Potemkin AI.

Generative AI and HR Tech. My current perspective.

I was at the Unleash Conference this week. I was also at the France v SA rugby. It was great to catch up with old friends, limited partners, other VCs, portfolio companies, vendors and more. Over the years the event has grown, as has the industry. It is an excellent event, well played Marc and team. It was also a brilliant game of rugby, the memory of seeing the Bokke win will stay with me for ever. Well played Siya and team.

The conference was all generative AI. The din was loud as the chants of Allez les Bleus in the Stade de France.

I am excited and nervous about the potential of AI, generative and non-generative. I’ve spent chunks of the last year or two learning about different AI methods and their potential applications. I’ve talked with many AI experts and read widely. I am not an expert, and I’m just starting to understand the opportunities and current limitations. I am convinced that almost all HR tech vendors will need to apply AI to be competitive. But saying you need AI is like saying you need weather, it is accurate, but not very useful.

I don’t share the same levels of enthusiasm about most of announcements and demos as many others do.

In 1797 Catherine the Great went on a tour of newly annexed areas of her growing empire, including the Crimea. Potemkin, was the governor (and her lover). The story goes that:

“Potemkin erected phoney portable settlements along the banks of the Dnieper River in order to impress the Russian Empress and foreign guests. The structures would be disassembled after she passed, and re-assembled farther along her route to be seen again.”

Modern historians reckon that the fake villages themselves were fake news, spread by the British press at the time, but the metaphor of the Potemkin Village has stuck (I’m a bit of a history geek, if you are, check out the Empire Podcast).

It is remarkably easy to add a thin layer of Gen AI onto an existing product. It imbues hitherto rather dreary applications with a new sparkle. It creates an aura of boundless possibilities. It must be fun to be in pre-sales today, because this demos like a dream.

Part of the reason for my scepticism is the time line. Open AI released ChatGPT in Q4 2022, and enterprise versions around the middle of this year. For almost all incumbent vendors the roadmaps for 2023 were largely nailed down before GPT hit the market. Most incumbent vendors are not packed to the gills with AI engineers and product managers with deep AI backgrounds. It is really hard for incumbents to build anything meaningful that is production ready in a couple of months.

Generative AI is part of the AI story, but it is not all of it. Generative AI is good at somethings, and at others it is at best useless, and often dangerous. It can’t really do maths properly, for instance.

I keep coming back to the words of one of the pioneers of AI. If you want a challenge read Joseph Weizenbaum’s Computer Power and Human Reason. He built what was probably the first chatbot, Eliza. He published a paper about it in 1966. It crudely mimicked a psychotherapist, and he was shocked how people were taken in by it.

What I had not realized is that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.”
― Joseph Weizenbaum

I’m trying to figure out where the value creation and value capture will happen with AI in HRTECH. I’ve not worked it out yet. Is it with the AI infrastructure players, is it with the incumbent vendors, or is it with start ups? What are deep capabilities that are going to be hard to copy, that create enduring, scaleable businesses? Who will be disrupted? The process of software creation itself is changing. We will see innovations at multiple levels, front-end, back-end, in integrations and more.

My public service announcements

For end-users: Demos are not production software. Look beyond the magic wand and ask the tough questions about bias, hallucinations, audit, reliability, security, safety, support, privacy, data providence and performance. And ask what business problems is it solving.

For vendors: If you think that adding a thin layer of Generative AI on top of your existing product is going to win you heaps of new revenue, and give you new moat, you are in for a nasty shock. If it takes you a few weeks to do this, then anyone can, and will.

AI is going to mess with every layer of your tech stack (the front end, the back end, the middle bits and all your integrations). If you don’t have a plan, and you aren’t devoting a significant part of your product and engineering teams to AI related efforts, you are likely to find the next few years an uphill battle. Nifty demos are all well and good, but the real work is just beginning.

As usual, here is a tune for you. Lang Lang playing Debussy in Paris.

More about Debussy here.