Presentation in Venice for the Venetian Criminal Bar Association
[2] What I present today is relying a lot on the research financed by the Council of Bars and Law Societies of Europe (CCBE), the European Lawyers Foundation and the European Commission.
[3] A short summary of this is also present in the book recently published by the American Bar Association, where there is a chapter on European Perspectives.
[4] I will talk about why lawyers cannot just ignore using and financing AI tools. What are the most important group of tools at the European level that will affect all lawyers, including criminal lawyers working on everyday matters and not just BigLaw dealing with contracts governed by the laws of the state of New York.
AI is a very wide term. This is bad, because people will usually have very different things in mind when one is talking about AI or AI tools and makes it difficult to understand each other
But when we are talking about practical uses by lawyers, including criminal lawyers, we will understand each other very well:
We are not talking about specific technology used in tools, whether a tool is relying on large language models, transformers or convolutional neural networks.
We are talking, in general, about innovative tools that are made possible by recent advances in the state of the art of legal IT.
AI tools as used by lawyers have their own, specific risks, that makes it very different from the risk of IT tools we have been using since 1980s.
I will talk about these risks and its European aspects and how, I believe, that will affect our profession in the coming 10 to 20 years.
Let me cover some economic aspects as well, because these are surprisingly common for all lawyers in the EU, especially for criminal lawyers and determine our future the same way, not only in Italy, but also in other countries of the EU.
[5] In the economy, on a global scale, everyone is expecting AI tools to bring about great changes, to overturn existing structures and how we work
[6-8] Although the global market’s trust in AI tools took a beating recently,
[9] - it is still the number one hope, it is a kind of technological salvation: we expect this technological change to turn our lives to the better.
Or, at least, if we strongly believe it, we will be among the selected few whose life is helped by these tools more than harmed. [10] But we’re told, to turn that remote hope into reality, AI tools need data and tech companies developing these tools need investment, money.
Data is not lying on the street for free.
To find it, to generate it and to analyse it, to turn it into even more useful data, we will need, again, more money
In our profession, we have lots of actual and potential data, good quality data. We create lots of data: we are writing briefs to courts, and memos to clients, to authorities, we go to courts, we debate there, we interview, we argue, we analyse documents and evidence, we receive written decisions, we react to those and so on.
Actually, legal documents are one of the best sources for “natural language” data to be processed, because lots of people work on creating it, work with such documents each and every day and they are of good quality, even if we, ourselves, doubt it from time to time.
These are all great sources of data that can be analysed and used to automate not only legal processes of the society, but also fields outside law, because everywhere people use “natural language” to interact with each other, from economics to customer relationships. This general sense of hope of the society in AI tools, is not something that we, lawyers, can choose to ignore,
it’s not a decision of lawyers whether we want to use AI tools or not in our profession: it is the decision of the society in which lawyers play a specific role. First, if clients find a new way of communicating with each other, and they got used to that channel, we have to be available on that channel as well. Secondly, developers of the tools relying on legal data will challenge the legal profession even if there is a professional monopoly, a reserved activity. The developers of such tools try to make new income, from investors or customers, and will pretend to be able to replace lawyers or, at least some functions of lawyers. If there is any statutory monopoly, they will still try to mislead investors and customers what their technology is capable of, even if not for selling, but at least for marketing purposes.
They highlight their tools are better, faster, more reliable and cheaper than a hypothetical lawyer. They promise some kind of alternative to what we do and how we do it.
And we have to react to that: we have to send the message to the same audience again and again that these claims are misleading and the lawyer’s profession is still relevant.
But two months later, developers again will claim that due to some new magic we will never understand, the new version of their tool is now much better and reliable than the previous one.
The easiest way to dissuade the public from these lies is to incorporate new tools as best as we can into the work of lawyers, as many lawyers as possible.
But of course, this has its own costs.
Lawyers are pressured to finance the development in the tools and data they are using. And these costs are huge and becoming bigger, it’s a real “arms race”.
How can we finance that?
