Gen AI: Peak hype doesn’t mean you should sit back and wait

“This is happening. It will be done by you or to you: we are at an inflection point and we are also at peak hype, but those are not mutually exclusive.” 

I’m sat on a sofa at ILTACON in Orlando with Casey Flaherty, co-founder and chief strategy officer at LexFusion, who I’ve asked to help me cut through some of the noise around generative AI, which, as you can imagine if you weren’t there, dominated the conference in a big way.  

Flaherty, for those of you who don’t know him, is a former Holland & Knight associate who created competency-based technology training Procertas, before co-founding legal innovation marketplace LexFusion. One of the companies LexFusion has worked closely with is Casetext, which this year brought GPT-4-based AI assistant CoCounsel to market and was acquired by Thomson Reuters for an eye watering $650m.  

Flaherty and I have both spent the last few days having conversations with literally hundreds of people, many of whom have stressed that gen AI is over hyped and that things will calm down soon. In fact, as I held this interview, the Walt Disney Dolphin Hotel lobby was buzzing with people no doubt having precisely those sorts of conversations. 

They are right, of course. Just immediately before the conference, Gartner published its 2023 AI Hype Cycle and there gen AI is, right at the top of the curve, ready to slide down into the famous trough of disillusionment.  

With legal tech vendors clamouring to bring their own early gen AI products to market, some of which are little better than vapourware, it is no wonder that buyers are visibly overwhelmed. My question is, given that there is so much hype and froth, is that a good reason to take a wait-and-see approach, particularly in light of some of the risks involved? 

Inflection points and hype travel in pairs 

The days of having to persuade people that gen AI is a big deal already feel largely consigned to the past, but for anyone still in doubt, Flaherty says: “The advances are very real, and the excitement is not only genuine but well-founded – this time it is different and I believe that we’re in a new super cycle. It was mainframe, then PC, then mobile, then cloud, now this, and every super cycle builds on a previous one and they move faster.” 

Flaherty says that it can be true that we’re at an inflection point but also that we’re at the peak of a hype cycle, commenting: “Genuine inflection points and hype cycles tend to travel in pairs. You can have hype without inflection, but never the other way round. An easy example is the dot.com bubble. It turned out to be true that simply owning a good URL does not a business make: you have to have an underlying business that works. But the digital real estate land grab has borne out to be real; Pets.com at peak hype was worth $400m – the business wasn’t worth that, but the digital real estate was.” 

What we’re seeing right now is an initial wave of excitement, and funding, and first-generation products, and Flaherty says: “Because the road is long, people get disappointed about what the products can do, which is where you hit the trough. But by the second and third round of funding and products, we will hit the plateau of productivity.” 

Early products to market 

What is clear – even if unhelpful – is that not all gen AI products being launched in this first wave will be a success. Flaherty says: “You can find 20 articles forecasting that the iPhone will be a flop and they were wrong. But Microsoft’s Window Phone and Amazon’s Fire Phone were a flop. Just because you produce something doesn’t mean it will be successful.” 

Vendors that were already working with large language models – such as Casetext – have an advantage, although Flaherty says that they were also smart in pivoting their entire business in the run up to OpenAI’s launch of GPT and in creating a very specific set of tools rather than “trying to do everything.” 

CoCounsel launched in March, and Flaherty observes: “You’ll find that LexisNexis and Thomson Reuters didn’t come out with their announcements until May and there were filled with words like ‘beta’ and ‘preview’, which is appropriate. The question is not whether the technology is applicable, it’s how much are they investing and how long it will take them to get to commercial grade applications. If you take a view of history, it will be short, but compared with consumer expectation, it will seem long and disappointing.” 

For many companies without the existing infrastructure, the road to product may well be long, but Flaherty says: “That doesn’t mean what they are saying today is bullshit, it just shouldn’t be surprising that most of what is available isn’t that mature or impressive.”

Being ahead is better than behind 

Arguably none of this really helps buyers in their decision making, but Flaherty says: “It’s early days but with something that is this massive you want to be ahead of it, because all of a sudden, you will be way behind. 

“So much of what you’re going to be doing now is learning and experimenting and exploring and developing an understanding. You’re going to be spending more time than you have vetting new products or new versions of existing products, but it is time you need to spend, and you need to explore in a secure manner.” 

Here Flaherty give a mention to LLM-governance platform Lega, which enables law firms and enterprises to assess and potentially implement gen AI technologies. Flaherty says: “You really do need to be looking at the Casetexts and the Macros and the new stuff from Disco and Litera, because it’s not that everything everywhere has changed, but that change will start to happen faster than we have ever experienced before. It might feel like a waste of resources because most of this stuff is not that mature and game changing, but being ahead is way better than being behind.” 

Fear of failure, or worse 

One of the deterrents for buyers to engage with gen AI tools and products is a fear that they will get it wrong. That they will breach privacy rules. That they will screw up big time in ways that they possibly haven’t even conceived of yet. 

On this point, Flaherty is unwavering. “When Richard Susskind came out and said that email would be the dominant communication for lawyers he was labelled dangerous and insane. People said we can’t adopt email because it creates issues for data privacy and client privilege and that is 100% correct. You are also creating a record that is discoverable, which is not just a risk, it’s a downside and a cost.  

“All of those data security, privilege and labour employment issues are real issues that are going to be actual costs, but objecting to using large language models on that basis is like objecting to email – you’re not wrong in the narrow sense, but you’re definitely not right. 

“This is happening, and it will be done by you or to you. We are at an inflection point and we are also at peak hype, but those are not mutually exclusive. The road to productivity remains long, but it’s a lot shorter than it used to be.” 

caroline@legaltechnology.com