Capability, not customer: How Travers Smith is protecting itself from AI degradation

IP claims threaten to weaken AI products – could going model-agnostic be the solution?

Law firm Travers Smith is building an AI model-independent UI to avoid vendor lock-in – and the fallout of looming IP claims.

Generative AI rocketing into the public consciousness spurred businesses worldwide to make bold claims about their artificial intelligence investments this year. Was it exciting? Yes. Did it make headlines? Definitely. Did the benefits match the claims? Debatable.

Companies from Coca Cola to Octopus Energy are using the technology behind ChatGPT to enhance their own efforts, but Travers Smith CTO Oliver Bethell argues that that is setting yourself up for failure - despite his company also using the GPT models.

"Our view is that there's going to be a need to be able to be model independent, because of the [IP] claims that are going be coming towards OpenAI and all of these other vendors around copyrighted data."

You might not expect to turn to a legal firm for AI innovations, but Travers Smith is building a specialised team from the ground up because the senior leaders think AI will have "a fundamental impact" on the sector in the long term.

"We need to have the capability in-house to be able to build and deploy the product, rather than just being a customer of it."

The company has already open-sourced its first product, a GPT-based chatbot interface that businesses can use without giving away their data.

Copyright, claims and court

As well as need, the company was also driven by differentiation - "We want to make sure we're not buying off the shelf products where we're only as good as our competitors" - and concerns about copyright.

"If, with any content that you create through GPT, it's possible to determine that that model has been trained on copyrighted data, it's possible that your output may contain contamination from that. So, could that be deemed copyrighted content?" Oliver asks.

"Lots of law firms are queuing up to buy Harvey, but Harvey is an OpenAI startup. So, it's baked in using the OpenAI models, that's not going to change. If that model degrades as a result of the copyright infringements, what impact does that have on those products that are all built on it?"

It's a good question, and one that still has to be answered. Generative AI is a legal grey area, and there is no precedent in courts to suggest how any future ruling might go.

It goes without saying that no CIO wants to see important systems crumbling as the AI model they use falls victim to copyright claims - and yet, plenty of products coming to market now are "just thin UIs on top of the underlying model." That makes them vulnerable.

"We think we're going to need to be able to swap out that model for different ones and potentially build our own...with clean data. I think provenance of data is going to become really, really important."

A new approach

Despite utilising it themselves, Oliver admits that the chatbot interface is "fundamentally quite limited."

The aim in the next iteration is to provide more functionality, higher scale and "orders of magnitude greater processing," which will require a variety of different user interfaces. Plus, of course, model agnosticism.

Computing sat down with Oliver to see an early preview of the newest product, known as Analyse. As the name suggests one of its main functions is document analysis, which is in high demand in the legal sector.

"A BD marketing exec or manager might take couple of days to sift through 30 or 40 pitches [to clients]. Actually, they can run this and then within minutes get the data back and start to construct a report."

Knock-on effects

The tools legal teams use today require a high level of sophistication and knowledge. What happens if a vendor produces a product that allows in-house teams to use them directly - cutting out the legal sector from certain aspects of work?

"That is a real risk," says Oliver. "It could go one of two ways. It either means that work goes to software, or you could end up in a Jevons Paradox, where the cost of delivering the services drops to the extent that actually there's lots more work to do."

Much like IT leaders at Google Cloud Next last week, he expects the human skills that can't be automated to become much more highly valued in the future.

"Project management, negotiation, relationship management - all of those things will become increasingly necessary. And they will just be supported by teams of lawyers using technology assistance."

The possibilities around AI are seemingly endless, but it could also be the biggest shift since the internet went mainstream. By building its own capabilities, rather than relying on larger tech vendors, Travers Smith seems to be taking the right approach to handle the coming disruption.