Will that change the way lawyers work, they cooperate with each other? 3. [11] To see where we are and what freedom we have, let’s take a look at some numbers. How much can lawyers spend on financing technology and AI tools? The legal market (6910 including lawyers and notaries) in the EU is bit less than 1% of the GDP of total services (which makes up 80% of the total EU GDP). Of course, we all know that the importance of the legal sector goes beyond a mere 1% of the economy. Italy is the third in terms of absolute “legal services” turnover per country, it is a “legal powerhouse”. [12] But how much of this income do lawyers spend on software and legal technology? The closes figure we have is an industry estimate of $ 26 000 million worldwide that also includes tax and accounting spending, includes both legal databases and legaltech. So, this is not the legal AI tools market, but a theoretical maximum of spending. Let’s take a look at the exact figures from the top three worldwide vendors in this market: 64-74% of their income comes from the US, not the EU. EU is only about 13%-25% of their total world income. This shows US firms spend a lot more on IT. Why is that? Let’s make some guesses.
[BACK TO 11!] Their legal turnover is twice that of the EU, they are stronger in legal export than any EU country, just like England itself is playing in a different league than any single EU economy.
But even compared to their bigger legal services market, the IT spending of lawyers on the top three vendors is disproportionately larger.
[13] The size of typical law firms in the US (and especially in the UK) are very different from that of the EU.
The first slide shows how many employees the law firms of each size category (micro, small, medium and large enterprises) have. In the EU, most people in the legal services market work in micro enterprises and the importance of large firms in terms of employment in this sector is very small. While in the UK, this is the opposite and the US is between the two.
[14] If we take a look at the market share of the turnover of law firm categories, that’s even more pronounced.
And that’s especially true of Italy. Italy is very strong in small businesses in general, and that is also true for its healthy legal market.
The typical law firm size for criminal law practice, is a small firm, even in the UK and the US.
While I don’t have statistics for this, but probably, larger firms spend on IT a lot more than small firms.
Why could that be a problem for the long term? Because the legaltech market will focus first on the needs of large law firms, on BigLaw and that also means, on the US and England.
For law firm that want to subscribe its lawyers for all the top products that e.g. Thomson Reuters currently has, including all the editorial content that could be useful for a practice, all the databases, all the generative AI tools with CoCounsel, that would cost a ballpark figure of around $3000 per month at least on list price. [15] To understand the problem in a practical way, let’s take a quick look at some of the tools provided by the large vendors and see what’s needed to make them work and whether they could help small law firms as well.
I don’t know the Italian legaltech market, so my apologies for not reflecting on whether you have these tools or not, but that’s not the point now. I just want to show you some typcial tools, what’s so expensive about them and what’s not so expensive.
The “AI tools” for lawyers is just a tip of the iceberg. For these tools to work optimally, you will need a whole legaltech value chain behind it. You can buy and use only “AI tools”, but for these tools to serve lawyers, you have to start with having resources from the bottom of the pyramid.
While public data, the first tier should now be free, and is indispensable for the lawyers’ work, more and more tools from higher tiers become indispensable as well.
To be able to provide more valuable advice you will need either lots of experience and learning, which is very time-consuming, or have access to added-value and premium content, which is not public, and guarded by publishers, relying on countless hours of editorial work.
That’s a serious market advantage from which only the largest legaltech providers and database publishers can benefit.
Let’s see some examples for such tools.
[16] The first one is just an example of editorial content that can enhance criminal law practice, including draft briefs and memos or practice notes. There is nothing really interesting or surprising in that one. The content is not necessarily the best, it’s just that having this content at the publisher can greatly enhance the legaltech tools at the next levels.
[17] Like with this simple technology: you may use either your own templates or pre-made templates if you have to create a large number of documents, connecting it with databases etc.,
[18] Some other comfort tools can assist connecting some elements of your content in your “unstructured” document with legal cases, citations.
[19] Even better, when you enhance the computer-readable structure and content of your document in a tool that is designed to handle lots of documents as part of a court case, called a case management system.
It can extract information, help you locate relevant information and …
[20] even build an overview of a case,
[21] connected with research tools, and giving you …
[22] a heatmap of weak links in your case, linking arguments.
