Skip to main content

AI is biased - well, that's the whole point!

Artificial Intelligence is a misnomer.

Does anybody have an idea of what human intelligence (the non-artificial one) is? Over here in Germany Intelligenz is absolutely not the same thing as intelligence represented with a capital I in CIA, for example. That CIA-one was most probably the intelligence that was meant to be artificially emulated in the 50s, 60s, 70s... And quite successfully so.

Today, AI is widely perceived to be an imitation, a parallel realization, or even a substitute of the human intelligence that suposedly sets us apart from every other rock. Remember the times when your science-inclined parents explained to you how the behavior of snails is nothing more than the output of a hardwired machine? Same for the canary, the turtle under you bed, and the dog. (It was hard to believe how the malevolent fits of my cat were hardwired and not real, genuine evil. But it was.) Intelligence was nicely reserved for us humans. That appears to get questioned. The definition of intelligence was usually a list of things unintelligent conversation-equipment would not be able of doing. This nicely excluded everything that didn't talk. Fish were out. Insects, obviously! too. But on closer inspection it didn't really hold. And it becomes every more clear that a real definition of intelligence - the human one with a 'z' in german - is sorely lacking.

Let's briefly dip into the hotly debated problem of bias in Large Language Models - which blew us laypeople away when they popped out of nothing earlier last year. We all played with the most popular AI-Apps and they are truly impressive (some users lamented the lack of accuracy when answering content-related questions. But that is not what those systems are built for, anyway.).

There sure enough is a widely documented problem when using those tools to sift through CVs or when asking for a suggestion of what to look for in a perfect candidate for an engineering position. The results unsurprisingly perpetuate clichées and stereotypes as the data-base is a western-centered a pile from a male-dominated past (past we hope) and it is filled with the everyday patterns of prejudice.

It is important to remain aware of this, but it is not at all surprising let alone malevolent. The AI we amateurs tinker with is a system trained to find - and repeat - patterns. It is probably best compared to our 'intuition' or, yes, bias we as humans are endowed with. The pattern of driving a car is - prejudice, bias... somehow. We repeat patterns we were trained to recognize without using our reasoning brain, which is too slow and too close to consciousness. You realize that best when you apply this competence in the wrong environment. I emember the headaches after the first day of driving on streets in the UK. The trained patterns have to be overriden by reason, by inellect - maybe 'intelligence'. Or to call Daniel Kahnemann to help - the software we today know as AI (or large language models) emulates quite nicely the fast part of thinking, not the slow part. So while we are 'thinking fast and slow' we have to stay aware of the fact that the AI-apps of 2023 are pattern-reproducing devices, they are bias-repeaters, they are thinking fast but there is no reasoning.

Reasoning, the slow part of thinking, is left to us.

For now.


Popular posts from this blog

Academics should be blogging? No.

"blogging is quite simply, one of the most important things that an academic should be doing right now" The London School of Economics and Political Science states in one of their, yes, Blogs . It is wrong. The arguments just seem so right: "faster communication of scientific results", "rapid interaction with colleagues" "responsibility to give back results to the public". All nice, all cuddly and warm, all good. But wrong. It might be true for scientoid babble. But this is not how science works.  Scientists usually follow scientific methods to obtain results. They devise, for example, experiments to measure a quantity while keeping the boundary-conditions in a defined range. They do discuss their aims, problems, techniques, preliminary results with colleagues - they talk about deviations and errors, successes and failures. But they don't do that wikipedia-style by asking anybody for an opinion . Scientific discussion needs a set

Left Brain, Right Brain

At a wonderful summer night I was lying in the grass, my little son beside me. We were staring into the dark sky, debating infinity, other planets, the origin of everything, observing falling stars that were whizzing through the atmosphere at a delightfully high rate. Why did we see so many of them that night? What are falling stars? What are comets. Why do comets return and when? The air was clear and warm. No artificial lights anywhere. The moon was lingering lazy in the trees across the river. Some fireflies were having a good time, switching their glow on and off rather randomly - in one group they seemed to synchronize but then it was random again. It reappeared: a few bugs were flashing simultaneously at first ... it started to expand, it was getting more. A whole cloud of insects was flashing in tune. Are they doing this on purpose? Do they have a will to turn the light on and off? How do those fireflies communicate? And why? Do they communicate at all? My son pointed at a fie

My guinea pig wants beer!

Rather involuntary train rides (especially long ones, going to boring places for a boring event) are good for updates on some thoughts lingering in the lower levels of the brain-at-ease. My latest trip (from Berlin to Bonn) unearthed the never-ending squabble about the elusive 'free will'. Neuroscientists make headlines proving with alacrity the absence of free will by experimenting with brain-signals that precede the apparent willful act - by as much as seven seconds! Measuring brain-activity way before the human guinea pig actually presses a button with whatever hand or finger he desires, they predict with breathtaking reproducibility the choice to be made. So what? Is that the end of free will? I am afraid that those neuroscientists would accept only non-predictability as a definite sign of free will. But non-predictability results from two possible scenarios: a) a random event (without a cause) b) an event triggered by something outside of the system (but caused).