How I Learned to Stop Worrying and Love the Chaos: By Erica Andersen

Finextra

4 hours ago

Or: AI Confessions from the Keynote Stage
What a difference a year makes. Last week, I found myself on stage at the AI World Congress, delivering a keynote to a room full of people who, twelve months ago, were probably telling anyone who'd listen that AI was going to solve world hunger, cure cancer, and maybe even fix their corporate expense reporting system.
Fast forward to today, and suddenly the same crowd is singing a very different tune. The other keynotes? Let's just say they weren't exactly radiating optimism. Microsoft, Oracle, IBM, McKinsey – the usual suspects – all took their turns at the podium to essentially deliver variations of the same message: "AI is hard. Our systems don't work. Where's our ROI? We're confused and slightly terrified."
Welcome to reality, folks. Population: everyone who actually tried to implement AI.
The Crybaby Chronicles
Now, I don't want to sound unsympathetic. Actually, scratch that – I do want to sound a little unsympathetic, because here's the thing: we've been saying this for years. AI isn't just software with a fancy hat. It's a completely different beast that doesn't play by the rules you learned in your Computer Science 101 class.
These organizations have been approaching AI with a software-only mentality, and then acting shocked – shocked! – when things don't work like a traditional database query. AI systems can fail silently, which is terrifying if you're used to error messages that actually tell you what went wrong. They can also appear to work perfectly while delivering completely suboptimal results, which is like having a GPS that confidently directs you to drive into a lake.
You need an engineering mindset for this, not just a software background. Engineers understand that things break, that systems are unpredictable, and that you need multiple layers of protection. Software developers expect deterministic outcomes. AI gives you probabilistic chaos with a side of randomness.
The Economics of Artificial Anxiety
And then there's the money talk. Suddenly, everyone's discovered that running AI costs actual money. Who could have predicted this shocking development?
Here's the part that's going to make you really popular at parties: I think the big providers – AWS, OpenAI, the whole gang – are actually undercharging right now. They're burning through investor cash to grab market share. At some point, someone's going to want to actually make money, and those token costs are going to climb faster than a venture capitalist chasing the next unicorn.
But here's where it gets interesting. People are obsessing over ROI, but that's like asking what the ROI was on the first spreadsheet. Imagine trying to explain to someone in 1979 why they should pay for VisiCalc: "Well, it's like a calculator, but bigger, and it has boxes, and you can change one number and other numbers change too." Revolutionary? Absolutely. Easy to calculate ROI? Not so much.
The smartR Approach: Embrace the Chaos
When we work with our AI models in our company we've taken a different approach. We think of AI as Assistive Intelligence, not Artificial Intelligence. The difference isn't just semantic – it's philosophical. Instead of trying to replace humans entirely (which is where most people run into trouble), we augment what people can do.
Think of it like having a really powerful, occasionally unpredictable intern. They can do things that are hard or impossible for humans, but you still want someone experienced reviewing their work. The magic happens when you combine AI's raw computational power with human judgment and oversight. You get something better than the sum of its parts, and you avoid the nightmare scenario of full automation gone wrong.
The Great Data Myth
Here's another sacred cow we love to slaughter: the obsession with perfect data. Everyone keeps saying, "Your data needs to be in order first." Well, guess what? Your data is never going to be in order. It's a beautiful, chaotic mess, and it always will be.
But here's the plot twist: AI can actually help clean up your data. Instead of spending months (or years) trying to organize everything perfectly, you can curate good datasets from your underlying messy data. The AI helps with the cleanup process. It's like having a really good research assistant who can find the good stuff buried in your filing cabinet of chaos.
People telling you that you must clean all your data first are essentially creating expensive busy work. They're making money off your preparation anxiety while you could be getting actual results.
The VC Reality Check
We also love talking about how the AI engine companies probably aren't going to make the ridiculous money that had VCs practically hyperventilating with excitement. This technology is going to become commoditized and open source. The real money – the sustainable, long-term money – is going to be made by people who figure out how to actually apply AI to solve real business problems.
That doesn't mean these foundational tools aren't important. They're absolutely crucial. But making venture capitalist levels of money from them? That's going to be tough when you're competing against open source alternatives and every tech giant on the planet.
Privacy: The Chickens Come Home to Roost
And speaking of uncomfortable truths, let's talk about privacy. We've been banging this drum for years, pushing private and secure models while everyone else was happily shipping their data off to the big cloud providers.
Well, surprise! A US court just told OpenAI they can't delete anything – including conversations people specifically asked to be deleted. Your private messages might become public evidence. But surely GDPR will save us, right?
Wrong. America doesn't care about your data protection laws. The worst that will happen is some European official will impose a token fine, give a stern speech about showing those big bad tech companies who's boss, and then everything will continue exactly as before – except now your private information is scattered across the web like digital confetti.
If you want your data to stay private, don't send it outside your ecosystem. It's that simple.
The Swiss Cheese Philosophy
Here's the thing about AI mistakes: they're inevitable. Sometimes what looks like a mistake to one person is actually a reasonable answer to someone else. That's just the nature of the beast.
I like to think of AI implementation using the Swiss cheese model. Imagine multiple layers of cheese stacked on top of each other, each representing a different safety barrier. The holes represent vulnerabilities. No single layer is perfect, but together they provide protection.
AI should be another layer or two of cheese in your stack. It adds protection and capability, but it shouldn't be the only layer. If you're going to remove human oversight, you better be absolutely confident that the AI can't break your entire system.
The Bottom Line
We're constantly upgrading our proprietary platform, based on real-world implementation experience. We've learned that AI works best when you stop trying to make it behave like traditional software and start treating it like the powerful, unpredictable tool it actually is.
The companies crying about AI being hard aren't wrong – it is hard. But it's hard in an interesting way, like switching from building model airplanes to building rockets. Sure, they both fly, but rockets require understanding thrust vectors, fuel chemistry, and the uncomfortable reality that sometimes they explode spectacularly on the launch pad. The physics are different, the margin for error is smaller, but when you get it right, you're not just flying – you're reaching orbit..
The key is approaching it with the right expectations, the right safeguards, and maybe a sense of humor about the whole thing. Because if you can't laugh at the absurdity of trying to teach machines to think while we're still figuring out how human thinking works, you're probably taking this whole AI revolution thing a bit too seriously.
And trust me, after listening to those other keynotes, we could all use a little less seriousness and a lot more practical wisdom about what AI can actually do – and what it definitely can't.
Written by Oliver King-Smith, founder and CEO, smartR AI