At the time of writing (who knows, the whole grift-fest might have blown up by the time this article is published) huge investment plans are being announced, seemingly every day, for some datacentre or other to support the “AI boom”. (I put that in quotes because I don’t believe it, but that’s not really the point of this article.)
Apparently, the reason for those massive investment plans is that the big AI companies need “more compute” – that is, ever-increasing amounts of data, chips, and data centres – in order to make their products work.
If only they can buy enough “compute”, claim the AI companies, all their problems will be solved, and we’ll inevitably reach The Singularity in the next six months.
To use a word a good deal more polite than the one I normally use to describe the AI industry, this is hokum.
Now, to be fair, in the early days of the AI industry, it wasn’t. Or perhaps, more accurately, I didn’t know enough about it yet to form that view.
That’s because, despite a 30 year career as an accountant, I have remained relatively optimistic and upbeat about what humans can achieve. I was as excited as anyone else when those “magic black box” AI technologies came out of the woodwork in the early 2020s, because I like to try new things.
It’s always easy to say “no”, of course. But during my career, I’ve generally found that, while this sounds like the safe option, it’s usually anything but. Saying “no” to everything guarantees that you stay pretty much where you are while the rest of your industry passes you by. You’re on the road to irrelevance if you carry that stance to extremes.
Now, if your business is selling antique-style scarves, say, which are hand-dyed and hand-stitched using the original methods and machinery from the 1870s, you probably should say no to every new idea about how you can make your production more efficient – in that sort of business the tradition – the inefficiency, if you like – is the main reason people can sell scarves which functionally serve the same purpose for £2,000 on a boulevard in Paris and 99p in a branch of Primark.
That extra £1,999.01 is, in part, to compensate you for the relative inefficiency of your traditional methods because if buyers didn’t do that, their only option would be a 99p scarf from Primark.
But if you’re reading this, it’s unlikely you’re selling £2,000 scarves from a Parisian boutique, so we can put that option aside.
Saying yes…with conditions
So, to anyone who isn’t in the Parisian scarf-selling sector, saying “no” all the time is rarely a good idea, even if you are an accountant.
Because if your business isn’t innovating, it’s highly likely one of your competitors is. And that means, at some point in the not-too-distant future, your business will be left behind. Ultimately you might not have a business at all.
For that reason, in the early days of a new idea, you should ideally by trying to say yes, if you can, to get it to the point where you’ve got enough of a proof of concept to make a final decision with a much better chance of making the right call.
Now, that doesn’t mean every manifestly stupid idea should get the green light. But it is to say, if the idea isn’t completely barking, you should probably try to let it run for a bit if you can.
In general, a little bit of early stage proof of concept work is relatively inexpensive and, you never know, it might turn out to be the most brilliant idea your sector has ever seen. Taking a chance here and there is unlikely to materially hurt your bottom line, and it might just be the foundation for an exciting new chapter for your business.
There is another important reason to take this approach where you can.
Assuming you employ competent, hard-working people, the idea probably isn’t completely mad. By green-lighting a little bit of exploration, you’re showing them respect and appreciation, which is rarely a bad thing.
But if you kill the idea on the spot, your team will start dreaming about their idea when they should be thinking about your business and your clients, and there will always be an “if only we green-lit that project in 1997, imagine what sort of a business we could be today” vibe about the projects which were summarily canned.
Your best option is to do a bit of exploration (with not a lot of money, and not a lot of time, until you know there’s something worth taking forward).
However your green light should come with some conditions. That way, you can either kill the idea if the conditions aren’t met, or give yourself the confidence that there could be something in the idea which you want to explore further.
Impact
One of those conditions should depend on the potential impact of the idea on your business.
It’s a matter of personal taste, but if someone proposes an idea which might not transform the world, but which is fine as far as it goes and doesn’t cost much to implement, my default setting is just to tell people to get on with it. There’s no need for formality, reporting, and meeting conditions if you’re moving the stationery cupboard from one side of the Reception desk to the other – just get on with it.
For more serious projects, though, business impact is critical, otherwise you risk key members of your team frittering away their time and energy on projects which are inconsequential in the context of your business as a whole.
Here I’m probably looking for a minimum 5% impact on the top line or a 2-3% impact on the bottom line, for a first cut.
Anything less than that is unlikely to generate an RoI on the time, effort and money which will need to go into the project.
So the first port of call is a “back of an envelope” calculation of the potential impact.
Again, if it doesn’t have that 5% impact, but is inexpensive, you might as well do it anyway. It won’t cause any harm and it means the person whose idea you green-lit is more likely to come forward with another idea in the future – and that could be the one which turns out to be transformational.
But if you’ve got a 5%-er, the next stage is to do a proof of concept.
As a general rule, if something doesn’t work small-scale, it won’t work large-scale.
