The AI in Law Debate: Shaver v Inkster and beyond
In September 2023, Denis Potemkin of Majoto organised and chaired ‘The Great AI [in Law] Debate’ between me and Nicola Shaver.
I am now managing to get around to posting about that on here to add to my series on Legal Chatbots.
The debate motion was:
Generative AI already enhances the delivery of legal services, and will revolutionize it within the next decade.
Nicola Shaver was for the yea’s and Brian Inkster (i.e. me) was for the nay’s.
The debate covered:
The short view: Is GenAI already useful, or still just a toy?
The debate began with whether there are real-world use cases for it, or is the tech still too immature to be truly useful. Are law firms and legal teams just jumping on the latest hype wagon without any substance behind it, or are they doing the right thing in exploring and experimenting with this tech?
The long view: Is GenAI just another hype cycle like the blockchain and metaverse, or is it different this time?
Then we moved on to debate whether there is something different about AI in terms its capabilities or adoptability, versus previous hype cycles. Will “remarkable” turn into “useful” for legal work? Will GenAI achieve its promise eventually, or are there inherent limitations in the tech that will limit its usefulness, or at best it will be just one cog in a large toolbox? Will real change require a different kind of technology altogether?
But in the meantime…: For legal teams or law firms who side with Nicola rather than Brian, or just want to experiment – should you buy or build your own?
Next up was a debate on the immediate practical question, of whether law firms should build their own systems e.g. using open source LLMs and training it on their data, or buy ready-made solutions like Harvey (whatever that may be!)
Final thoughts
Denis concluded the debate by requesting a brief answer to “How many lawyers will GenAI replace in 10 years time”?
Favourite Bits
Denis Potemkin gave his favourite bits from the debate on LinkedIn:
My favourite quotes from Brian:
“Another potential use case was to translate legalese into simple language. I think lawyers should be doing that as a matter of course without having to run what they do through chatGPT.”
and “It may well get better, but it may get worse.”
My top quotes from Nicola:
“The best solutions we’re seeing right now are additive. They’re ones that are drawing on different types of machine learning, as well as other types of technology to produce something that is robust and genuinely useful”.
and “I’m an optimist. I do think to work in legal innovation you have to be an optimist.”
Watch the AI in Law Debate Video
You can watch/listen to ‘The Great AI [in Law] Debate’ here:
The Debate continued and got a lot more heated on LinkedIn as others joined in.
The problem with LinkedIn debates (and indeed social media debates in general) is they tend to be very much of the moment. All those moments will be lost in time, like tears in rain…
I like to preserve those moments for posterity. I can do that in this old fashioned thing called a blog 🙂
So here are the tears, without the rain, covering:
- The Build or Buy AI in Law Debate – Jake Jones
- The Buy v Build Survey – Legal OS
- Wait and See – Frank Thomas
- We have to be willing to experiment – Kevin van Tonder
- Analysis, Luck and Use Cases – Shawn Curran
- Head to Head in the AI in Law Debate (with a translation perspective) – Deborah Parry do Carmo
- Is Environmental Responsibility the Elephant in the Room? – Deborah Parry do Carmo
- What did Alex just Waken up to?! – Alex G. Smith
The Build or Buy AI in Law Debate
Jake Jones (Co-Founder at Legal OS – AI for Inhouse):
Looking forward to watching this.
For now my position is clear: building in-house is a bad idea because:
– it’s (usually) a massive upfront commitment
– law firms don’t have the skills Inhouse
– the amount of maintenance required for solutions built Inhouse is always underestimated
– there are vendors out there dedicated to solving these problems; it’s all they’re thinking about; most offer reasonably priced pilots that mitigate the risk
– any gen AI solution is just the beginning; vendors will continually improve the software; I’d you built Inhouse, you’ll have to foot this hill
This is a small selection. Many more. Will watch and see if there’s a new perspective here. I’ve not yet spoken to a lawyer from a law firm who is using their own gen AI solution in “production” and getting great value from it.
Basically, why buy a hotdog from an ice cream stand?
Shawn Curran (Director of Legal Technology at Travers Smith):
https://github.com/Travers-Smith/YCNBot
Over 100 clones now Jake Jones, including from companies who actually sell software (they have been kind enough to let us know they took our code as a base for their own Enterprise chatbot offering). Lets be honest with ourselves, OpenAI have done most of the work here, we’re all building the thinnest of thin UI layers on top of their API’s. If our lawyers aren’t getting value from our front end of GPT-4 how on earth would they get value from a vendors! I think there’s a market for buying and building, lets not be dogmatic about it.
Me:
Jake Jones, As you will discover, when you do watch, that is (not surprisingly for the Nay’s!) also my position 🙂
Jake Jones:
Shawn Curran, couldn’t disagree more. Using examples of thin interfaces doesn’t mean all vendors are doing this. At Legal OS, we absolutely are not.
The question “if they aren’t getting value from our front end, how would get they value from a vendor’s” doesn’t say much. So you have a UI for GPT-4? Fine. Try a RAG query on a complex document set with a single API call to GPT-4. This is a very basic use case and it will be inconsistent and unpredictable. If you’d like examples of this failing, I’ll jump on a call with you no problem.
There are much more complex use cases that a thin product wouldn’t begin to be able to handle, not to mention the importance of the experience layer in an AI product. Most legal use cases need governability, auditability, predictability; with thin applications of GPT-4 you get none of these.
There’s a market for buying and building, sure. I’m just bullish about the best vendors cracking significant problem spaces in legal with generative AI; I haven’t yet seen any evidence to feel the same about law firms building. That’s not to say they can’t do it. Of course the right team can. I just believe it’s a lot less likely.
Denis Potemkin (Founder of Majoto | contracts that people understand and love | faster deals more trust through design and automation):
Jake Jones , Shawn Curran , sounds like there’s a deep dive debate brewing here!
Clare Fraser (Product developer – building LLMS using public domain legal data to support access to justice):
Agreed. And as Nicola Shaver said in the debate, it’s not as expensive as you might think it is, and lawyers are excited about this tech so why shouldn’t they give it a go? There’s a snobbery about the use of APIs – oh you didn’t build an LLM, you just customised it – not necessary IMO. What matters is if it is helping the users.
