What family lawyers need to know about ChatGPT

If you've felt slightly giddy watching the capabilities of AI explode over the last few months, you're not alone. As a founder at adieu.ai, it's my job to be across this stuff, and I can hardly catch my breath between announcements.

To help you make sense of all this I wanted to put together something short and sharp to cut through the noise and get straight to what it is, why it matters, and what it means for family lawyers. Most of my perspectives are based on first-hand experience and data from more than a thousand matters in our private beta, where we are developing autonomous AI paralegals to assist family law firms.?

What it is

Don't mistake ChatGPT for a chatbot. It's actually a Large Language Model (LLM) under the hood, which is immensely more capable. Although chat happens to be the most convenient interface to it at the minute, the model sitting underneath can do a whole plethora of things, many of which haven't yet been understood or even discovered.?It's also not the only game in town. Whilst OpenAI's ChatGPT is the one you've seen in the headlines, Google actually pioneered the core approach (Transformers, which are the 'T' in GPT), and there are a multitude of AI labs big and small currently engaged in an AI arms race for dominance.

In its essence, a large language model is an autocomplete system. LLMs aren't new, but what they've been able to do once they reached a certain size caught even the most optimistic AI researchers by surprise.?Although these language models are exceptionally good at understanding natural language, it is the 'G' in GPT which makes things really interesting: Generative. GPT actually creates things from scratch that didn't previously exist. That means it can write poetry and jokes, draft letters, construct arguments, and summarise documents with astonishing detail. GPT-4 (which was only released last week) passed the bar exam in the 90th percentile (ahem), got a near-perfect score on the SAT, and can do complex calculus. It can find bugs in my code, lecture me on what I did wrong, and even write its own code in pretty much any language.

It's important to understand that GPT is fundamentally a reasoning engine, not a database of facts. This means that - much like us - it sometimes reasons its way to conclusions that are incorrect. The AI term for this is "hallucinating" and many people have been shocked by the confidence and earnestness with which it will try to convince you of something that is patently false; but that's because they misunderstand the nature of this new paradigm, which reasons its way forward rather than simply regurgitating stored facts. And indeed, this is the very thing that makes GPT so interesting: its ability to perform general purpose tasks.

Why it matters

This technology is ubiquitous and incredibly capable. This is going to be everywhere very soon, and some of it will be very, very powerful.

Two weeks ago, OpenAI made the model that runs ChatGPT available to any developer to use. We are currently in those eerie, quiet minutes before the tsunami hits. Over the coming weeks and months, everything you touch and interact with is about to become AI-powered in some way. Most of this will fall somewhere on the spectrum between useless and irritating, as every company on the planet scrambles to stuff AI into every product crevice they can. Some of it will be immoral and appalling as the scammers and haters are given the digital equivalent of nuclear weapons. But some of it will be wonderful, solving real human problems in ways we couldn't have imagined only a few months ago.

This ubiquity isn't just in the products you use though. Anyone who can type has free and direct access to this technology right now (if you've not yet signed up and used ChatGPT yourself, this is your cue). This means your clients are using it. This also means your competitors are using it. There is no opt-out button for this - your clients and competitors have already made the choice for you.

Let me give you a stark illustration from my own life. In the space of the last 10 days, the way that my children approach school has significantly changed. My wife is an occupational therapist who helps children develop core skills for school, and we've developed a way to use ChatGPT as a specialist tutor in each core subject. Our kids now have one-to-one tutoring everyday, in every subject. They can ask as many questions as they need, and use their own homework as the material that their tutor works through to further explain concepts, provide examples and test their understanding. The quality of the tutoring being provided is remarkable and available immediately as each new concept is being learned. We're only 10 days in, but you'd have to be pretty bloody-minded to think that this isn't going to have an impact on how we approach schooling, both over the next 10 weeks, and the next 10 years. [Do let me know if you'd like to offer this to your own kids - I'd be happy to share it].