Large-scale is where you drive your efficiencies and make a return on your investment. But there’s nothing magic about scale – if something doesn’t work when it’s small scale, it won’t work when it’s large scale either.
The step ladder
Going from an idea to a full-launched product is like a step ladder. You have to take the next step before you can take the one after that, for example.
And if you think about it, there are multiple steps to any project which is going to have a significant impact on your business as you move from the “prove it in the lab” stage right through to having a finished product on supermarket shelves.
For example, in the early days, you might use only the staff members working on the project to direct how the project goes. At the later stages, you will almost certainly want to involve a survey group, or a taste panel, made up of ordinary members of the public to double check that the idea your team had stands a chance of making it in the real world.
Whatever those stages might be for your business, once you’ve green-lit an idea, there needs to be an understanding that the project is only green-lit for each stage in turn. Giving the OK up front doesn’t mean one of your team has the authority to build a new factory in Outer Mongolia – it might just be OK’ing a train ticket to visit the Mongolian embassy in London to start researching potential sites.
If you take this approach, however, you will find that some projects – no matter how promising they seem up front – gradually run out of steam. Generally that’s because a good idea in a lab is not necessarily a good idea, or at least a do-able idea, in the real world.
When a project starts to run out of steam, that means the momentum has turned downwards and, usually, the law of diminishing returns has started to kick in.
If that happens at Stage 2 of a six-stage process, I can guarantee you that this project will never be successful in the real world. The worst thing you can do is persist with an idea that clearly isn’t delivering results. The kindest thing to do for everyone involved is to congratulate them warmly on the good idea they had, reflect that the real-world data isn’t coming in as people had initially hoped, and wind the project up completely.
Neither you nor your staff should be upset about an idea not working.
Just having an idea good enough to be green-lit in the first place means you’ve got someone there with some real potential for the future. You should be nurturing them, not demoting them, when an idea is canned at Stage 2 of 6 – probably, they’re the people who will come up with an even better idea next time.
Not more, but different
However, what I usually find is that an idea which is stuck at Stage 2 won’t get any better if you throw more resources of the same type that were deployed up to that stage.
Ultimately, there’s an economies of scale question to be resolved, of course – will an idea which works in a lab also work on a fully-automated assembly line in a vehicle manufacturing plant, for example.
But the way you get round a problem about things not working in a lab is rarely “more lab”.
Almost always there was some fundamental misunderstanding about the process, some conceptual blockage, some principle which doesn’t work as well in practice as it does in theory at the heart of the problem.
“More lab” doesn’t address any of those issues. It just piles up costs on a project you are eventually going to have to can anyway. That’s just not smart.
Think of it like being a championship dancer.
The difference between being an OK dancer and a champion dancer isn’t knowing the steps any better. You needed to know them already to get up to “OK” level.
Going from “OK” to “champion” means working on your musicality, the grace and elegance with which you use your body to tell a story, the way you draw influences from the greats of dance history to delight other championship dancers even though most of the general public won’t even spot the little bits of homage you’re paying to one of the greats.
None of those things are “doing steps harder”. They are entirely different activities.
And that, in a nutshell, is the issue I have with AI, and the industry’s claims that, if only they had more computing power they could solve all their platforms’ problems.
Ever since I experimented with AI early doors, I’ve found it underwhelming (issues of intellectual property theft and morality aside).
And as investment has poured into the sector, the results have not become any more whelming (is that a word?) than they were before.
Going from Ver 3 to Ver 4 has become a yawn-fest – and, if anything, performance is getting worse from one model to the next, which is hardly a positive sign.
More importantly, if “more compute” was the answer to AI, we’d have solved all of AI’s conceptual problems well before now, after spending the first few hundred billion dollars, and we’d just be in scaling-up mode, which I accept would take significant capital investment.
But normally, you’d only put in that investment to scale up a project which was already working in principle.
You don’t pour investment into an idea that’s run out of steam in the hope that scaling up will fix a conceptually broken model.
That’s never going to work. The only people who like it when that happens are the bankruptcy lawyers because they are the people who stand to gain the most, ultimately, from a multi-billion dollar investment into technology which fundamentally doesn’t work.
Throwing ever-increasing amounts of the same resources as you’ve deployed up to now in the hope that this will fix any product development problem is at least unwise… and possibly insane. You need to do something completely different to reach excellence.
Perhaps start by thinking less like a tech exec and more like a dancer. At least dancers understand that excellence is unlikely to be reached by doing more of the same, and more likely to be reached by integrating ideas you haven’t work on up till now into the equation.
Alternatively, you could just decide it’ll never work and can the project before it bankrupts your business.
But, whatever you do, don’t keep doing what you’re doing.
As Albert Einstein said, the definition of insanity is doing the same thing over and over again, while expecting different results.
I wonder if tech people have ever heard of Albert Einstein? All the available evidence suggests they haven’t.