Shawn Curran:
Jake Jones,
- RAG Query on complex document set, no worries, that will be the Analyse product we built and are offering on a shared source basis. Our template feature is growing every day to get the right qualitative and quantitive prompts to do RAG status, and a variety of other things – https://www.linkedin.com/feed/update/urn:li:activity:7098310726598975488/
- The importance of the experience layer. Absolutely, couldn’t agree more, that’s why we’ve been building a data labelling asset for the past 3/4 years and are very keen to retain it for our own benefit! https://legaltechnology.com/2020/10/19/next-level-ai-travers-smith-launches-open-source-contract-labelling-tool-and-why-legal-ai-vendors-will-probably-hate-it/
- Governability, Auditability and Predictability. Let’s take the one at a time. Governability, see https://arxiv.org/abs/2306.11520 on how much focus we put into governance of the models. Auditability, believe it or not, little old ‘shouldn’t build tech’ law firm introduced this concept to our industry by open sourcing data labelling platforms to ensure full auditability of training data used to build models. https://www.youtube.com/watch?v=URWI0ifJNYQ and then Predictability, have a look at Analyse screenshot, an early sign of our ability to make the human-in-the-loops verification job easier. https://media.licdn.com/dms/image/D4E22AQFOZbfYVkEy0Q/feedshare-shrink_2048_1536/0/1692369156937?e=1697068800&v=beta&t=u4VWdMYtV4gG5SlxdOSB2uegA63arLJ7R7dw68QnvCk
This is off the back of 5 months of experimentation and education with an Enterprise chatbot we developed ourselves, something only being made available to the market now. Were we supposed to wait for the ‘software companies’ to catch up so we could execute our own AI strategy? Or sit on a waitlist for months?
Oh and on building, if AI isnt enough, we even built the first, and only, cloud-to-cloud API-to-API migration tool on our recent move from NetDocs to iManage. Something that doesnt even exist ‘off-the-shelf’.
However, having said all this, building software is not for every firm. I wouldn’t be so arrogant as to think this is a binary problem with dogmatic views applied across the board. Some firms can and do build, and do it very well, others cannot, and they buy. Vendors should accept the issue is more nuanced in my opinion. 🙂
Jake Jones:
Shawn Curran in no way am I suggesting that there cannot be successful building projects internally. If law firms successfully deploy their own software that has a meaningful impact on the productivity (and qol) of the lawyers in the firm, this should absolutely be celebrated.
As I said in my initial reply, I’m just bullish about the best vendors cracking the problem space in the most scalable, sustainable way. I believe vendors with deep experience in building delightful, usable experiences will be the “winners” here, and I predict law firms ultimately adopting external tech.
Of course, I could be wrong; but I might as well share my opinion to stimulate lively discussion. I’d rather share my perspective (& predictions) and be proven wrong, rather than not share.
I’m a vendor and I am seeing some brilliant work from my team and other vendors, work that is a long, long way ahead of the thin “AI” products that are little more than a front door on ChatGPT; so of course I’m going to promote this and believe in it.
By RAG I’m referring to retro Al augmented generation, not red/amber/green status.
If you’re doing well with what you’re building, kudos. I’m not commenting on individual projects. I’m stating my position on the “build Inhouse or purchase from a vendor” debate. Nothing more.
Me:
Shawn Curran, Jake Jones, As I said on the podcast there are lessons to be learned from Clearspire and Atrium when it comes to build your own: https://thetimeblawg.com/2020/03/07/atrium-a-post-mortem/ – Don’t be a Clearspire or an Atrium.
-+-+-+-+-+-
The Buy v Build Survey
Legal OS conducted a Survey on LinkedIn on Buy v Build. The results and further debate:
Martin Bregulla (Legal Product Manager @ Legal OS | Freelance Legal Tech Advisor | Fully-qualified Lawyer):
I think for law firms it depends on a lot of factors: what freedom does the “innovation team” get (both financially as well as regarding the projects they work on), how big is the team, what is their technical and product expertise?
It also depends on how well the innovation team is integrated within the law firm to be able to rely on legal experts’ know how and being strategically aligned with the law firms’ overall goals.
It’s a tough balance to find but if you can find it, there is a chance that you might have success building your own solutions as a law firm.
If any of the above is missing, I think it’s next to impossible to develop anything that will have a noticeable impact from inside of a law firm and buying legal tech would be the way to go.
Legal OS:
Thanks, Martin.
In your experience, if an innovation team decided to build this themselves, to what extent would this consume their resources for the following 12-24 months (or for the lifespan of the product/project).
Martin Bregulla:
Completely depends on the desired outcome. But if they were aiming for a scalable and marketable product, then it would require the same amount of resources as it would a startup, so a team of at the very least 5-7 people (likely quite a few more even) working full-time for many months doing focussed work for the technical side as well as significant effort from the legal experts to include legal know how.
I don’t see how an endeavour like this would come easier for a law firm than for any other company.
-+-+-+-+-+-
Wait and See
Frank Thomas (ASEAN legal trainer. Former BigLaw thriver and survivor. Proud husband and father of 3 Alpha humans. Steadfast skeptic of new Tech that overpromises and underdelivers..):
Wait and see is the best approach.
I am here in Asia, better culture.
Americans rush then crash and burn imho.
Remember Amara’s law: We always overestimate in the short term and under in the long term.
Just my 6 senses
-+-+-+-+-+-
We have to be willing to experiment
Kevin van Tonder (Focus on legal department performance | Legal Operations | Legal Technology | Change | Commercial Contracting):
I think healthy scepticism is good but we have to be willing to experiment to see whether it is helpful for one’s particular needs or not. At the end of the day, it is another potential tool that can be used within a lawyer’s workflow. Good debate. Enjoyed it.
-+-+-+-+-+-
Analysis, Luck and Use Cases
Shawn Curran (Director of Legal Technology at Travers Smith):
‘ChatGPT is the starter not the main course’ – actually in my view it’s likely to go in the other direction towards canapé, because more and more websites are restricting access for the large base models to continue to learn, or are putting paywalls up, so these large base models will naturally regress. If websites like Getty Images are fair game for synthetic image creating AIs, the public internet dies, and we move to a private internet future with everything of value hidden away.
Having said that, that’s not the rationale Brian is using on why this technology won’t be successful but rather ‘hyped technology before hasn’t had the impact it promised so this won’t either’ which I really struggle with as sound analysis of the situation.
There is absolutely no doubt if the worlds public and private data is provided to a single base model with enough expert fine tuning that this tech is a revolution for law but that won’t happen, GPT-4 will regress due to a lack of available data and GPT-5 is impossible to release due to uninformed consent of enterprise data going into the GPT models.
Brian might be right, but purely by luck rather than detailed analysis of the situation.
And only until we get back to GPT-4 levels legitimately!
