Simplifying contract reviews
She highlighted IronClad, a tool that uses OpenAI’s latest release GPT-4 to simplify contract review, as a prime example of how AI can speed up tedious tasks for solicitors.
“For a lot of contractual agreements, there is very little time to turn them around, particularly when you’re working in-house. This tool is life-changing in terms of speeding up contract review. It allows you to ‘redline’ numerous contracts to align with a company playbook, and will match your requirements to what is set out and show the changes you need to make,” she said.
“A big caveat is that any output will always have to be reviewed and checked by professionals. This type of tool brings the role of a lawyer into a whole other sphere. It is very much about having the skills in terms of quality input, as well as knowing how to analyse the output for accuracy.”
Leveraging AI
Chris Murnane (legal solutions director at Johnson Hana) said that start-ups were currently trying to leverage AI to compete with IronClad, which is the biggest global player in the area of contract-management software.
He referred to Luminance, which was built out of a proprietary large-language model (LLM) at the University of Oxford. It has developed a fully-automated, AI-powered contract-negotiation tool for nondisclosure agreements, called ‘Autopilot’, which has featured on BBC News.
“A lot of what we have seen so far is technology and AI being used to streamline and amplify what lawyers do. Autopilot is an example of lower-value work being taken off lawyers altogether. Everything below high-level, master service-type agreements in-house is ripe for automation,” said Murnane.
Dublin-based Johnson Hana is operating a new outsourcing model that can realise and release more value for lawyers.
“It comes from the idea that you can disaggregate legal advice from the process. We look at in-house legal teams – and increasingly law firms – that are spread too thinly and resolve that through bespoke legal solutions,” Murnane explained.
Easy peasy
At the summit, Murnane shared how Johnson Hana used AI to examine around 1,000 contracts for ‘repapering’ for a corporate client. (‘Repapering’ involves reviewing and carrying out risk assessment and remediation of contracts and other documents in order to comply with shifting company policies or regulatory changes.)
“We ended up negotiating 300 contracts within three months – a process that would otherwise have taken that client a year to do,” Murnane said.
A significant development he has observed in the past year is LLMs no longer needing huge datasets to be able to train documents.
He cited eBrevia, which can now train on a set of a minimum of seven documents, which means it can be used for a small due-diligence of 20 or 30 contracts, for example.
“It will be up to young solicitors to be early adopters and champions of these types of technologies. Demonstrating the efficiencies they bring to your team may convince partners to use them. This will free you up to do something more exciting, or place you as the tech-savvy lawyer in the team,” he said.
Generally, AI tools are geared towards in-house teams but, in Murnane’s view, some will be adopted by law firms.
“A lot of larger international law firms are developing LLM tools that rely on their own internal datasets. I believe the future in the legal profession is LLM models built on internal knowledge – not pulling data generally from the internet,” he predicted.
Data crunching
Paula Fearon (head of project services at McCann FitzGerald) told the summit that it is actively engaging, piloting, and testing generative AI tools to identify where they can assist the firm in the future.
Fearon is part of a specialist group at the firm set up in 2015 to streamline how it deals with bigdata projects.
“Unlike previous tools, generative AI has the potential to have application in almost every aspect of our practice. It’s not possible to deal with the volume of data that clients are creating in the traditional way, and it’s important to stay informed on where we can rely on technology to do the heavy lifting,” she said.
“We currently use e-discovery software most often, as it has been around the longest. As it has reached maturity, we have found use cases far beyond what we originally used it for. There is a full range of AI and other tools available now on e-discovery platforms, which allow us to generate a robust and consistent output much faster.”
In the can
One of Fearon’s favourites is an imagelabelling tool, which scans all of the images and photos in a dataset and names them, based on the AI’s interpretation of what’s in them.
“I haven’t seen a dataset yet that hasn’t contained holiday snaps or baby pictures. There’s a real trend towards professional and personal communications being intermingled. So being able to access tools to identify swathes of irrelevant material is really valuable,” she said.