What it means for family lawyers

Ok, explaining what this is and why it matters is much easier than working out what it will mean for family lawyers; but having observed lawyers and their clients using AI across more than a thousand matters, I can probably give it as good a shot as anyone. We have a saying in the startup world for working with things that are uncertain and emerging: "have strong opinions, loosely held". This is the posture I'd recommend taking if you want to end up on the right side of history. It's also the reason that my views on this will almost certainly evolve as things unfold. Here goes...

1. You should be using AI now

And I mean now as in today rather than this month. Right now you can use both general AI capabilities via the public ChatGPT interface, and specific capabilities embedded in AI-enabled products like adieu.ai. The number of things you can do using the public ChatGPT interface is already immense, and your key challenge is to work out how to leverage it for your firm. To help with this I'll let you in on two little secrets:?

1. Getting AI systems to do what you want them to is primarily about trial and error. It's not pretty and it's probably not what you imagined - but there you have it folks.
2. You use human language to make language models do things.?Not code; but good old fashioned English words. Now if only there were a professional group with a strong command of language who specialised in clear and unambiguous written communication - imagine what they could do...

Fork out the extra twenty bucks a month for ChatGPT Plus - this will give you access to GPT-4, which is far more powerful than the free version's GPT-3.5. The sooner you start to explore and iterate, the sooner your firm and your clients are going to benefit.

In terms of specific AI-enabled products, at adieu.ai we're developing autonomous AI paralegals in a private beta and the results we've seen over more than a thousand matters have been nothing short of transformative for our firms. This isn't yet open to the general public, but we're always happy to talk with new firms who are interested in joining the beta or the waitlist. We were doing this stuff long before it became cool, so are a fair way ahead of the curve, but you'll no doubt start to see plenty of new startups and old incumbents announcing all kinds of things over the coming months.

2. Treat AI as a junior

One of the reasons that I'm so bullish about getting your hands dirty is that it's not hard to do it in a responsible way. We already have a working model for how to leverage the work of others: supervision and oversight. Simply treat the AI as you would a junior staff member, and provide the appropriate checks and balances. I will however warn you that I often find this relationship flipping into one where I become the protege and the AI my mentor; but unless you have nothing left to learn I think this is actually healthy - just as long as you remember who is ultimately responsible for the work.

3. AI will work in client-facing roles

This is already happening today in adieu.ai. Forget any cold robot-voice cliches that may come to mind - the way our AI paralegals are able to communicate with clients is so warm and empathetic they put most humans to shame. They also have awareness of the firms in which they work and the communities they work in, and draw on this when talking to clients. Our average conversation time between AI paralegal and client is just under 10 minutes, and this usually gives the paralegal enough material to get the client to initial disclosure with no further input needed.

Utilising AI in client-facing roles comes down to great character design and a keen awareness of client-experience, so those firms with a coherent brand will fare better than those who are more vanilla.

4. Some of the gains will be very significant and very soon

We've already seen a 30x improvement in the elapsed time from engagement to initial disclosure in our private beta, and have a goal to reduce the time to filed Application for Consent Orders by a factor of 10. These are staggering numbers and you'll find it hard to compete with firms who can offer this to clients if you can't.

5. Boutiques will benefit more than big firms (controversial!)

This may sound counterintuitive, but it's something I've observed over many matters. When you have technology that is equally accessible to big firms and boutiques, boutiques will generally benefit from it more. Why? Because they're simply more agile and less invested in the status quo. One of the reasons this feels counterintuitive is that up until now, it was the big firms who made the most noise about innovation, sponsored the events, and could actually afford the technology. But when everyone has access to something, what counts is what they do with it, and for whatever reason, boutiques just tend to do more.

This has been a surprise for us - but many of the firms embracing AI the most in our adieu.ai beta have turned out to be small firms in regional areas of Australia. Whether this is peculiar to AI paralegals or a more general trend in AI remains to be seen.