Me:
Shawn Curran I think my analysis of the situation is a bit more involved than that! My real issue with it is finding actual use cases for it in legal practice. I go into that in a bit more detail here: LLMs in Law: Hype v Magic
… and that is just the latest in a series of blog posts by me on the topic: Chatbots
Shawn Curran:
There are definitely use cases. I agree with you on blowing away doc auto investment for a clause from Wikipedia, but on the ‘legal’ reasoning side, when you give it all the data and ask the model to reason on it, it’s doing some amazing things.
Me:
… and you will find my comments on regression (slightly different from your point) here: https://www.linkedin.com/feed/update/urn:li:activity:7104056939910713344?commentUrn=urn%3Ali%3Acomment%3A%28activity%3A7104056939910713344%2C7104262011252867072%29&replyUrn=urn%3Ali%3Acomment%3A%28activity%3A7104056939910713344%2C7106636961209049088%29&dashCommentUrn=urn%3Ali%3Afsd_comment%3A%287104262011252867072%2Curn%3Ali%3Aactivity%3A7104056939910713344%29&dashReplyUrn=urn%3Ali%3Afsd_comment%3A%287106636961209049088%2Curn%3Ali%3Aactivity%3A7104056939910713344%29
… If you can give me some actual useful and real use cases over at LLMs in Law: Hype v Magic you could win a magic wand 🙂
Nicola Shaver (CEO and Co-founder, Legaltech Hub | Author | Innovation, AI, Legaltech Leader, Advisor, Investor | LLB, MBA | Fastcase 50, 2021, ABA Women of Legaltech, 2022 | Adjunct Professor):
I did actually enumerate a fair number of genuine use cases during the debate…
Me:
Nicola Shaver, Indeed. I’m interested in law firms actually doing things with them that is better than using existing (non GenAI) tech as opposed to what they might be able to do with them. Hopefully Shawn Curran will tell us how they are actually using LLMs to great benefit at Travers Smith.
Nicola Shaver:
Brian Inkster I’m in fact working with law firms and their lawyers on the ground to do this – my answers come from real-world deployment and use cases with firms that have hired me for this purpose.
Clare Fraser (Product developer – building LLMS using public domain legal data to support access to justice):
Brian really enjoyed the debate, well done to you and Nicola. Ok so you asked for real use cases which Nicola provided so… does that not mean that there are actual useful and real use cases?
Shawn Curran:
Brian Inkster might be worth following our AI page to learn more. https://www.linkedin.com/posts/travers-smith-artificial-intelligence_travers-smith-is-pleased-to-announce-the-activity-7098310726598975488-qkcn?utm_source=share&utm_medium=member_desktop
Me:
Clare Fraser, Possibly. If Nicola can ask the lawyers to comment and provide details of how it has benefited them there will be magic wands to be had! I will remain sceptical until I hear the real use cases from the coalface. I’m always a bit wary of third party accounts e.g. Harvey!
Nicola Shaver, Get the law firms and their lawyers to confirm what they are doing with it and the benefits to them and I will very gladly hand out magic wands to them 🙂 [N.B. Unsurprisingly, three months later and I still await a response on that]
Shawn Curran, Thanks for the link to that page: “At this stage, we have not utilized this technology for client work, but we are preparing to discuss how Analyse can potentially support our service delivery for them.” So not a use case yet but an experiment that might lead to one?
Shawn Curran:
Brian Inkster Use Case – “a specific situation in which a product or service could potentially be used.”
https://www.oxfordlearnersdictionaries.com/definition/english/use-case
Me:
Shawn Curran, My search has been, clearly over optimistically as it turns out, for “how LLMs work in an actual real law firm”. I will leave potential ones to the legal futurists!
Shawn Curran:
Brian Inkster this saved me about 10 minutes!
Me:
Shawn Curran, Can’t imagine a lawyer ever having to do that as part of their day job!
Shawn Curran:
Brian Inkster, summarisation example based on https://www.crofting.scotland.gov.uk/corporate-documents
Blocked
After this point Shawn Curran blocked me. I seem to have this effect on law firm legal technologists when I mention use cases 😉
Thus, I didn’t have the opportunity to respond to Shawn’s summarisation of Crofting Commission documents. I can now do so. Firstly his prompting is bad. He mentions the Crofting Reform (Scotland) Act 2010 but omits the Crofters (Scotland) Act 1993 (which is far more significant). The Crofting Commission do not deal with “commercial aspects of the industry” but regulate the industry. A leading crofting law specialist knows what the Crofting Commission does and does not need a chatbot to tell them.
However, the main purpose of Shawn’s examples appears to be to show that ChatGPT can summarise documents. I knew it could do that. It is the usual “use case” that legal technologists or futurists come up with to apparently show conclusively and irrefutably that ChatGPT will replace junior lawyers. Rubbish. Real lawyers don’t want to read summaries of documents that might omit crucial information. Any lawyer worth their salt will read any and all relevant documents in full and extract therefrom any parts relevant to the specific legal argument they need to make. To do otherwise is a folly and potentially professional negligence. So it is not a real use case it is a daft case.
Also by limiting the summarisation to two particular documents the significant criticism at the time of the Crofting Commission by Audit Scotland was sadly lost.
Thus, if anything, Shawn Curran demonstrated the shortcomings of using ChatGPT in legal research.
-+-+-+-+-+-
Head to Head in the AI in Law Debate
Deborah Parry do Carmo (Lawyer-Linguist • CIoL-certified • Legal and financial translator: Dutch and Portuguese to English • In progress: Specialist Paralegal Qualification in Wills, Trusts & Executries (Scotland) • Triple national: UK/SA/PT):
Head to Head
I thoroughly enjoyed The Great AI [in Law] Debate between Brian Inkster and Nicola Shaver.
If you’re a lawyer, paralegal, lawyer-linguist, legal translator, legal process engineer (or equivalent), it’s well worth the 45-minute listen.
Both debaters put their positions across clearly and respectfully. Brian as the skeptic, and Nikki as the optimist.
Translation
Translation was only touched on briefly. Nikki mentions at around 9:50 that genAI can translate legal documents from one language into any other language, and that’s particularly useful if you’re working on cross-border matters.
I could write a post (and more) about that statement, but I will say that to categorize translation output as useful, you have to be able to rely on it. To be able to rely on it, you’d need a professional working knowledge of both languages in law yourself, alternatively a lawyer-linguist or legal translator to verify that for you. Otherwise, you’re flying blind. What appears well translated can be—and often is—very deceiving. LLMs haven’t cracked the code of comparative law and won’t anytime soon. And there are still the ‘normal’ issues like hallucination, omissions, and gibberish, not to mention confidentiality concerns.
