The middle ground isn’t sexy. It doesn’t dazzle with cheerleader-style enthusiasm, nor is it arch and cynical, like the cool kids. But when enthusiasm turns into hype, and the cynics reveal themselves to be little more than reactionary doom-mongers, clutching their grandmothers’ pearls, taking a measured, centrist view starts looking decidedly attractive.
And it’s where we find ourselves in the generative AI storm.
High pressure
Because, as the figures are showing, hundreds of millions have got swept up in the Chat GPT wave, only to fall back, disappointed and a bit cynical.
Like the fads, bubbles and assorted other grifts that came before it, AI has the perfect credentials to create that storm.
First, crypto died back enough to reveal a load of now-wealthy grifters, hungry for the next big thing and with just enough understanding of large language models (LLMs) to be drawn to the sector.
Then, the heady combination of easy-to-grasp potential (who doesn’t want to talk to a computer?) and Silicon Valley doing what Silicon Valley does, blew the whole thing high enough for everyone to give it a whirl.
And in a world where eyeballs mean dollar signs, millions of users all flocking to see what AI could do for their business made lots of people get very excited.
But when the use cases weren’t immediately obvious, and people didn’t quite know how to make AI stick, the too-cool-for-school kids came out from behind the bike sheds and helped to give it a kicking. Armed with the weary cynicism born of everything from the metaverse to self-driving cars, they booted it for lack of real purpose, with an extra dig in the ribs for its environmental impact.
Why today’s AI isn’t the product
Because AI’s generic appeal is also its problem. Despite Chat GPT’s easy-to-use interface, it’s only masquerading as the product.
So, investing in it now could be disastrous. If you don’t really understand the data that makes your business unique and haven’t spent the time getting it into a structure that can be well-understood by this type of technology, you’re pouring cash into a big output of nothing. And you’re probably launching a substandard product that actively irritates your users.
Because although AI is sometimes touted as a kind of alchemical fairy dust, the reality is more graft than grift. There’s a huge amount of work to get from owning some unique data, to people sitting down at one of these interfaces to have a good chat with it.
You need to understand what great answers look like. You need to think about structure, context, different kinds of search and everything that goes into building the prompt that’s good enough to generate that great answer. You need to design the interface that will support your user to ask the right questions. Lifting and shifting someone else’s format is unlikely to cut the mustard for your unique data.
And what do those users need to see in order to trust the answer? For example, the writhing mass of information that swirls around the legal profession seems like a good candidate for AI. But while someone might blindly regurgitate Chat GPT for their kid’s year nine essay, lawyers aren’t going to risk their professional indemnity quite so easily. Answers will need to cite the relevant references, together with links, so that professionals can check back for themselves on the validity of what’s been suggested. The output here isn’t a grab and go – it’s a prompt to help users to go off and do their own research.
Is AI for you?
Of course, for all the hundreds of millions of people who’ve toyed with Chat GPT and fallen away, some organisations with high quality, large data sets are already feeding it the fuel it needs to create great answers.
More study of the people using it every day is vital. What are they getting out of it? What roles do they work in? What need are they using it for? How are they using it? What value does an LLM actually add for them?
That’s how you’ll find out the potential value for your own organisation. If your data is numeric, generative AI isn’t appropriate. If your text-based information is unstructured or repetitive, your users will struggle to get a decent answer to a nuanced question.
But for our clients who have well-structured, high quality data, generative AI is revealing exciting new horizons. We’re working with them, through that process of breaking it down and building it back up until the answers are right. Our quality assurance system is testing and retesting, benchmarking the results against professionally-written questions and answers. Because confidence in the answer is crucial.
If you just use Chat GPT to pull together some information that isn’t life, death or professional suicide, enjoy.
But in all honesty, from our position in that unsexy middle ground, we’re advising all but a very specific few of our clients to watch with interest. AI isn’t simply big, shiny and magical. And nor is it carrying us all to hell in a handcart.
For the majority of organisations, the smart move is to hold back on major investment for a while until generative AI becomes more efficient and commoditised. The picture then should be clearer and calmer, with the developments that will enable more businesses to use it more effectively.
