I know that you can’t start a story with a date, but in this case, precision is necessary. On November 28, 2022, I attended a conference on the application of technology to the legal world invited by my friend Laura. She told me “I’ll leave you a corner and talk about what you want. The only condition is that you come without a computer”. As a good Madrilenian, I signed up for a bombardment with two pipes so I asked him to tell me what a corner (it turned out to be a corner in which to sell knowledge like the one that gives perfume a smell) and what were the hyper-innovative topics on which they were going to talk: Kanban, processes, brainstorming with improv cardsLinkedin Legal Selling, customer journey and things like that. She still didn’t really know what to do among so many Anglo-Saxon proposals. Laura told me that she could talk about my big mistakes or a story of overcoming. The first proposal was overwhelming for me and the second one corny, but she put me on the track of what to choose. I remembered the tweets that showed that GPT wrote better football chronicles than Rajoy or those of writers and scriptwriters who were astonished by this tool. So I proposed to do something that no one was interested in except me: tell my unbiased experience using GPT to write legal texts. I cheated on another friend, Maite, with whom I shared a moment of closeness as if it were a piece of micro-theater. Despite setting up a presentation à la Pimpinela (Maite was the prudent one and I was the techno enthusiast), we didn’t have a great acceptance. In the moments of greatest dramatic tension, we did not reach five people, and this despite the fact that the result of the experiment was much better than I ventured when I proposed the topic. The GPT Playground in its version 3 back then, using Davinci as its engine, produced some reasonably decent leases (even with a certain humor and context), some pitiful lawsuits, but absolutely formidable contracts and legal texts in English. It was the first time in the legal world that an artificial intelligence spoke to us in Spanish. All the available payment products are Anglo-Saxon and required, until now, an unpaid training effort that was well above the performance that was obtained. The language barrier and a different legal system had been protecting us from the technological assault on the legal profession. I understood that that advantage was over.
I concluded from that tiny corner of the world that generative, cheap, accessible and quality AI in Spanish was going to change the legal world in the medium term and I went to a classical music concert. As a good headline would say clickbait“what happened next will surprise you”.
Two days, two, after our performance, OpenIA launched ChatGPT and not a day has passed since then in which at least one legal event -webinar-coffee with legal cakes has not been held in which the challenges, threats and opportunities of GPT for the legal world are discussed. Not to mention the millions of messages, articles, tweets and chat experiences with this intelligent bot, crowned with the disturbing experience last February of the integration of GPT in its version 4 with the Microsoft search engine, Bing, to which journalists turned into a terrifying and egomaniacal AI in the style HAL from “2001: a space odyssey”. ChatGPT was the cover of the Time and generative AIs have turned, who would have thought, Microsoft into the incumbent, the modern one, in the search engine market in the face of the consolidated Google, who, hurried by events, announced its own version, Bard that was released wrongly loudly in the reply that was used as advertising. Bad start for its AI and its internal review systems, which led to the collapse in Google’s price due to the blunder. The penultimate chapter of this mad race was the presentation of GPT4 at the beginning of March with new capabilities, such as generating a website just by “reading” a pencil drawing or combining text with images. To continue disguising that they are a company that left philanthropy to seek profit, OpenAI published a paper with all the novelties that was more like an advertising brochure than a scientific text.
of this hype I draw several conclusions. The first, that I am a visionary who will never take advantage of her visions. And the second, that we are a disaster in predicting tsunamis despite the fact that they have been announced for years. Open AI, the owner of Generative Pre-trained Transformer, GPT, was founded in 2015 as a non-profit entity that ceased to be a non-profit when Microsoft put it out of poverty (it consumed money like a coal locomotive). Precisely this consumption of resources, together with mediocre results, called into question its viability and in us the doubt as to whether this hectic scenario was going to be possible in the short term. But, as we all know, it has been. That it caught everyone in a cherry is surprising, but that Google, with its resources, was in the thick of a fig tree, says a lot about this tsunami effect of slow-cooking but critical-impact technologies. As much as we love it, we have a hard time putting ourselves in improbable events, especially if they are contrary to our business.
Returning to the legal world, in this term, in addition to animating numerous gatherings, GPT3 passed the US Bar Access Exam by scratching (although he nailed it in his GPT4 version), one of the most demanding, in the world and has already been used in the drafting of sentences. For me, this remains an anecdote. For now, access not only to extensive databases but also to a correlated interpretation in natural language calls into question our system of excellence based on memory, from one’s own career to the system of opposition to the superior bodies of the state. But it is also that GPT has an API, a universal connector, that allows us to “place” this AI behind any service, from a WordPress blog to an automatic legal consultation system. I know that you will tell me that you still make many mistakes, but it is that GPT, the tool, not the chat, allows you to be trained with your own information repository (finetuning), for a ridiculous price and with a non-existent technical barrier if it is done using any of the Microsoft services. I can turn GPT, my website or my app, into a divorce expert and let you draft the demands with minimal supervision. The galactic office within the reach of anyone, even those who are not lawyers. The wild opening of legal services without lawyers involved. Lawyers, hopefully, will become the human interface for AIs, which will only have those who can afford it. A society divided into two classes: those who can afford privacy, security and dealing with a human, and those who cannot. And, up to here, my prediction for today.
Those who know the guts of these systems, like magicians who know the trick, are astonished. For them, it is nothing more than a relational system, some stochastic parrots that drop correlated words without understanding what they are saying. The question is that if a parrot is able to effectively replace a significant part of the work of lawyers, what does that say about us?