GPT in general and the legal industry in particular: Report from an interesting roundtable.

Yesterday at noon (local time), a roundtable on "Beyond ChatGPT: What Generative AI Actually Means for Law Firms, In-House, Legal Tech and More" was held in the US. Legaltech News Editor-in-Chief Stephanie Wilkins welcomed an illustrious crowd of guests.

Without a doubt, the surprise hit came from Pablo Arredondo, Co-Founder and CIO of Casetext. Not only did he announce the simultaneous live launch of GPT-4, but also of "CoCounsel", an application from Casetext that already relies on OpenAI's brand new language model. He raved about the new GPT version, predicting that it would be simply impossible for law firms not to use the new technology because it was so much better than humans (!) To be fair, he added that he had analyzed his own litigation files with CoCounsel, which alerted him to highly personal errors and gaps in reasoning.

Casey Flaherty, Co-Founder and Chief Strategy Officer of LexFusion, initially tried to put things into perspective: GPT is ultimately just one of many available language models that has the hype on its side. However, he also emphasized that the threshold to market maturity had been overcome; if ChatGPT was the trailer, the movies would now follow. His advice: "This is happening now; it will be done by you or to you." 

It was also exciting to hear that Flaherty, unlike the other panelists, very much expects technology to partially replace lawyers.

Darth Vaughn, Litigation Counsel and Legal Innovation and Technology Operations Manager at Ford Motor Company, took a different view: he, too, is excited about new technology, but stressed that at all times domain knowledge must remain available in the organization to control the new tools.

A large indirect impact on advocacy was seen by Jae Um, Founder and Executive Director at Six Parsecs. She, too, sees LargeLanguageModels as indispensable in law firm practice. However, she emphasizes that the use of ultra-fast high technology means the end of traditional billing based on time and material. The law firms are urgently required to try out alternative remuneration models, whereby it will probably come down to trial&error.

In summary, a common ground can be found in the fact that from the point of view of US commercial law firms and providers, the use of state-of-the-art technology is indispensable. Which language models will ultimately prevail cannot yet be judged. I have not heard of any reservations about training models of foreign providers with own law firm content (=client data). And whether this constraint is transferable 1:1 to the continental European market with its completely different legal system could not be addressed in this intra-American format.

ChatGPT and Academia - a tense relationship

After the first two days of the IRI§23 conference were primarily about practical usability, as far as the respective streams were related to ChatGPT, the focus on Saturday was on academic topics.

The first was the question of the protectability of language models and their results under copyright law. The answer to the latter is probably complex enough, there would have been no real need for the appeal for a general abolition of copyright in the field of science.

Highly interesting, however, were the reflections on the subject of teaching and on the subject of citation. With regard to teaching, the thesis was put forward without contradiction that the age of "academic homework" is finally over. No teacher can be expected to take responsibility for assignments that are actually written by highly developed chatbots.

And the scientific citation, is it also endangered? - I mean, yes certainly. How should a citation be composed if the result is not repeatable? After all, here there are massive differences in the new technologies compared to the common practice of citing web pages with an exact date of retrieval. But I also wonder if the question is really that prominent. Will there be legions of scientists using ChatGPT to write their papers? If so, then only in marginal areas, and there a copy&paste from Wikipedia should be sufficient. With appropriate reference.

Conclusion: Some questions can be solved, others only in the more distant future. It may be discussed further on academic ground.

ChatGPT - Revolution or bubble?

At "IRI§23", the International Legal Informatics Symposium in Salzburg, ChatGPT was the star. The deserving organizers had not quite expected this, but the number of visitors to the corresponding streams was high.

But what exactly is so special about this new technology?

• The ability to correctly interp'ret human language, say some.
• The ability to even write texts in a flawless form, the others.
• ChatGPT revolutionizes database searching, according to some; or not.
• It is a system that should be used only by experts because of its error-proneness - or, according to another opinion, only by non-professionals.

And so it went on, and every opinion was well-founded! There was even disagreement on the question of whether ChatGPT would tsunami-like overrun the legal world or whether it was just a risky bubble.

At least concerning the last question I have a clear opinion: ChatGPT (and the epigones) are too big to fail. The billion dollar investment leaves no other development. And after all, never in history has a new technology found so many users so quickly.

Conversational AI is making history, but no one knows exactly what form it will take.

Allowed to be wrong sometimes

For years now, the question has been discussed again and again in relevant formats whether lawyers will need programming skills in the future. I have always denied this - to acquire and maintain good legal knowledge is enough to fill an evening, there is no room for another discipline (apart from exceptions).