6. Regulation will shield us from the more extreme downsides

This is as much about regulation and culture as it is about technology, and for better or worse (probably a bit of both), regulated industries aren't as sensitive to technological changes (in fact, recent data from the US shows that while technology has generally driven down the cost of things since the year 2000, the cost of service in regulated industries such as healthcare has instead risen significantly despite huge technical advances).

As little as 12 months ago it was a truism that those working in the creative industries were safest from AI and automation - but the last 9 months have been a bumpy ride as they've watched AI image generation technology seemingly come from nowhere to produce everything from oil paintings, to comic books, to hyper-realistic photographic imagery (the new version of Midjourney was released yesterday and its images are stunning).

We're simply not going to see that kind of disruption in the legal sphere - and that's a good place to be. It means that you can be clear and purposeful in how you use AI to benefit your firm and clients without worrying whether you'll be driving an Uber in 12 months time because human lawyers have been made redundant.?

7. We won't see job losses in family law any time soon

A new study was released yesterday with the conclusion that GPT will significantly disrupt nearly 20% of jobs which "may see at least 50% of their tasks impacted. The influence spans all wage levels, with higher-income jobs potentially facing greater exposure". You'd be excused for feeling some level of anxiety reading this, as it isn't merely speculating about what might happen in the coming decades; this is a study on the current capabilities of GPT.

I've been watching the impact of AI on firms for nearly two years and I can tell you the number of jobs lost so far during our private beta programme: zero. The reason for this comes down to the definition of "exposure" in the study, which doesn't equate with job loss: "exposure is defined as driving a reduction in time it takes to complete a task by at least 50%... we estimate that GPTs and GPT-powered software are able to save workers a significant amount of time completing a large share of their tasks, but it does not necessarily suggest that their tasks can be fully automated by these technologies."

When AI paralegals start performing the most tedious and time consuming work in a firm, the team isn't left twiddling their thumbs; they simply get on with doing more valuable work, of which there is no shortage. I don't have the data to speak with as much confidence when it comes to other types of legal work, but I suspect the pattern will be similar for some time to come.?

8. Smart, resolution-focused firms will actually be more profitable

Again, this may not be obvious at first glance, but we've seen it consistently in the adieu.ai beta. This is simply a matter of leverage - you're able to create more value with less inputs.

One of the reasons that it isn't immediately obvious that AI should lead to more profitable and sustainable firms is the long shadow of the time-cost billing paradigm, in which revenue can only increase by having more people spending more time on something. We've already seen a steep rise in the popularity of fixed-fee billing recently, and AI is going to force this issue even more. If you invest in technology that enables you to provide the same or higher quality work for your clients at a lower internal cost, is that work less valuable to your clients?

As an enterprise whose mission is to enable separating families to reach resolution 10x sooner, we have a very particular view on how this new value should be distributed: Firms should make more and clients should pay less. That's not a typo - we aren't simply redistributing the existing value here, we're creating new value which can be shared in a way that benefits both firms and their clients. The numbers we've seen in the beta are that firms make 2-3 times more profit on AI-assisted work, while at the same time offering clients fees that are 20% lower. That's the sweet spot we want to see: firms becoming more sustainable, and providing better outcomes for their clients.

Not everyone shares my views on this, including Allen & Overy who in announcing their new GPT-based drafting AI were very clear that it "would not reduce billable hours or save money for the company or clients" while in the next breath stating that it would be "saving time at all levels". I appreciate that their PR team doesn't want to attract the wrong headlines, but clearly the benefits will be accruing to someone.

To wrap

These times are as exciting as they are uncertain, but I think we're going to be more than ok.

Follow me or connect on LinkedIn to hear my insights and learnings as this unfolds. You can have a peek of what we're doing at adieu.ai and chat to us about joining our private beta if you're interested. If you're curious about using the AI-tutor framework I mentioned above just let me know. If anything I've written has piqued your interest and you'd like me to write on it further I'd love to hear about it.