As I was drafting this piece, Cloudflare experienced an outage that froze the AI tools I needed for background information. The outage lasted several hours.
Imagine if I’d been working on a brief due in an hour.
While the legal profession debates AI hallucinations, privacy risks, and the adoption curve, some fundamental threats that lurk beneath the surface are being ignored.
The fact is AI development could stall and systems malfunction, as explored here. And as explored in a future post, is it possible we never get to the point where the efficiency gains from its use fully offset the time needed to verify output?
Legal in particular risks overreliance on something that may not always deliver. Like the people of Pompeii, many sit blissfully unaware of the dangers from believing the volcano above us will never erupt.
Where We Are Today
Here’s where we are today.
- Vendors touting the magic of AI, developing ever increasing and more sophisticated programs running off AI platforms to do more and more mysterious and mystical things.
- Pundits and commentators urging legal adoption at warp speed.
- Predictions that AI will disrupt what lawyers do and how they do it by a generation that barely has any practical or hands-on experience.
- A belief in an ever and exponential expansion in what AI can do.
- Billions of venture capital dollars being poured into legal AI companies and farcically inflated valuations.
- Law firms investing million in AI systems that replace core functions and ways of doing things.
There’s an ever-increasing reliance on these AI systems to practice law and an assumption that these systems will not only work flawlessly but exponentially get better.
But to make all this AI work you need vast numbers of computing chips. You need data centers that are expensive and time consuming to construct. You need the energy to power these huge data centers. More energy than the existing systems can supply. That means new power plants.
The Numbers Just Don’t Support Complacency
Some numbers from a recent Business Insider article entitled The AI Bubble You Haven’t Heard About highlight this potential coming infrastructure crisis.
- Data center developers have proposed 26 projects to major US utilities which they say will require some 711 gigawatts of power beyond present capacity. That’s almost the same amount of power demanded by the continental US at the summer peak.
- If every already permitted data center came online the electricity demanded could be as high as over 293 terawatt hours. That’s about the same amount of power needed for the state of Florida for a year.
- There is simply not enough energy capacity to power all the semiconductor chips the data centers say they will need.
Put simply, there’s an infrastructure gap between the energy, the data centers, and chip capacity and the coming need.
And closing that gap won’t be easy. As a former utility lawyer, I know you just can’t clap your hands and say here’s more power. (There is nothing longer than a state Public Service Commission proceeding to approve a new power plant. It’s like watching paint dry.) Increasing capacity means building new utility plants. It takes time to get regulatory approval to construct these plants and time to build them. Lots of time.
And all these estimates are based on the assumed continuing development of AI and increased needs. If both plateau, then we are left with a huge amount of capacity but stagnant demand.
It’s a recipe for a volcano eruption for which we aren’t prepared
And Then What?
The impact of the shortfall could be huge.Think of a fast-growing city. The new population inundates the existing road infrastructure. We all know the result. Think Atlanta, Nashville, LA, Miami. You sit in traffic bottlenecks for hours and hours, frustrated. Wasted time and inefficiency.
The same is true for AI. If the data center and utility and chip capacity can’t meet demand, the AI systems we all depend on will be slowed. Or worse, the existing capacity will be throttled (think brownouts when capacity is reached on a hot summer day). Like the outage that frustrated my work on this article, AI systems could just stop working for a time. Certainly, models would get more expensive to run. Development will slow down and tools that depend on newer models will start to stagnate. Training slows or stops. Capabilities plateau. And if that happens, we will be left with no longer needed capacity.
What About Legal?
For AI tools to work effectively in legal, they not only need to be accurate (more on that in a later post) they need to work at warp speed.
- Lawyers and legal professionals can’t sit around for hours waiting for AI to supply a new contract draft.
- They can’t wait days for an AI tool to help them prepare for a hearing or finalize a motion on time.
- A litigation team facing a discovery deadline can’t tell the judge that Harvey was down.
- Clients won’t accept “the AI crashed” as an excuse for missed deadlines or shoddy work. How can firms meet client expectations when they’re relying on AI that may not deliver as promised?
- The alternative fee business model with which many firms are experimenting depends on AI doing a lot of the work. What happens if AI can’t do that work or at least do it in the expected time frame?
- And it’s no defense to a malpractice case that there was an outage.
Most of the legal world is not focusing on the risk the infrastructure volcano will erupt. They aren’t making contingency plans if and when AI becomes constrained due to energy shortages. They aren’t asking about what happens if the GPU supply becomes threatened. Or if the cloud provider runs out of capacity. They don’t wonder if uses can be scaled to meet peak and emergency legal demands which often happens in litigation.
Law firms are making million-dollar bets on AI infrastructure they don’t control, with vendors who can’t guarantee uptime during peak demand. Legal AI is like Pompeii before the volcano. Once it erupts, the cascading effects will be throughout the industry. Law firms could be left with expensive tools they can’t use. The legal AI boom will be just another tech bust.
A Few Prophets
Of course, not everyone in Pompeii is ignoring the potential pitfalls. Recently, even Google CEO Sundar Pichai sounded the alarm that no company would be left unscathed if the AI boom collapses and that there are elements of irrationality in the market, according to an Reuters article.
On the other hand, fortunately there are plenty of entrepreneurs out there looking for innovative solutions. Entrepreneurs like Lance Dieken, a tech savvy businessman with some 30 years in the IT industry and more advanced degrees than I can count. His company, XYK Financial, is looking to solve the speed and capacity problems by going smaller scale. Building smaller data centers and then powering these centers with generators that depend on existing and underutilized energy sources like combined heat and power (CHP) close to where the centers are.
XYK is looking to modular and quickly deployable systems instead of behemoth data centers. Think local. The smaller scale and decentralized approach enable faster build time and cost-effective geographic deployment. “The project,” says Dieken, “leverages existing geothermal sources and fiber infrastructure to minimize capital expense and maximize speed deployment.”
XYK’s systems could be used to supply the needs of hospitals, universities, and factories. And law firms: distributed systems like XYK’s could provide the backup capacity needed when big data centers hit peak demand during major work crunches. And these smaller centers could spur growth and jobs in rural locations, close to the data centers.
Dieken’s “keep it simple approach” allows him to move fast, not worry about over supply and meet the needs where they are. We will need this kind of thinking to meet the infrastructure problems.
So, What Do We Do Now?
Like the ancient city of Pompeii, legal sits complacent to the real fundamental risk towering above it with AI. In the age of the nonstop AI hype machine, legal risks an overreliance on something that has some fundamental issues, as this morning’s experience taught me. One can only imagine if the outage was longer or more pervasive due to systemic infrastructure problems or something else
What can legal do to survive? Be aware of the volcano that could and likely will erupt. Understand that there may be limits to how far AI can go despite what everyone is saying. Yes, we should all be cognizant of the immense benefits that AI can bring. But like good lawyers, let’s look at the potential problems and our collective exposure should the AI tools plateau or even crater. Ask the hard questions.
It’s probably safe to assume, as many say, AI is as bad today as it’s ever going to be. But take that with a grain of salt: it may not necessarily continue to exponentially improve or even work as well and as consistently as more is added. Let’s not lose our heads.
Smart firms should diversify their AI dependencies, maintain analog backup processes, and avoid betting their client service on infrastructure they can’t control.
And not believe everything the vendors promise.
Stephen Embry is a lawyer, speaker, blogger, and writer. He publishes TechLaw Crossroads, a blog devoted to the examination of the tension between technology, the law, and the practice of law.
Melissa Rogozinski is CEO of the RPC Round Table and RPC Strategies, LLC, a marketing and advertising firm in Miami, FL.