In relation to genAI being helpful for ‘translating’ legalese into plain language, I particularly enjoyed Brian’s comment that lawyers should aim to write that way anyhow. Indeed, if we all understood what legalese is (and isn’t), then maybe more lawyers would. In the comments, I’ll post the link to my blog post ‘What is legalese?’.
The hype behind generative AI is driven by huge advertising budgets—we see the same in translation with the tech evangelists, not translators, saying how phenomenal adaptive NMT is—so it was good to see a skeptic and an optimist both take the floor in this debate.
I knew before I watched what side of the debate I’d probably fall on: far from being a Luddite, I embrace technology in my translation work. I was one of the early adopters of CAT tools, and my toolbox doesn’t stop there. I’ve been a paid subscriber to DeepL since 24 March 2018. But I’m never swept along by or get overly excited about tech. I’m a cautious user, unimpressed by hype. And I’m a strong proponent of machine-in-the-loop thinking (the technology must work for me in my workflow; I’m not the human in a loop).
By the end of the debate, it was clear Brian and Nicola shared some common ground, but I came away firmly in Brian’s court.
Thanks to both of you and Denis Potemkin, I enjoyed it and was impressed by how calm, measured, and prepared you all were.
Now, back to translating an annual report, with ChatGPT nowhere in sight!
Nicola Shaver (CEO and Co-founder, Legaltech Hub | Author | Innovation, AI, Legaltech Leader, Advisor, Investor | LLB, MBA | Fastcase 50, 2021, ABA Women of Legaltech, 2022 | Adjunct Professor):
Thank you for weighing in, Deborah Parry do Carmo , and I very much appreciate your expertise in the translation space.
Where I am seeing this used is very specifically to translate a clause or two rather than large numbers of documents. From my two decades working in law firms I know that an issue is the continued use of Google Translate for client work so that lawyers can understand incoming documents in a matter – obviously problematic from a security perspective. Having a secure deployment of a domain-specific LLM (which has been reinforced with legal content and is able to “understand” legal concepts) is preferable for this sort of narrow translation use case than public sites like Google. Another use case is the translation within a language of the feminine to masculine case or vice versa, in languages where this is a feature. In other words, if you are using a document from a past transaction or litigation matter and the relevant party was a women but this time the relevant party is a man, having software that can rapidly change the grammar is a time-saver.
If you listen again, I did not make the comment about legalese and plain language. Brian raised this in relation to other comments he has received online.
Denis Potemkin (Founder of Majoto | contracts that people understand and love | faster deals more trust through design and automation):
Nicola Shaver, you make a good point that lawyers use basic web tools like Google and Google translate A LOT, so genAI (in the right shape and form e.g. a domain specific LLM) might be very imperfect but still a superior tool to Google.
I think in some areas, like doc automation (getting a draft agreement prepared), Brian Inkster’s argument holds water: that there is other tech that is more effective if done right.
But there are two factors that seem to favour AI: one is accessibility, the other is the lack of domain specific tech for certain tasks (translation perhaps being one such example).
So using genAI could be a valid exercise in those cases: it’s imperfect but it’s more accessible.
I think that still leaves open the question of ultimate value: how much value and efficiency are you going to get from AI given the imperfections and the effort in putting it in place? I think there is no single answer for that – it’s going to depend on use cases, tech used, controls, context etc. And of course with time the equation should swing in AI’s favour even if it might not yet.
Nicola Shaver:
Denis Potemkin, I think honestly it’s quite hard for people to weigh in who haven’t actually been on the ground trying to get lawyers to adopt new technology (I’m not talking about Brian here, I know he runs a law firm). It’s all well and good to say document automation is superior but the reality is that anyone working in law firm innovation can tell you these tools have been clunky at best and really have not gained traction or adoption by lawyers because they don’t mimic the way lawyers work. The smart drafting solutions are often superior here and ALL of those vendors immediately realized how impactful gen AI would be to their solutions and started building it in. Again; I am a huge believer in the importance of UX (see my recent article on this) and do not, in fact, believe in magic. I believe in hard work and getting things done. But one of the truly transformative aspects of this technology is that it can make what has appeared to be cumbersome and difficult before (like getting lawyers to “fill in the blanks” on a questionnaire w/o actually seeing what document they’re producing as they “draft”) far more intuitive. Anyone who deals In change understands how important that is.
And I am not suggesting – nor do I ever in the debate – using ChatGPT off the shelf. I’m talking about leveraging LLM technology that has been reinforced with legal content and data and that has been made secure for legal environments. Depending on use case, these are also able to display source materials and say “I don’t know” if there is no answer in the dataset at which they are directed – some can even now ask clarifying questions if the prompt is inadequate (I see new start-ups on a daily basis building with this tech).
Using an LLM for complex drafting is non-sensical if it is generating a new agreement every time – but building this tech into existing UX around drafting will allow for starting from a template and then shaping it in ways that are faster and more intuitive than questionnaires. Etc etc. This is just the beginning and I really think anyone who looks at the world around us right now and thinks that this technology will NOT have a considerable impact in any profession where language is core has their head in the sand, and stubbornly so. I think the question is, how soon will that impact be felt, and in which legal environments will it take hold first.
Denis Potemkin:
Some interesting points there Nicola Shaver on doc automation and UX. Something we’ve always done differently at Majoto. Will come back and write more fully on that!
Deborah Parry do Carmo:
Thanks Nicola Shaver.
We already have far superior translation tools to Google that also considerably outperform the likes of ChatGPT (there is research on this).
I hear you on leveraging secure LLM technology.
My point about being able to rely on the translation output before calling it useful—wherever it’s coming from—stands though, as does the fact that none of these systems have cracked the code of comparative law. I have a post coming out soon with a good example of that.
Whether we’re talking about a clause or a set of documents, a case can turn on a word or even a comma. For determining what needs to be professionally translated, sure, genAI can help sift, but a paid subscription to DeepL or even ModernMT (which I wouldn’t use myself because it’s also hyped beyond belief) already does this confidentially and handles it better. Research in translation already shows that the quality of ChatGPT output in translation is getting worse, not better. I don’t have the links handy but they’ve been circulating here in recent weeks.
I have no time to listen to the video again, but accept my memory must have failed me. Apologies. I’ve tweaked my post in relation to the plain language issue.
We also have existing technology for the grammar use case you mention. We’d have the first translation in our CAT tools and could run our translation memory through the new translation file and immediately fix the gender (and any other) changes. It shows as what’s called a fuzzy match. We’ve been able to do that since around the time I started translating (2002) when I left law after emigrating, if not before.