Today I say: that was too short-sighted. Just as marketing professionals need to have a deep understanding of the processes around digital advertising, lawyers of the future need to have a good understanding of the relevant digital technologies.

They don't have to be the ones writing lines of code for e-discovery suites. But they do need to be able to assess what to expect from the various technologies that are usually grouped under the umbrella term "artificial intelligence." Lawyers and notaries, corporate lawyers, judges, prosecutors and administrative lawyers - all of them will, in the course of their very own work, be facing the question of the use of technology for certain process or procedural steps and will have to assess for themselves what they can expect from it and what the risks are. By this I don't mean the eternal data protection issues, but issues such as bias, training sets, confidentiality, and most especially error calculus.

This also creates a whole new set of challenges for legal academia, but even more so for continuing legal education. For a long time, lawyers have seen time constraints as their central problem. The tension will only be resolved if ... but that is another story.

"The 'robojudge' is technologically impossible."

With this clear statement, the Executive Director of the Center for Legal Technology and Data Science at Bucerius Law School in Hamburg, Dirk Hartung, tried to direct the discussions about the digitalization of law into more realistic channels.  "We tend to be concerned with the social impact of technologies, the use of which is far beyond any meaningful period of consideration," Hartung said.

The occasion for his lecture was the 30th EDV-Gerichtstag (EDP Court Day), which for well-known reasons was again held virtually rather than in Saarbücken.  Prior to this, Florian Matthes, professor of computer science at the Technical University of Munich, had given a highly interesting talk in which he screened the main technologies that are currently shaping the development of legal informatics. Matthes did not hesitate to address controversial topics. When he describes Natural Language Processing (NLP) for legal texts as an essential field of research, hardly anyone will disagree with him; his appeal to rely (again) more on rule-based expert systems, on the other hand, is likely to meet with headwind.

The topic of NLP also reveals a topic that ran through the entire event. One might agree that "computer scientists and lawyers must become friends" (Maximilian Herberger) if one is concerned about the digitalization of law. However, the very mention of "understanding" legal texts as a link between the processing of existing texts and the generation of new ones triggers fundamentally different associations in lawyers and technicians: the linguistic-logical understanding of a legal issue literally collides with the mathematical approximation analysis that is meant by "understanding" from a technical point of view.

Also of direct interest to Austria is the issue of anonymizing court decisions as a prerequisite for their publication. Work on automating this labor-intensive process has been going on for some time in both Germany and Austria. It is interesting to note a thesis of the judge Isabelle Biallaß: Anonymization not only has to be humanly controlled, it also has to be scaled, depending on whether the anonymized text is to be made available to science or industry.

That could still take some time.

Who is being harmed by bad campaigning?

Last week, I reported on this space about the sale of Kira to Litera and the intention to release a standalone product for corporate lawyers under the Zuva brand. The media release emphasized that Kira technology would continue to be operated and developed under the new umbrella.

Somewhat surprisingly, UK vendor Luminance responded with an aggressive campaign urging existing Kira customers in the legal sector to switch, portraying the Kira-sell in a distinctly different, negative light. One could read from the Luminance press release that it was a child disposal, the fact that the existing management would perform a strategic advisory function was left unmentioned.

Several media are now asking what drove Luminance to this activity, and who actually benefits from bad campaigning. On the one hand, this is coffee-bag reading, because without knowing the reactions of the customers, one will not come to a conclusive answer. In any case, the commenters seem to assume that Luminance itself will come out of this damaged. However, I think that the issue shows the enormous pressure that even established legaltech companies are under. Is it the concern of falling behind in development resources? Is it fear that their own customers will switch to Litera's comprehensive software solution because they appreciate the all-in-one concept? - Everything is possible

Markets on the move

As reported by alm, Legaltech group Litera is acquiring industry leader Kira, effective Sept. 1 of this year. The product and brand are to be continued, and at the same time Litera is aiming for market expansion. Why and to what extent can this also be significant for Europe?

Kira is a specialist in AI-based automated document analysis. Main use case is due diligences / data rooms in transactions. Accordingly, Kira's customers are mainly law firms, globally.

Litera, which has grown enormously through numerous acquisitions, describes itself as a leading provider of legal workflow and workspace software. The focus is clearly on speed and effectiveness of legal work towards client service and retention. Litera sees itself as one-stop-shop for any type of legal work. That's why Litera aims to leverage Kira's core expertise in its own software suite. 

The big difference, however, is that Litera also operates in the corporate law market, and by spinning off its own new startup Zuva, Kira is also expected to follow this path in the future.

Consultants on both sides of the Atlantic won't be too happy to hear that.