Category: Artificial Intelligence
Nov 25
Of Intelligence Spaces And Human Like Intelligence
If one believes the common narrative in the media, Hollywood films and today’s popular science literature, then artificial intelligence (AI) will lead us into the apocalypse. The Terminator and Superintelligence will soon destroy us or at least enslave us all. AI will be smarter than us, and with it we will have created our last invention.
But here is the paradox: on the one hand, we humans dare to be so smart that we can create Superintelligence, but at the same time we are so stupid that we cannot test it and stop it.
Well, Superintelligence will be something we are unlikely to experience in our lifetime. And it will probably look very different from what we imagine today. A false conclusion is that we always compare and equate such Artificial General Intelligence (AGI) with human intelligence. We ourselves have difficulties defining what intelligence is and what intelligence we humans actually have. Not to mention the intelligence of the animal world.
We have always defined intelligence differently over time. “An AI will be intelligent when it defeats the chess world champion.” Well, and as soon as she did it with Deep Blue against Gary Kasparov in 1997, we quickly redefined her, because this “Brute Force” computing we didn’t really see as intelligent. And the next level – an AI that can defeat a Go world champion – was then again only probabilistic algorithms, so not really what we want to understand as intelligence.
A common current definition describes intelligence as such:
Intelligence is the general mental ability that includes, among other things, the ability to think logically, plan, solve problems, think abstractly, understand complex ideas, grasp quickly and learn from experience. It does not consist of pure learning from books, subject-specific academic competences or the ability to pass tests. It is about more comprehensive and complex skills to understand the environment, make sense of it, and find out what to do.
If that’s too long for you, then you can use Max Tegmark’s (author of Life 3.0 and founder of the Asilomar Conference on Artificial Intelligence) definition, who describes intelligence as
the ability to solve complex tasks.
Even these definitions are vague, because we can already divide our own human intelligence into several subcategories, each of which can and does look different. If these could be expressed in algorithms, they would have little in common. Types of intelligence we are talking about include social, emotional, cultural, artistic or spiritual intelligence. In addition there are other forms like body intelligence, or the one that becomes swarm intelligence when we combine the brainpower of several people.
As we can see, we are talking about different intelligences that make up human intelligence. Each of these intelligences in turn occurs on a spectrum. Not all of us have high social intelligence, but perhaps strong emotional or spiritual intelligence. We humans are each unique in our characteristics in this multidimensional intelligence space.
Animals, on the other hand, have partially similar, and then completely different, manifestations of intelligence, which makes them adapted to their habitat. Some of these animal intelligence dimensions like our overlapping, but only partially.
We can already see: we are floating somewhere a much larger multidimensional intelligence space with our comparably small multidimensional intelligence, and the total size of the intelligence space is something we cannot estimate. Animals and other living beings probably are floating in another part of this entire intelligence space, partly overlapping with ours, but mostly not.
And now we are adding machines that will be equipped with their own set of dimensions of intelligence. Neither will it be human-like, nor should it be our goal. Just as we should not invest huge sums in the development of wood-like materials when real wood is so much cheaper and easier to create by simply growing, so it does not make sense to artificially recreate human-like intelligence. We have enough of that anyway and it is much easier and more fun to create – in the form of babies.
But there is a reason why we want to artificially recreate human-like intelligence: to better understand our own intelligence and our brain and body.
The bigger goal should rather be to create AI in these areas of the intelligence space, which we have not yet been able to cover purely through human intelligence. If with this AI, created by us humans, we are able to cover these areas of the intelligence space, which we cannot yet grasp and use, only then can we speak of an intelligence explosion. Just as over thousands of years the great effort of mankind was to make more and more energy accessible to us, so it is our time to make more and more intelligence accessible.
More available intelligence can solve more problems better and faster. Through natural intelligence we have probably reached a limit, or at least the limits of growth rates of intelligence, but with artificial intelligence we can maintain this intelligence growth. And today we need nothing more than smarter approaches and more brainpower.
This article was also published in German.
Recent Comments