background

Principles for a Legal AI Future

Uwais Iqbal2023-04-14

We’ve put together 5 principles for navigating the future of Legal AI to help you and your teams make sense of everything that is happening and will happen with AI in legal.

🖐 Our 5 principles for navigating the future of Legal AI

🔥 There's been a lot that has been happening lately in the world of AI and legal. It's super exciting to see that Legal AI has actually become a defined category! There are some real shifts taking place in attitudes towards AI across the legal industry.

🤷 The pace of developments has been unrelenting. It's been really hard to keep on top of the daily model releases let alone make sense of what this all means for us in legal.

🤔 We've been doing some hard work thinking and trying to make sense of everything. We've come up with 5 guiding principles to help you and your organisation navigate the future of Legal AI.

1️⃣ Your data is your moat

The only real protected asset of value when it comes to AI is data. Everything else in the supply chain of producing and consuming AI will become commoditised.

We're already seeing falling hardware costs for compute and advances in efficiency with new algorithms. Today, you can train a BERT model from scratch for $20. In 2018, it would have cost around $10k.

Along with developments in the open-source community the cost of creating AI will fall significantly to the point where creating the best-in-class AI will become universally accessible. The only differentiator that will remain will be data.

Law firms and legal teams need to realise this and start getting their data into shape.

2️⃣ Narrow AI over General AI

There is an unspoken assumption in the research and AI products coming out of big tech research labs. That bigger is better. Most of the current AI research is plagued by this bias of wanting to build General AI. It's kind of in their best business interests to get to building a General AI first so they can maintain market position. Whether we can even get to General AI is still just a hypothesis.

What happens if you let go of this pursuit of General AI? Can Narrow AI models focused on a niched domain with a singular task out perform a trillion parameter model?

For us in the legal industry we are at an advantage in that legal language tends to be much more structured, regular and predictable than general language. We can niche down our models and take advantage of the nature of legal language to build smaller, more efficient and effective models.

3️⃣ Know your Jobs

There will always be a new model just around the corner. Today, the talk of the town is Generative AI and Large Language Models. Tomorrow it will be something new. Technology will always change. Your jobs won't.

The jobs legal professionals have to perform are stable over time. Getting smart and knowing your jobs will mean that you can cut through the hype and effectively evaluate if a new breakthrough in technology will enable you to do your existing jobs in a better way.

4️⃣ Education is the enabler

The real barrier to adoption is AI illiteracy. We are plagued by misconceptions, misinformation and things we have learnt about AI from the marketing material of AI vendors and the headlines of sensational media articles. The apparent complexity of the technology makes it easy to fall into fallacies and mistakes in thinking.

We tend to fear what we don’t understand. Technology is already complex enough and AI shifts how we look at the world.

We are all using the same buzzwords but are speaking entirely different languages. The first step to getting ahead in adopting and embracing such a radically novel and revolutionary technology like AI is to develop the language and ideas so we can think about and speak about AI in a meaningful way.

Undoubtedly, there are endless applications where AI can be applied in legal but we will never be ready for that revolution if we speak different languages. “The Revolution will be complete when the language is perfect.” - 1984, George Orwell

5️⃣ Appreciate the event but master the process

An AI model is a process not an event. An AI model is an end-product. It's a result of a process brining together different components in a particular way and combining them. This process isn't magic. In fact, it's actually repeatable!

In the past few weeks we've had ChatGPT, LLaMa, Vicuna, Alpaca, Koala and Dolly to name just a few. There will be always be a new event with AI just around the core - you won't be able to keep on top of it all but that is ok.

What's important is being able to master the process. If you can master the process of creating AI rather than obsessing over the latest event, you'll be able to see the real opportunities. If you can master the process, you can bake in your needs and concerns around data privacy, security, transparency etc rather than being at the mercy of a third-party.

💭 Closing Thoughts

We are in a position to leverage AI to take the legal industry into the future. We are at a real pinnacle moment in history with technology and language. For the first time, text data is as computational malleable as financial data or image data. This will have a significant and lasting impact across the legal industry. Making sure we get our principles right will help guide us so we can still progress forwards while navigating the winds of change without getting swept off of our feet.