Increasingly, technology is directly having an impact on the types of work McCann FitzGerald is being presented with by clients – which essentially means dealing with a new legal landscape, Fearon noted.
“We’re seeing whole businesses being run on TikTok, contracts being concluded on Instagram, and customer service being deployed via FaceTime. Data extracted from smart devices, such as Alexa, have become part of the evidential pull in cases. Emojis are now routinely used in the professional world, and have even found their way into the courtroom. In one recent US case, a specialist was appointed to help the judge to interpret emojis,” said Fearon.
Eating up the internet
In his presentation, Barry Scannell (consultant at William Fry) told attendees that there was a “huge amount” of copyright-infringement litigation going on at present in relation to AI, particularly in the US.
“We’re at the very start of the AI journey, and intellectual property is the biggest issue from a legal perspective. On the input side, where AI is being trained to basically eat up the entire internet, most cases are in the US. All claimants are essentially saying the same thing: that the use of their data in the LLM training was reproduced without permission, and that is copyright infringement,” he explained.
“Text and data mining (TDM) often involves copying large amounts of copyright material. The problem with some datasets released through this process is that they have been used for purposes that are 100% commercial.
A lot of AI companies are getting themselves into hot water over that,” Scannell says.
There are various TDM copyright exceptions around the world that AI companies rely on to avoid litigation, Scannell noted.
In Europe, for instance, commercial TDM is permitted under the 2019 Directive on Copyright in the Digital Single Market, provided the works are not explicitly restricted by the rights-holders.
Legislative progress
At the summit, both Scannell and Redmond discussed the status of the EU Artificial Intelligence Act and its implications.
Political agreement was reached in December that it will be based on a risk-based, tiered system, where high-risk AI systems will face stringent regulations, and that there will be mandatory impact assessments.
Non-compliance with the rules could lead to fines ranging from €7 million or 1.5% of global turnover, to €35 million or 7% of global turnover.
“A lot of organisations will have to carry out impact assessments. This will mean strategic shifts for businesses. It is going to increase their operational costs and administrative burden, and there will be massive sanctions for non-compliance – worse than for non-compliance with GDPR,” said Scannell.
“Now that political agreement has been reached on the main part of the act, it is going to have to be tweaked to get to the detailed technical requirements.”
When asked for her view on the AI Act, Redmond said that “so far, it has frankly been a bit of a mess … It has been so stop-start. We were all shocked and surprised when political agreement was reached. There is clarity in relation to certain parts of the text, but the core part covering high-risk models is still up in the air, so I’m on the fence about it. I expect we will have this text this year, and the act may be enforceable in late 2025.
“Europe is aiming to be the first in the world in regulating AI and replicating what happened with GDPR in how it influenced data-protection regulation in other jurisdictions.
“When it comes into effect, the AI Act is going to be a total game-changer. The same kind of concept and processes under GDPR will apply, but in a different way. The high-risk analysis required under the AI Act is almost like the GDPR’s data-protection impact assessment on steroids.”
Sorcha Corcoran is a freelance journalist for the Law Society.
FOCAL POINT
IN-HOUSE VERSUS PRIVATE PRACTICE
During a panel discussion on the rise of the in-house function at the Future of Legal Practice Summit, attendees learned about the key differences between working in-house and in private practice.
“When you’re in-house, you’re part of the business and part of the solution. You get to see things through, from start to finish. In private practice, you have to manage the expectations of lots of clients and make sure you hit targets in terms of billable hours,” said Sarah-Jane Clifford (senior legal counsel at the Coca Cola Company), who has experience of both sides.
“You don’t have to produce long memos of advice for in-house clients. Speed of response is key. Clients just want you to point out the risks involved so they can make a business decision and move on quickly. Traditionally, in-house lawyers have been generalists, but this is changing as regulation increases. There are now more specialists, in areas such as privacy and environmental law.”