Are you a big AI business vendor making terrible AI business decisions? We can help

The word Microsoft does not appear in this article. Why would you think otherwise?

Opinion Congratulations! As CEO of a giant tech company, head of a sovereign wealth fund, or a VC bored with megayacht leapfrog, you have billions of dollars of other people's cash to spend. You want to make a difference. You want to be a success.

Most of all, you want to look good doing it. Right now, nothing looks better than AI to do all three.

This columnist understands your needs. Nothing says forward-thinking, innovation-minded builder of the future than a big splash in artificial intelligence. It’s an exciting field in exciting times, promising big cash returns and reputational enhancement for big investors. I would be failing in my duty, though, were we not to offer some guidance in making a good deal. That means understanding the pitfalls.

Short of spaffing half a billion dollars into the void to discover you’ve bought into a bankrupt crock of ordure, how do you know what to avoid? You may not care very much if you see your investment as primarily marketing and the careful evaporation of money into the pockets of advisors, partners and consultants until years later it along with all the workers are all gone. In that case, you know your business better than we do. Enjoy.

If, however, you are steering a giant tech firm which has gone all-in on AI, investing in an AI failure could look bad, very bad, or truly awful. Is it that you don’t understand how to assess an AI proposition, which would be bad, or is that AI can’t live up to the hype? What would be truly awful would be if the AI investment was in technology in the bullseye of your speciality - code generation, say - and that just a couple of years ago you’d decided to align your core missions and integrate the tech with your flagship cloud services and AI platforms. Seriously, how bad would that look? Yeesh.

So, if you don’t want to align your core mission with bankruptcy, here’s what to do. First, do a proper analysis of the technology. Is it real, or is it a branding layer atop offshored humans? This would be a bad sign. If it does look like it’s trying to do what it claims, is it working? Are you sure what "working" even means here? Yes, this is hard with AI, especially when there’s no such thing and no-code AI app generation that always works. There’s no such thing as any approach to code generation that always works, so what are you buying - and when you’ve bought it, what are you selling?

As with any large tech investment, a wise organization will insist on a proof of concept pilot before proceeding to plunder the petty cash fund. POCs by themselves can be pointless, even misleading, if they don’t have scope and goals forensically designed and results rigorously verified. As the more astute observer will already have noted, forensic rigor isn’t a primary attribute of business AI, which explains a very great deal.

You can go right back to the Turing test as an example, a loosely deformed thought experiment that has forever been forced into uncomfortable configurations in an attempt to prove something. Where AI is most successful, again as widely noted, is where it is tightly designed to do a specific task with quantifiable results. Code generating AI certainly generates code, but that’s not the point. Saying, say, that 30 percent of your code base is AI generated means nothing without context: pre-NT Windows was largely written in assembler, and it was a very bad idea forcing a complete redesign.

Code generating AI isn’t a standalone black box app generator, it is a team member that needs expert human collaborators. If it speeds their work and ups quality, then fine. If not, it’s a problem, not a solution. If you’re coding in an organization where it is politically impossible to report that your artificial helpmate is anything but, then both you and your upper management have got highly real problems.

The only useful POCs for AI coding will be where the time to produce an app and the quality therein, compared to the total cost of whatever mix of carbon and silicon, actually did the job. That means a standard app that can be built in a mix of ways, in a public arena where nothing can be hidden, and competition is all. Cyber security teams have their Capture The Flag, autonomous vehicles were born in the DARPA Grand Challenge: AI code generation needs its own Robot Wars.

That’s the final tip for those looking to make a difference: be successful and look good in AI. Before you drop a bundle on a marketing-led hype machine with the intrinsic health and lifetime of a chocolate cake at a kindergarten, spend a fraction of that on a big public competition where app teams can face off with the best AI coding help they can muster. That’ll prove the POC everyone, especially you, needs to validate AI-assisted coding, and to make a framework for future development that won’t end up a costly, embarrassing mess. That is, of course, if the outcome of such an idea wouldn’t make you look even worse.

You know your job better than we do. ®