Me:
Deborah Parry do Carmo, You make a good point here which is that if there is existing tech doing a decent job there is not necessarily a need to use new tech with an AI sticker on it. Often that new tech might not be as good and the ‘AI’ might just be AI washing.
Deborah Parry do Carmo:
Thanks Brian Inkster, that’s exactly what’s going on in translation. It might change, of course; the new tech might outperform the existing tech or, more likely, be integrated in the existing tech to enhance it some way. But there is nothing newfangled happening right now that makes me sit up and say wow.
Unfortunately, as translators, we can’t compete with the advertising budgets of certain hype-peddling intermediaries trying to line their pockets at our expense. That’s our issue, not the tech itself. We love tech generally as a group, have been using it for decades, and certainly don’t have our heads in the sand but sweeping statements about what tech can do need to be rebutted. Our livelihoods depend on it.
Me:
Nicola Shaver Not sure what document automation systems you have been trying to implement but I don’t have experience of “clunky”. A system that produces the same document templates each time populated with data input once per specific transaction can only be better than a GenAI solution. As I said there might be benefits for a GenAI layer on top but you really have to get the document automation layer sorted first. Any problem with adoption has been poor implementation and the same will be true of GenAI solutions. The vendors did not necessarily realise how impactful GenAI would be to their solutions and started building it in. Those that did simply jumped on the hype bandwagon. Especially the start-ups. Many established players are sitting back and watching the chaos unfold. Anyone who deals in change should understand the importance of incremental improvements and making the existing technology you have work to its fullest potential before adding something new to the mix.
Deborah Parry do Carmo:
There’s another big-picture element that I’ve thought about overnight that touches on the importance of incremental change (sometimes being on the outside looking in at debates helps, even though I obviously don’t have the in-depth legal tech experience). I’ll post this evening or tomorrow about it.
Nicola Shaver:
Brian Inkster Incremental improvements will only get you so far. I take it you haven’t read this, which shows how close we are on this topic: https://www.legaltechnologyhub.com/contents/llms-do-not-obviate-the-need-for-ux/
But I have heard first-hand from lawyers that they don’t like drafting without seeing the document they are drafting, which makes the questionnaire format problematic. There are newer solutions that make this easier but the point stands – the combination of gen AI and document automation that allows for drafting from a precedent will be an improvement on current providers.
Denis Potemkin:
Nicola Shaver, I’ve heard from some consultants that users hate the Q&A model for doc drafting, and having experienced genAI, are asking for a single prompt-based approach. I’ve also heard one or two legal teams saying that they’re only looking at doc automation solutions that have genAI embedded (which I think is the wrong approach but it’s more support for the notion that conventional systems built on conditional logic and Q&A forms are considered awkward).
I’ve always thought this, which is why Majoto from the beginning has not been built on this logic, but on (a) generating a starting template from 1 question, (b) allowing the user to fine tune inside the document rather than through a form, and (c) an automation method that does not depend on the usual conditional logic which is very hard to put together. That’s without any AI.
You can imagine how pleased I am that after 3+ years, my belief that this approach is a better UX, is being vindicated!
So I agree that most automation systems are clunky and inaccessible. Not all of them. And perhaps for some the “clunky” approach works, or at least good and disciplined implementation can overcome it.
Me:
Nicola Shaver I know that continuous incremental improvement can in fact get you very far https://thetimeblawg.com/2014/07/04/improving-for-reinvent-law-london-2014/ – I noticed a post by Gav Ward this morning with an excerpt from Atomic Habits on the gains made over time by just 1% improvements a step at a time. Sometimes the attempts to leap frog with new tech will set you back rather than take you forward.
I am not a fan of the questionnaire format but understand it may be more of a BigLaw thing (as I have said before a world I gladly do not inhabit). Document automation done properly (in my view) involves known (to the lawyer) templates that are filled when called up with data that will vary from transaction to transaction but has been input once for a particular transaction to be used numerous times on numerous different templates relative to that transaction as and when required.
This method usually involves little if anything further requiring to be done to the document once produced. Not sure what GenAI could add or improve to that. But always open to seeing and considering suggestions that could layer onto an already highly optimised document automation system.
…. link to that post by Gav Ward https://www.linkedin.com/posts/gavward_onepercentbetter-activity-7107647956861771776-7L7g?utm_source=share&utm_medium=member_android
Deborah Parry do Carmo:
Brian Inkster: Enjoyed those slides, and seeing how you’ve gradually kept layering and coming up with innovative ideas. Had time to look at plug and play now, very impressed(!) and the will kiosk idea is a very clever way of providing a much-needed nudge to people to get their estates in order.
Me:
Deborah Parry do Carmo, And that was 9 years ago. You should see what the slides look like now with another 9 years of incremental improvements at Inksters!
Deborah Parry do Carmo:
Here’s just one example of what NMT and GenAI can’t yet handle in translation (and why I weighed in on the debate). We come across examples all the time.
https://www.linkedin.com/feed/update/urn:li:activity:7108015721497997312/
ChatGPT (first go): The phrase is in German and translates to “Mangle at the Tattoo Booth” or “Wringer at the Tattoo Stand” in English. Utter codswallop for my needs.
Will it get better? Probably, but only in closed, specifically trained systems. Otherwise the dog will just start eating its own breakfast as unqualified “post-editors” accept nonsense and feed it back into the system.
Is all the hype around NMT and GenAI in translation justified? No, yet the continued lies backed by huge advertising budgets are pushing down rates to sweatshop levels and driving talented translators away from the profession.
The problem is NOT the tech (we translators know how to leverage it), it’s the hype merchants.
Graeme Johnston (Software to map work – before that a lawyer):
Machine ITL rather than Human ITL – 💯
Tereza Afonso (BA in Law | MA in Translation | Legal Linguist and Translator | English and Spanish-to-Portuguese Legal and Financial Translation | Proofreader | Translation Tutor | Researcher):
Thanks, Deborah.
I’ll watch the video when I have some free time. As you say, “LLMs haven’t cracked the code of comparative law and won’t anytime soon.” And the outputs are mostly garbage. Sometimes, helpful for exploratory purposes, but with subpar quality.
I’ve conducted extensive experiments with both DeepL and ChatGPT, and neither has proven particularly reliable. In particular, ChatGPT’s performance with Portuguese leaves much to be desired, as it struggles to distinguish between Brazilian and European Portuguese, much like Google NMT.
