Ai Hallucinations - Is AI the 2026 Musket?
The 16th century musket had a misfire rate of 40%. We still went to war with it. Sound familiar? AI is our 2026 musket !
With new technology, comes enthusiasm. Early adopters are risk takers and those who manage will indeed take a real advantage on the battlefield, business or war alike.
But here is the “uncomfortable truth” – 😡 Oh no! Not again the “uncomfortable truth”! This is written by a human, for god’s sake! Strike that! –
When a new paradigm arrives in the world, especially when it is challenging a lot of what we know, we tend to try it for everything and more importantly we try it despite its known limitations. AI is by design a Stochastic system. It does bring random in its answers. We know that. We know that from day one. Did we care? Nope. Did we use it as if it were reliable deterministic solution? Yes. Did it blow up to our face? Yes. Just like the Musket. And just like the musket, we keep trying because, when it does work, it is a killer tool. The famous “unfair advantage” that LLMs keep putting in everybody’s lazy post. If AI works for your problem then yes, it is a game changer!
The topic of the recent time is all about Hallucination control. After all, like this 15th century weapon, we don’t appreciate much when it blows up at our face. So, we keep working on it. The difference with the old gun: it is Stochastic! You can take the problem any way you want, the problem is integrated in the tool. It is not a flaw, it is a design.
So:
The Design: AI is a stochastic system. It is probabilistic by nature.
The Error: We treat it as a deterministic, reliable IT solution.
The Result: It blows up in our face. Just like the musket.
When an AI “expert” is telling you that they have solved the hallucination problem and made your AI totally reliable, especially using AI to fix AI, they either mislead you or the problem is suitable for a Symbolic AI solution. In which case, we are talking about a proper IT architecture, not an “Agent” doing Mission Impossible on its own.
If you want to go beyond the AI-Musket syndroma, you’d better take into account what AI really is and start building serious architectures around it. Let’s call that orchestration to make it sound fun, but we all know what we are talking about, right?
Did AI ever “misfire” on you? How did you fix the barrel?
We build AI solutions for Businesses. Solutions as in …”Software Systems using AI for what it does best so that it does not blow up to your face” kind of thing.
Talk to you soon.
