HiTech with Jan Coomans: Artificial Intelligence and what it is(n‘t)

Even when we’re talking about the most advanced AI models, the definition of the word intelligence is being stretched significantly, a bit like Tesla calling its cruise control system an “autopilot” or “fully self driving”. My car without cruise control is also self driving, just not for very long.  An adaptive cruise control like Tesla’s simply delays the crash, so to speak. If it were an actual driver then it would be able to drive a twisty dirt road in a snowstorm or navigate an intersection in Bangkok where the traffic lights are out. It can’t. Likewise, AI seems like it’s intelligent but only within very tight boundaries where no actual thinking is required. Like the robot at the Tesla factory doing most of the work, AI can give us the next level in automation but it can’t be genuinely creative any more than that the robot can itself decide to build something other than a car.

Which is still a technical achievement we can pat ourselves on the back for as a species, don’t get me wrong. I find myself pretty eager to delegate something that requires no actual thinking to an AI model. At least in those cases where the checking and correcting of the output doesn’t take more time than doing it all myself in the first place. There are plenty of tasks where these programs and models can be used, it’s just unfortunate that they are often used in ways that are counterproductive and frustrating. Take customer support, for example. Customer support is the one thing companies hate spending money on, so no wonder that they jumped at the opportunity to replace a human who had almost no way of helping the customer with an AI chatbot which has absolutely no way of actually helping you. It’s all part of the decline in quality of pretty much all that is online – which is pretty much everything.

But where Artificial Intelligence drops the facade and reveals itself to be closer to Real Stupidity is when you fact-checking it. There are countless examples to be found of cases where it proceeds to be utterly, but confidently, wrong. A child knows that there is no such thing as a food recipe that contains gasoline, or rocks, but the AI doesn’t. The fact that generative AI can be plain wrong on the simplest little things means that you also can’t trust it to tell you something on more complicated topics where you don’t know the actual answer. AI is good at generating an answer that looks like a human who knows what he’s talking about. It won’t use “hmm” or “maybe” or “I think that…” so as an end user you get the impression that it should be right because it expresses itself without a hint of doubt. Instead, what the AI generates is based on what it ingested during its training and that includes nonsense, parody and sarcasm. It doesn’t understand those things because its “intelligence” is only skin deep at the moment. Calculating the most likely answer based on whatever human-made content it was fed with no real way of differentiating a straight answer from sarcasm.

Of course, AI isn’t going away and it will continue to be a hot topic for a while. The companies who build the models will continue to make them more accurate and more useful. But as with any hyped-up technology its capabilities are overstated and the inherent limitations are ignored. When we as consumers inevitably get tired of being bombarded with this particularly viral acronym, presumably we’ll go back to just giving useful software regular names again and the term AI will drop in the internet search rankings as has already happened to otherrecent technobabble like “blockchain” or “web3”.

That’s not to say that there’s nothing to worry about, artificially generated content by any name can be disinformation rocket fuel, especially when let loose on social media where the algorithms make sure no to upset their users’ confirmation bias. Whether it’s all that or a simple investment in AI companies on the stock market, I strongly suspect nothing little good will come of it in the end. Or perhaps my natural brain has just been “trained” to be skeptical by existing for over 44 years on planet earth which is filled to the brim with unreliable data.  Just to maximize the irony though, I’ll translate this English-written text into Russian via ChatGPT before sending it off to the editors. If nothing else, at least it’ll annoy them. That, it seems, is currently AI’s most reliable feature.