I understand you’re well aware of the limitations and inaccuracies of machine translation, but it’s always valuable to acknowledge them. When it comes to plain language, I contend that it is a task for translators/linguists, and everyone would benefit if lawyers reconsidered their approach to clear and concise communication from the ground up.
-+-+-+-+-+-
Is Environmental Responsibility the Elephant in the Room?
Deborah Parry do Carmo (Lawyer-Linguist • CIoL-certified • Legal and financial translator: Dutch and Portuguese to English • In progress: Specialist Paralegal Qualification in Wills, Trusts & Executries (Scotland) • Triple national: UK/SA/PT):
𝗕𝘂𝘆 𝘃𝘀 𝗯𝘂𝗶𝗹𝗱
Inspired by yesterday’s lively debate (see the link below), I started visualising my upcoming move to Glasgow and gradual return to law alongside translation.
Is the ‘buy vs build’ debate in legal tech a real issue for me?
𝗡𝗼𝘁 𝗱𝗶𝗿𝗲𝗰𝘁𝗹𝘆
While both sides put forward compelling arguments, the financial aspect tips the scale for me firmly towards ‘buy’.
As Brian Inkster mentioned in the debate, most law firms, especially future solo practitioners like me, lack the financial resources and in-house expertise to develop custom AI solutions. In my case, opting for off-the-shelf legal tech, much like I currently do for translation, is the pragmatic—and only—choice. The debate is secondary.
𝗕𝘂𝘁 𝗶𝗻𝗱𝗶𝗿𝗲𝗰𝘁𝗹𝘆?
If I’d been on the fence, something nagging at me overnight would have pulled me over to the ‘buy’ side: the idea that 𝗰𝗵𝗼𝗼𝘀𝗶𝗻𝗴 𝘁𝗼 𝗯𝘂𝗶𝗹𝗱 in the world of genAI has more significant consequences than we might realise.
Law firms deciding to build genAI systems are embarking on a resource-intensive journey. The data centres, high-powered computers, and energy-hungry infrastructure they’ll need, directly or indirectly, consume vast amounts of electricity and water.‘ChatGPT gulps up 500 milliliters of water… every time you ask it a series of between 5 to 50 prompts or questions. […] The estimate includes indirect water usage that the companies don’t measure—such as to cool power plants that supply the data centers with electricity.’
𝗜𝘀 𝗲𝗻𝘃𝗶𝗿𝗼𝗻𝗺𝗲𝗻𝘁𝗮𝗹 𝗿𝗲𝘀𝗽𝗼𝗻𝘀𝗶𝗯𝗶𝗹𝗶𝘁𝘆 𝘁𝗵𝗲 🐘 𝗶𝗻 𝘁𝗵𝗲 𝗿𝗼𝗼𝗺?
So, is the choice to build genAI solutions a selfish one?
While I understand that some law firms seek control and exclusivity with AI tailored to their unique needs, shouldn’t this reported excitement to build around genAI raise environmental alarm bells?
The AI arms race shows no signs of abating. Meta is developing an AI model to rival GPT-4, OpenAI’s latest large language model. This begs the question: How do law firms even know if the underlying technology they’re investing in now is the best? Aren’t they getting ahead of themselves?
By jumping on the genAI bandwagon, are law firms overlooking the unintended consequences these technologies may bring?
Climate change is a ticking time bomb, so law firms can’t disregard the environmental ramifications of their choices. Are they considering the bigger picture if and when they get behind ‘build’ for legal tech? Are they trying to be mindful stewards of innovation? Curiously, I haven’t come across this in any buy-vs-build threads I’ve explored.
Has it been overlooked? Conveniently ignored? Are EIAs conducted before diving into ‘build’? Or is the environmental impact less severe than it seems?
Legal tech isn’t my field, but I’m keen to hear the answers as someone returning to law and a product buyer.
Tereza Afonso (BA in Law | MA in Translation | Legal Linguist and Translator | English and Spanish-to-Portuguese Legal and Financial Translation | Proofreader | Translation Tutor | Researcher):
Thank you for your post, Deborah.
We need to think about the consequences. People tend to see only the colorful side of things. They forget that LLMs emerged and continue to develop on the back of underpaid crowdworkers and that the carbon footprint is immense, as is the consumption of other resources. We need to think about where we are going and how. It would be good if the euphoria passed and we started to think about all this, about ethics, about privacy…
https://www.cbsnews.com/news/artificial-intelligence-carbon-footprint-climate-change/
Me:
We (Nicola Shaver and I, with Denis Potemkin in the Chair) never even touched on the environmental impact of GenAI in The Great AI Debate. This post by Deborah Parry do Carmo raises a very worthy point. Maybe one for Denis to take a deeper dive on in a future episode of #theSlot?
Nancy Myrland (LinkedIn™️ Coach For Lawyers | Legal Marketing & Business Development | Content, Social Media, Podcasting, Video & Virtual Presentation Consultant | Individual & Group | Speaker, Trainer & Advisor for Lawyers & Law Firms):
This is a great discussion to have, Brian. The environmental impact is discussed when one takes a deeper dive into GenAI, but it is not embedded in every discussion, nor is it easy to fathom unless one is in touch with environmental resource issues.
Also, there is no way the current iterations and underlying technology of GenAI are the best. Knowing it *should* become smarter and smarter, as well as more efficient as time goes by, no current technology or version will ever be the best, will it?
I also think buy and build will start to look a lot like each other as “off the shelf” software will become even more sophisticated in its allowance for customization. When have we not looked back on technological advances and seen how large and expensive they used to be before there was mass adoption and development?
These are incredibly interesting times and discussions to take live and take part in.
Me:
Interesting thoughts Nancy Myrland. There is a school of thought that thinks GenAI might get dumber and dumber, as it is excluded from scraping certain content and trains on its own output! Time will tell!
Nancy Myrland:
Brian Inkster, I have read that also. I don’t think we can even begin to imagine what will come next, or how developers will figure out how to get around that issue to feed the beast with intelligent content.
Denis Potemkin (Founder of Majoto | contracts that people understand and love | faster deals more trust through design and automation):
Perhaps not dumber but (a) more prone to error as it picks up ai generated content with hallucinations and (b) more anodine and banale. I also worry about the impact of propaganda and misinformation since it’s a black box much more so than browsing. So tech and people that moderate and tune to just make it work, is going to be essential. Already there is a whole economy of training/moderation sweatshops which is another social impact that needs to be unpacked!
Me:
Denis Potemkin But that makes it dumber, no? I used that word because a number of the articles suggesting this dumbing down have used it.