[23] And the last one is an example of an integrator tool - a large language model like those behind ChatGPT where you can ask the language model questions about your case and it can take into consideration the structure you have built in relation to your case, analysing the available evidence, helps you with comparing and contrasting your facts with available case law findings, piece by piece. [24] At least we see that there is a strong drive for these large vendors to serve the robust and competitive US BigLaw market, to invest in product development.
Reading the annual reports of the large legaltech vendors, it is a clear pattern that revenues from traditional legal databases is decreasing, but revenue from this new class of value-added legaltech tools can offset this decrease and even achieve a couple of % growths. There is a strong, innovation-led competition in this market, at least for the premium products. But as long as premium products are available, in time, these developments may trickle down to the mass market of lawyers as well.
The premium product providers or the BigLaw firms are not interested in serving this wider market, but at least lawyers can clearly see what functionality are useful for their practice, but is too expensive. That also means legaltech startups and contenders will try to sell the same features to the mass market.
This benefits the small firms as well, including, criminal lawyers IN THAT MARKET. [25] Unfortunately, all the legal databases and most of the legaltech innovations are language- and jurisdiction specific.
There is no guarantee that any innovation from the US and UK market can be transplanted to other countries.
These smaller markets have to have their own incentives.
With regards to one technical class of AI tools, large language models: we see how well e.g. ChatGPT speaks non-English languages based on relatively small training material, but they are still useless at best in replying to legal questions to laws they were not trained on. [26] Because in the EU, the national laws of member states are fundamentally different. We have three common law countries and civil law countries with fundamental differences in law, such as Germany and France… Even with harmonisation and the European Convention on Human Rights, there is no “ius naturale” that can provide practical answers for any EU jurisdiction…
There is no possibility for cross-jurisdictional “transfer learning”, which would mean that a tool trained on the law of one EU member state can be used to answer questions on the law of another. That’s a very difficult problem with very little hope to be solved. Another problem is that gathering all the case and national laws of the EU member states is a very costly enterprise, with considerable differences in what the government and courts publish. There are already very good initiatives, but they are far from being complete and they are without strong political backing.
A third problem is the difficulty in integrating with different court systems. The way lawyers have to interface with different electronic court solutions is very different from country to country and these solutions need constant maintenance, because court solutions also constantly change with time.
In short, there is no single market for “AI tools” or even legaltech tools or legal databases for lawyers in the EU. Software developers face challenges in terms of scalability and profitability at the EU level. [27] That leads us back to the problem of who will finance the development of these legal tools? Will the “AI tool” market for criminal lawyers in Italy be big enough to finance itself? If not, who will do that if “not doing the development” is not an option?
The EU or the governments may subsidize some of these solutions, because they also have a stake in lawyers being able to connect to electronic government services and participate in those proceedings as far as its absolutely necessary.
But will they provide lawyers with user-friendly electronic government and court solutions? Will those tools be adapted to what the lawyers need?
There’s plenty of evidence that if lawyers have to use the same tools as citizens, this can go wrong a number of ways:
no features available for substitution, for access by legal assistants or trainees, or for authorisations to access the client’s data based on a power of attorney;
no special rights online that lawyers are otherwise granted in the “brick-and-mortar” court or prison system (access to documents, high-security areas etc.)
This problem is more acute in criminal law and it is very unlikely that the government will ever make tools available to lawyers at any time that enable lawyers to successfully challenge the police or public prosecution in analysing evidence or otherwise effectively find weakness in the government’s position. It is highly unlikely that the development of AI tools will be cheaper for lawyers in the long term. It may also lead to a situation where the lawyer’s market becomes stratified: there will be lawyers who can spend more on advanced AI tools and those lawyers who will have to use the bare minimum. It can also lead to a situation where lawyers working outside criminal or family law, will be forced to rely more and more on foreign laws just to do their daily jobs, because these foreign jurisdictions are so much better supported with AI tools. Large businesses may continue to force even small businesses and also their law firms in EU countries to use a specific AI-friendly jurisdiction, wherever relying on national law is not mandatory.
