32: The first question really flooded the chat yesterday.

00:00:35: It basically boils down to this.

00:00:37: If I put my client's invention details into a cloud-based AI tool, am I basically leaking my client's trade secrets?

00:00:44: And the answer is it depends on which kind of cloud you are using.

00:00:49: You need to understand the three tiers of AI security or you are probably walking into a malpractice trap.

00:00:55: Tier one is consumer AI.

00:00:57: This is, you know, the three versions of JetGPT, Gemini, Claude, DeepSeq and so on.

00:01:03: If you paste your client data here, you are paying with your clients data.

00:01:08: These models typically train on your inputs.

00:01:11: So if you use this for client work, you are basically gambling with your license.

00:01:15: just don't do it.

00:01:16: Tier two is enterprise AI.

00:01:18: This is where you either pay for some kind of enterprise plan for your general purpose chatbot like chatGPT or you buy a license to one of the commercial patent drafting tools out there which then in turn typically feed the data into one of the big generative AI providers.

00:01:38: And here you typically have a contract that says that they will not train on your data.

00:01:44: This is safer.

00:01:46: But the data still leaves your building, right?

00:01:49: It probably sits on a server in the United States.

00:01:52: As a patent attorney working with confidential client data, you have to figure out for yourself which of these architectures is best for you and which meets your client's expectations also, of course.

00:02:04: If you're a European patent attorney like me, there is also specific guidelines on the use of generative AI for patent attorneys published by the EPI.

00:02:15: I'll put a link into the description.

00:02:16: The third tier is sovereign or local AI.

00:02:20: This for me is really the gold standard because this is where the AI runs entirely on your own computer or within your firm's internal network and no data is leaking into the internet.

00:02:31: I actually made a whole separate video about how to install a local AI on your computer and I'll link that up somewhere here.

00:02:39: And by the way, if you are unsure which of these architectures is best for you and your patent law firm, I have something for you.

00:02:48: If you need a safe AI deployment plan that really specifically fits your circumstances, I'm doing one-on-one coachings just for that.

00:02:57: Link in the description.

00:02:58: The next big question was all about trust.

00:03:01: So if we, for example, trust a calculator because, you know, two plus two always equals four, why can't we trust AI in the same way?

00:03:11: And this is one of the most dangerous misconceptions in our industry in my view.

00:03:15: A calculator is deterministic, right?

00:03:17: So it follows rigid rules, logic rules.

00:03:20: That's why you can generally trust that the calculator will deliver the correct result.

00:03:25: AI on the other hand is probabilistic.

00:03:28: It's not calculating the truth.

00:03:30: In fact, it's not concerned at all with the truth.

00:03:32: All AI does at the most fundamental level is predict the statistically most likely next word.

00:03:40: That's it.

00:03:40: So, and when you treat a probability machine like a calculator, you become what I would call an intellectual tourist easily.

00:03:47: You know, you scan the file, you press a button, the AI generates a thirty pages of very fluent text and you just nod along and say, looks good, I'll sign it.

00:03:59: And this in turn leads to a psychological gap which is called, in the research, satisfying.

00:04:05: I've talked more on that in a video, I will link somewhere here.

00:04:09: Satisfying is basically the psychological effect that occurs when you accept a text just because it sounds plausible and that's the case with AI.

00:04:19: But because the AI is just predicting the most statistically likely.

00:04:23: next word, you end up with a statistically average patent if you don't pay close attention.

00:04:29: And we all know that patents are not about average, they are about the outlier, you know, about protecting inventions.

00:04:37: So in the end, it's all about using AI responsibly in a way that makes you, the patent attorney, a better thinker, and it doesn't make you think less.

00:04:47: And that brings me to the third question.

00:04:49: Many people were asking, what if the AI hallucinates?

00:04:52: And we all know it by now, right?

00:04:54: AI sometimes make up non-existent prior art documents, or it makes up an entire board of appeal decision.

00:05:01: In the end, for me the solution is simple.

00:05:04: You never use AI to outsource fact finding and you never use AI to outsource decision making.

00:05:11: You use it only as a scribe to produce text and as a provocateur.

00:05:15: to challenge your own thinking.

00:05:18: And that's actually a very powerful use case for AI, especially, you know, with the reasoning models we have nowadays.

00:05:25: So for example, every time when I've drafted a set of claims for one of my clients' inventions, I always let my AI take on the role of the devil's advocate, right?

00:05:37: So I tell the AI to be a very, very strict EPO examiner and to raise clarity objections.

00:05:43: Most of the time the first of objections contains at least one angle which I hadn't had on my radar.

00:05:51: I then revise and fine tune my claims and then I repeat the cycle until the objections from the AI get kind of silly.

00:05:59: That is hallucinated.

00:06:01: And that's fine because I'm in the driver's seat.

00:06:04: I'm the expert who judges in the end what to do with those objections.

00:06:09: Now, if you also want to use AI as a tool for thought that sharpens your own thinking, comes to my seminar.

00:06:15: It's a three hour online session exclusively for patent attorneys and I'll teach you basically everything I know from roughly three years of intensive playing with AI tools and setting up and perfecting my own AI toolbox.

00:06:30: Link in the description.

00:06:31: So my takeaway from search and examination matters, twenty twenty six was clear.

00:06:36: AI assists.

00:06:37: humans decide.

00:06:39: The tools are getting sharper and that means our judgment needs to get even stronger.

00:06:44: Now I want to hear from you.

00:06:45: So what's the number one burning question about AI and patent law that you still haven't gotten a straight answer to?

00:06:53: Drop it in the comments below.

00:06:55: I read every single comment and I might just pick your question for the next Q&A video.

00:07:00: Thanks for watching.

00:07:01: Stay sharp.