Denis Potemkin:
Brian Inkster, yes you’re right. I was just saying that even if one doesn’t agree it will get “dumber”, it’s difficult not to at least recognize those specific challenges, among others.
Deborah Parry do Carmo:
Thanks for the repost, Brian Inkster. If Denis Potemkin can find the right guest(s), it would be an interesting episode.
Clients sometimes ask law firms to provide their ESG credentials through RFPs, and I’ve been wondering today how a firm building its own GenAI would cover that in its response. The responses typically revolve around office energy, single-use plastics, paper use, travel, etc. There’s usually a target to reduce something by a fixed percentage by a certain date. But if they’re building GenAI, they’re consciously doing something with an adverse environmental impact. I’d be interested in what arguments they make to justify it when there are tech firms creating off-the-shelf software that, as Nancy Myrland says, is becoming increasingly customizable.
-+-+-+-+-+-
What did Alex just Waken up to?!
Alex Smith (Global Search & AI Product Lead (Senior Director) at iManage):
What have I woken up to caused by this conversation?!? It’s a good job Legal Service Design conference is happening in Helsinki this week to get us all back to the problems needing solving in Legal … many or most of which do not need LLMs etc … or at least “grounding” all the talk in the actuals of tasks, or processes or moments of truth for the client or ideally citizens. I feel this conversation could go to granular detail next. I’m looking forward to a social feed that week full of design not tech. The irony is that I can’t go to it because I have to deal with the impact of LLMs and everyone talking tech in a three day workshop on tech (which i hope the big tech brings back to problems) … but that’s been the story of the year. I’m hoping Justin North and Virginia Jones let me in on the Whatsapp channel or send me photos of the event/slides. I’m a bit resentful because this year has been tech chat not problems chat … I don’t have much time left so losing time to a year like this (not manifest by not being in Helsinki) does hurt a little.
Denis Potemkin (Founder of Majoto | contracts that people understand and love | faster deals more trust through design and automation):
Alex Smith sorry to give you such a rude awakening! It’s been fun and educational for me and I hope others. There are a couple of moments in the discussion that connect with what you’re saying: (a) Brian Inkster’s statement that law firms and legal teams should focus on getting the basics right (e.g. doc automation) and Nicola Shaver’s statement that the best use cases are additive, where AI is being used to enhance existing tech (which I would read as using AI to help get those very basics right, where the tech was too inaccessible or clunky to be able to do it up to now).
Oh and Brian’s quip that lawyers should be using plain English anyway, not relying on ChatGPT for plain language summaries.
So funnily enough, the GenAI debate takes us right back to the question of real use cases, identifying problems and getting foundational things right. For me, those foundational things are simplification, information architecture, design and plain language. I think GenAI has a role to play here, but as a servant to these things, not a master.
PS happy to help you out with the Helsinki fomo – we’ll find a way of sharing it with you.
Alex Smith:
I have yet to see a public deconstruction of a legal process yet with the right tech, people, process fit and how and where new shiny AI will help. I can’t wait for this to come together with the AI start ups etc and for each to contextually deconstruct all of this. That said i’d not expect this, this year as many have pointed out most of what we see is reaction and experimentation. I’d expect a lot to fall away but hopefully the detail will come as the “breaches in the wall” (ie interest of partners or companies) lead to hard yards of process design (or maybe ChatGPT auto designs processes). Next year should be more fulfilling than this year.
Look forward to pictures from Helsinki and some of the interesting slides/presentations.
Nicola Shaver (CEO and Co-founder, Legaltech Hub | Author | Innovation, AI, Legaltech Leader, Advisor, Investor | LLB, MBA | Fastcase 50, 2021, ABA Women of Legaltech, 2022 | Adjunct Professor):
This was intended to be a fun debate, the brief of which was: Is Advanced AI useful or not, will it have an impact on legal or not. Hence the questions and the format. As anyone who knows me understands, I am a hugely vocal proponent of user focused, problem-first design. Alex Smith himself knows this, we had a conversation just a few days ago about exactly this, he has read the article I published on the importance of UX just last week, we have been at the Legal Design Summit together, he has read a ton of my commentary about problem-solving and how critical it is to dig into the problem, and has contributed to much of it. He knows I am working directly with lawyers to identify problems and ascertain whether AI is necessary or not and whether it would add value or not. He is just stirring the pot, another fun skeptic.
Alex Smith:
Nicola Shaver Just got to this after 7 hours straight of meetings. Sorry if this comment has been misconstrued. I watched the entire debate end to end yesterday (instead of some of the rugby) and greatly enjoyed the “sparring” and the refereeing of Denis. I enjoyed the challenges and insights. I enjoyed the different perspectives from people at the coalface and people at slightly different coalfaces (from innovation teams to wondering how gen AI will fax a Scottish bank). My comment was more around the amount of comment yesterday on this that filled my feed upon waking this morning. And looking forward to the next session here that would be awesome to see at a process level. As you both said in comments, you’re both very sensible people, grounded in reality and I thought you kind of ended up somewhere on degrees of the middle. My comment was not aimed at either of you personally, though I am looking forward to testing ChatGPT on Crofting Law and crofting law processes soon. Your conversation triggered me some of people on the further ends of the spectrum who would not have this debate and do not have grounding in the reality of legal processes and who definitely do not see the nuance of questioning cool vs real world tasks. I look forward to further of these debates and possibly a granular level one … I think the doc auto vs LLM drafting stuff is an area that would pull out a lot of differences and debate that is only happening in DMs and chats. You just had time to scratch that one.
Nicola Shaver:
Alex Smith Thank you for the clarification – and especially, thanks for spending time watching this when you could be watching rugby 😂 . Agree it would be great to see a series of these that dig into the issues at different levels and from different perspectives. Great that Denis has opened the door to real conversations.
Majoto:
Nicola Shaver , Alex Smith we do think there is a series here, digging deeper into some of the topics (buy vs build, doc automation vs LLM, also some of the environmental and social aspects). And then getting Brian Inkster and Nicola back together to ruminate on all the fallout! Would love your input on what the questions in these further debates should look like – we’ll reach out to you.
Further musings on the AI in Law Debate post September 2023
I continued the debate on this blog with several posts after Nicola Shaver and I had our debate.
In October, Legal Geek was unsurprisingly full of AI in law. The search for use cases though clearly continued. The conclusion being “that people in law firms saying they are using GenAI are lying to you!” See: Legal Geek 2023: The one with Withnail and GenAI
Later in October, Richard Susskind was giving his views on AI in law at Strathclyde University. Susskind is, like Nicola Shaver, of course, a Yea. I gave my contrary Nay views to his talk at: The Future of Legal Services: GenAI, Drills and Holes? The Journal of the Law Society of Scotland singled out this particular blogpost as their Blog of the Month in December.
