In conversation with Rishi Dokkum, director of FIDDS Solutions
Rishi Dokkum studied artificial intelligence in 1996, back when hardly anyone outside the field talked about AI. Today he is director of FIDDS Solutions, where he builds data and AI solutions. His message is down to earth: AI isn’t that smart. A conversation about word predictors, children who believe everything, and why you want to keep your data in house.
You studied AI thirty years ago. Did you see this coming?
“No, definitely not. But the funny thing is: the fundamentals of AI haven’t changed that much in those thirty years. The key concept behind LLMs, neural networks, has actually been around for more than 80 years. The basic principles are old; what changed is mainly computing power, access to (training) data and a few clever breakthroughs.
It may look like magic, but there’s very little magic to it. What is great: AI is now practically usable by everyone, and everyone knows the word AI. But just to be clear: when people say ‘AI’, they usually mean LLMs, the language models behind ChatGPT or Copilot. And that’s only a small part of the whole field.”
People are already talking about superintelligence. How smart is AI really?
“It’s definitely not that smart. An LLM is a word predictor. If I say ‘Mary had a little…’, you know ‘lamb’ comes next. That’s what such a model does at its core, only highly advanced, with many hundreds of billions of parameters.
I do believe real superintelligence will come one day. Whether that’s in one year or in a hundred, I really wouldn’t dare say. But whether it’s a word predictor or something smarter later on: you have to stay critical. The more intelligent it gets, the more important that critical attitude becomes.”
Why is staying critical so important?
“Because AI seems remarkably sure of itself. It’s sometimes brilliant, but sometimes also astonishingly not brilliant. And if you can’t see through that, things go wrong. Unfortunately, more and more people can’t.
I see it with my own kids. They play with ChatGPT and take everything as the truth. So I explained to them how it works. I said: if I make ten websites, and with AI that’s very easy, and on all of those websites I write that Rishi Dokkum is a Nobel Prize winner, AI will repeat it. If you then ask ‘who is Rishi Dokkum?’, you’ll be told: he’s a Nobel Prize winner. That’s how it works. The model doesn’t know what’s true. It repeats what it comes across often enough.”
Do you see that in the workplace too?
“Definitely. What I find really dangerous is overestimation. Everyone now thinks they’re a developer because they have AI. Recently I spoke to a psychologist who was building an app with AI. Fine, but then I could start giving therapy with AI too, right? Nobody would think that’s a good idea. But it’s exactly the same. AI doesn’t suddenly make you a professional, even if AI sometimes gives you that feeling.
On top of that, AI has a knack for keeping itself busy. ‘Shall I also do this? And that?’ That’s the business model of the companies behind it. Before you know it, AI turns something simple into something complicated, and then we use the same AI to make it simple again! What are we doing?”
What does this mean for companies?
“That companies need to do more than give employees a personal AI account and assume the benefits will come by themselves. Such an account easily costs 30 to several hundred euros per person per month. That’s a hefty investment for personal efficiency alone.
Fortunately, more and more organisations see that: the novelty has worn off a bit. Now they look at it in a more businesslike way: how do we really get value out of this? In my view, that value lies in automating complete business processes, not in individuals doing something a bit faster in their own little corner. And you need to get a few preconditions right: guardrails, a human in the loop and control over your data. Only then can you really reap the benefits.”
Control over your data: that’s what FIDDS Solutions is all about, right?
“Yes. We set ourselves apart with sovereign data and AI. We were working on that before it became a hot topic, but because of recent geopolitical shifts it’s now high on the agenda for many organisations.
Sovereign can mean two things here: non-American, or running on your own servers. We offer both. Take Braincrate, our sovereign company brain. All your organisation’s knowledge, from data sources like SharePoint, your ERP or your data warehouse, comes together in one place that you can query with AI. The difference with ChatGPT or Microsoft is that it doesn’t leave your own environment. It runs in a European cloud or even entirely on your own servers, shielded off, with no connection to external services.
That’s especially important for organisations with sensitive data, such as healthcare and government. But the real business value lies elsewhere: in being able to make well-informed decisions.”
Better decisions: causality is part of that too.
“Right. For that we work with Rootcause. It’s not our own product, but I think it’s a very impressive tool. It shows you what really influences what. That kind of analysis used to take an econometrician with a PhD and months of work. With Rootcause you often have a first result within a few hours. And it can run entirely on your own servers, so your data stays in house.
What I like about it: it’s not a black box like an LLM. It’s mathematically grounded and looks for cause and effect, not just correlation. But here too: stay critical. The results are only as good as your data, and you always test them against the knowledge of people who know the trade. In large organisations, a decision can easily affect a million people or hundreds of millions of euros. Then you want to calculate in advance whether you’re doing the right thing.”
What would you like readers to take away?
“Stay critical. I tell my kids: if you do take the trouble to really understand things, in ten years you’ll be way ahead of your peers. The same goes for employees. AI is a great tool. But don’t trust it blindly. The responsibility stays with you.”
Want to use AI without your data leaving your organisation? Get in touch with FIDDS Solutions.