We humans are busy developing and already using Artificial Intelligence (AI) on a large scale.
That is sometimes useful, but also a problem. We still do not yet understand exactly how our own intelligence and our own consciousness work, but we are building forms of Artificial Intelligence that should not only work as well as our own intelligence, but even better!
That seems doomed to failure to me for now.
Of course, AI can be very useful! For the time being certainly for simple repetitive actions, calculations, designs and decisions. And now that the market is flooded with AI systems, developers will have to continue to innovate.
Almost all of the current AI systems are based on so-called ‘large language models’ (LLMs). This means that the developers copy huge amounts of text and other data, often illegally (a pressing ethical and legal issue!), and ‘feed’ it to their software.
With that, they have developed software that can not only literally repeat what we say, but can also predict what we are going to say before we have even said it! Sometimes it does so in a stunningly convincing way, but more often it works rather clumsily. This is not only because AI itself is not yet functioning optimally, but also because of the often-dubious data with which it is fed.
Only in recent years have developers discovered that simply feeding AI huge amounts of data is not enough – something that seems obvious, but that is apparently not evident for these often withdrawn and autistic young men – and they are now building in their first corrective mechanisms. They seem to have realized that there is more to intelligence than just copying, regrouping and predicting data (which was sufficient for a good chess computer), and that correcting and adjusting are also important aspects of human intelligence (gosh!).
They develop these corrective mechanisms with the help of cheap laborers in poor countries, who have to check the produced ‘intelligent’ texts, images and sounds for errors and for bad things such as bias, racism, sexism, etc. Hopefully it won’t stop there and they will also involve teams of specialists from scientific fields such as neurology, psychology and ethics. In short: the not-yet-so-intelligent results still need to be significantly refined.
And then there is the problem that Artificial Intelligence can lie very well and convincingly. After all, it is not interested in the truth or in standards and values, but in predicting what the most logical next words and sentences (or images or sounds) are that it has to produce at our request.
This too poses a major ethical and legal dilemma! AI-generated, persuasive disinformation has already had major undesirable effects on politics and on the way people interact with each other. And this is just the beginning.
Although Artificial Intelligence can produce beautiful sentences, the content often leaves a lot to be desired. And that makes sense. AI cannot imitate, let alone surpass, human intelligence by simply predicting what the next thing we humans would say or show will be, based on that enormous mess of (qualitatively dubious) data it is fed with. Not even with a built-in rudimentary correction mechanism.
That is not how intelligence works! It is a gross underestimation of the magnificent complexity of our consciousness, our thinking, our creativity, our emotional life, our memories, our experiences, our associations, our interpretations, our mutual interactions and connections, and so on. In short, of everything that together makes up our complex intelligence.
Maybe we should be a little less ambitious and first develop an artificial form of what we as humanity are much better at: our stupidity! Our unbridled nonsensical denseness!
Let’s first develop Artificial Stupidity (AS)!
After all, not much can go wrong with that (LOL), it has a much greater chance of success and we can have a good laugh about it too!
And then we know right away what NOT to do when we focus on Artificial Intelligence again…