Then in November we had Peak AI Hype at UK Safety Summit
Earlier in December ‘Inkster’s Law‘ was coined by Gav Ward: The 2:1 ratio of time needed to check for accuracy of AI generated material.
I then put GenAI to the test with my law: Inkster’s Law, ChatGPT and Hallucinations
The AI in Law Debate – To be continued in 2024…
The AI in law debate will undoubtedly continue in 2024.
Already this past week alone we have The New York Times v Open A.I. and Microsoft. This is an action for purported copyright infringement by Open A.I./Microsoft for using material from The New York Times to train ChatGPT. It is alleged that, as a result, ChatGPT can now reproduce full articles from The New York Times.
Gary Marcus has pointed out that the ramifications are much wider, with it being fairly apparent that GenAI platforms are being trained using copyrighted images: Things are about to get a lot worse for Generative AI
I decided to do an experiment on that point, given the nature of the featured image on this particular blog post. So I asked Microsoft Designer to give me an image of:
Roy Batty (portrayed by Rutger Hauer) in the 1982 Ridley Scott film Blade Runner, at the end of the film before he dies, when he says “All those moments will be lost in time, like tears in rain…”
It didn’t disappoint:
Interestingly, in October I couldn’t get DALL-E 3 to give me an exact likeness of Richard E. Grant as Withnail for my review of the Legal Geek conference. The same is true today. Perhaps GenAI has not, as yet, been trained on actual images from Withnail and I.
We also have recent revelations that Michael Cohen, a former lawyer to Donald Trump, used Google Bard (an AI assistant like ChatGPT) to provide his lawyer with fictitious legal citations. Cohen clearly didn’t realise the need these days to be a Legal Hallucinatory Detectorist.
Image credits: Roy Batty (portrayed by Rutger Hauer) in the 1982 Ridley Scott film Blade Runner © Warner Bros.
Reactions on The AI in Law Debate: Shaver v Inkster and beyond
On LinkedIn the following comments have been made:-
Denis Potemkin (Founder of Majoto | contracts that people understand and love | faster deals more trust through design and automation):
Brian Inkster, wow amazing. I love that you do that with your blog posts, it really adds depth and colour for the time poor. Thank you for doing it on this topic, the debate really did create some quality side discussions.
I’d love to continue this debate series and perhaps some of the people you mention might feature (I know that a couple of folk already expressed an interest).
Me:
Thanks Denis Potemkin. I hope you do continue the debate in 2024 and, as you suggested before, following debates with others come full circle back to me and Nicola Shaver. There is certainly plenty currently going on, as I highlight at the end of my post, to debate about. You will need to bring a copyright lawyer into the debate mix!
Denis Potemkin:
Btw you’ve picked up a lot of detail including things I remember I wanted to read into but never did, so this is really great. And very clever of you with the tears in the rain piece. One of my favorite movies (and scenes).
Me:
I like detail. ChatGPT doesn’t. 😉 I learned today (using traditional search methods) that Rutger Hauer altered the script himself and added the words “All those moments will be lost in time, like tears in rain”. Link to source material from those last 4 words the first time that they appear in the blog post.
-+-+-+-+-+-
Nancy Myrland (LinkedIn™️ Coach For Lawyers | Legal Marketing & Business Development | Content, Social Media, Podcasting, Video & Virtual Presentation Consultant | Individual & Group | Speaker, Trainer & Advisor for Lawyers & Law Firms):
Honored, Brian. Thank you! I have to admit I was a little worried when I saw that you tagged me, wondering if I said anything intelligent. 😅
Me:
Always a pleasure Nancy Myrland. I have, however, been known to quote people who would rather I hadn’t! 🤐
Nancy Myrland:
So far, I feel safe, Brian Inkster!
Me:
You need to hope it stays that way 😂
Nancy Myrland:
I will do my best to color within the lines!
-+-+-+-+-+-
Philip C. Hanke (Fusing Law and Technology to Drive Innovation | Director of Publishing at Weblaw | Legal Tech Educator):
“I’ve seen things you people wouldn’t believe”
Me:
If it was using ChatGPT then I could believe that 😉
-+-+-+-+-+-
Alex Smith (Global Search & AI Product Lead (Senior Director) at iManage):
“More human than human” is our motto.
Me:
Has Elon adopted that one yet for Tesla Bot? 😉
-+-+-+-+-+-
Graeme Johnston (Software to map work – before that a lawyer):
“But [comments] are like poppies spread,
You seize the flow’r, its bloom is shed;
Or like the snow falls in the river,
A moment white – then melts for ever;
Or like the Borealis race,
That flit ere you can point their place;
Or like the Rainbow’s lovely form
Evanishing amid the storm. –
Nae man can tether Time nor Tide,
[Except when Brian captures it all on thetimeblawg.com]”
Me:
Very good Graeme!
-+-+-+-+-+-
Frank Thomas (ASEAN legal trainer. Former BigLaw thriver and survivor. Proud husband and father of 3 Alpha humans. Steadfast skeptic of new Tech that overpromises and underdelivers..):
I like Nicola a lot- she lives in Brooklyn I think, where I was born.
But I vote you up Brian- a balanced prudent approach to random reckless toys- that we can fiddle with, but not EVER partner.
We are the stewards the law lords of our domain. We have those unique competencies that clients pay high dollars (or pounds) for.
Me:
Thanks Frank Thomas. I like that: “random reckless toys- that we can fiddle with, but not EVER partner.” Certainly a lot of fiddling going on with them at the moment!
-+-+-+-+-+-
Roquilly Christophe (Full Professor. Directeur du EDHEC Augmented Law Institute. Doyen Honoraire du Corps Professoral. Honorary Dean of Faculty chez EDHEC Business School):
I know now why you cry, but it’s something I can never do
Me:
Whereas ChatGPT hasn’t a clue why we do anything we do! We have a long long way to go to get to the Terminator’s level of understanding. Don’t listen to the legal futurists who say we will get there by this time next year 😉
Roquilly Christophe:
Agree and usually when I listen to futurists they already belong to the past.
-+-+-+-+-+-
Clare Fraser (Product developer – building LLMS using public domain legal data to support access to justice):
Ahem point of order Mr Inkster. The use case of translating legalese – yes lawyers should be doing it but they often don’t. Not you of course.
Me:
Those lawyers are unlikely to understand the output 😁
