A Realistic Path to Artificial General Intelligence - Part II
Published: January 19, 2020
A special kind of AI is progressing in a way that reminds us how human intelligence evolved. In this post I propose that we leverage it as the most promising path to Artificial General Intelligence.
But first thing first: wait, this is Part II of the article, if you missed it, Part I is here.
They will always be just machines
Believing that an Artificial General Intelligence could never be this or that or who knows what, if compared to us humans, is like believing that no artificial vehicle could ever run faster than the fastest animal on earth.
Many of our artificial vehicles run faster than the 61 miles per hour (98 Km/h) of the world's fastest cheetah and no one of us thinks it's strange, but ask anyone from two hundred years ago...
In the case of Artificial Intelligence, however, the matter is not believing or not that we will create superintelligent machines. The famous Nick Bostrom survey from 2013 shows that, at least between AI experts, the question is not if we will do it but when, given as a fact that sooner or later we will achieve the goal. Yet many aspects remain blurred and are often cause of confusion especially among the general public. I will not address concerns like the perils of superintelligent machines that realize we are useless and extinguish the human species in a matter of hours, as well as I'm neither going to focus on the opposite front where those machines instead result good and friendly but are perceived by us so different that a whole new set of discriminations and potentially strong social conflicts would emerge. Those are all valid worries and surely deserve specific coverage, but in this article I'm going to treat other aspects often way too much underrated and source of misconceptions that could potentially, as per my view on the matter, sabotage quicker progress in the field.
I am referring here to what is rarely talked about and yet it seems deeply instilled in our strongest belief system: no matter how much intelligent a machine could become, there will always be something, intrinsically exclusive of us humans, that machines will lack.
Generally speaking, that something has usually to do with stuff like self-awareness, creativity, ability to fantasize or dream or feel things like empathy, love, desire. We subconsciously give it as a fact that: no matter what, machines will never be like us, we humans will always be at a higher level on those fronts. They could solve the hardest math problems of the universe, win the hardest games ever, surpass us in diagnosis, analysis, consulting, become as intelligent as possible conceivable, precise, indefatigable, indestructible, faster, but they will never ever feel love, have inspirations or intuitions or create something truly wonderful which was not a mere recombination of what we humans made before.
The point is: this way of thinking is flawed for at least three main reasons, the last one of which is the central point of this article's thesis. But let's start from the first:
first: most of the things that we perceive as intrinsically exclusive of us humans are not even desirable when we talk about Artificial General Intelligence. In many cases we aspire to invent something that is more intelligent than us, not at all more, or even equally, emotional, sensible or creative as us. Consider a self-driving car: there is nothing creative we would like to see in such a device. Surely we want it to be very intelligent, at the point of being able to predict way quicker than us what potential problematic scenarios are about to materialize in order to prevent accidents better than we usually do. We are not looking for a car that zig-zags along the streets in creative ways while taking selfies to post on Instagram with the writing "See how good I am at avoiding people!!! :-)". So the real point here is not if they could or couldn't ever be as creative, or whatever, as us. On the contrary, what we should wonder is: is it really even possible to create something truly intelligent that is not at the same time also as self-aware and creative and emotional and sensible as we are? The answer to this could be surprising, as we will see in a few rows.
second: there are cases in which, on the contrary, we want exactly those qualities that are normally perceived as an exclusive of us humans. Consider a caregiver robot for the elderly or a simple lover/companion/sex robot. In those cases we surely want to feel the best ever empathy and human warmth of love and sex fulfillment that could ever be imagined. We are aiming here at getting strongly superior experiences if compared to what we are used to, not merely surrogates of those. Well, honestly there is nothing in the laws of nature that prevent us from achieving exactly that. Nature gave us rocks and meadows, but we invented way more comfortable beds, sofas, and chairs to sit and sleep; we were given very effective legs to move around, but wheels were not that impossible to be invented; we got horses to travel around the world, but it turned out that building the Airbus A380 was also possible; evolution gifted us with a voice, language abilities and a brain to use them, but using all that to make vocals on WhatsApp and TikTok on the phone also become an option in the end. What I'm trying to say here is that nothing so far has been demonstrated to be made by nature in a way that we humans could not augment, enhance or do it better completely from scratch. Nothing. Surely there are differences, obviously a plane does not fly like a bird, but saying that the love of a robot will always be just a fake simulation of the real love of a human is like saying that the running of a Ferrari 812 Superfast will always be just a fake simulation of the real run of a cheetah. See the non-sense? Believing that we humans are carriers of exclusive qualities that cannot be replicated nor improved in any other type of sentient being is not supported by any evidence, neither logical nor empirical. Instead, evidence of the contrary is astounding. In fact we humans are just the result of natural evolution and as such saying that nothing better than us can ever be conceived is like denying that our brains today are more intelligent than those of our ancestors from hundreds of thousands or millions of years ago. We are still evolving and even if natural evolution is very slow, nothing says that there is a limit which once reached no further evolution could be possible. Rather, the real problem is that natural evolution has limits that prevent it from evolving as extraordinary a solution as we humans can realize, and this is why we don't see birds carrying 800 passengers in nature, while there are planes that do it. History teaches us that every time man has managed to understand something about nature, he has been able later to make things orders of magnitude better than those made by nature itself, and so far this happened every single time, without any one single exception, although sometimes we need a lot of time. This happens not because nature is imperfect or because it uses a method, natural evolution, which is not that good. Quite the opposite, the method is among the absolute best that we have ever discovered. The point is that any method, however good, will produce results that will always and only be in line with what are the objectives that one wants to achieve, and nature has never had among its objectives to create the most intelligent men possible, or the most beautiful, or the best lovers or the most creative minds or the most extraordinary artists. Natural evolution has always just one goal: those who adapt best to the current context survive, the others die out. That's all. Not the stronger, or the most intelligent, or the most whatever: just the most able to adapt. The good news, however, is that we humans can use the same algorithms nature uses, the same evolution strategy, but changing the objectives according to what we want to achieve. That's why it's absolutely feasible to create an artificial intelligence that is orders of magnitude more intelligent than us, an artificial lover that is orders of magnitude more good at loving than us, an artificial artist that is orders of magnitude more creative than us, and so on. And yes we are talking here of real intelligence, real loving, real creativity, just orders of magnitude better.
"If an imperfect imitator, wishing to inflict pain, were to build himself a crude idol of wood or wax, and further give it some makeshift semblance of a sentient being, his torture of the thing would be a paltry mockery indeed! But consider a succession of improvements on this practice! Consider the next sculptor, who builds a doll with a recording in its belly, that it may groan beneath his blows; consider a doll which, when beaten, begs for mercy, no longer a crude idol, but a homeostat; consider a doll that sheds tears, a doll that bleeds, a doll that fears death, though it also longs for the peace that only death can bring! Don't you see, when the imitator is perfect, so must be the imitation, and the semblance becomes the truth, the pretense a reality."
The Seventh Sally, or How Trurl's Own Perfection Led to No Good ― Stanislaw Lem
- third: we are instinctively led to believe that qualities such as self-awareness, creativity, sensitivity are human characteristics that we have obtained either after developing a very strong intelligence, i.e. they represent the next, superior level that we have reached after achieving pure intelligence, or we perceive them as separate characteristics, disconnected from intelligence, which we also have developed in addition and in parallel to intelligence and which contribute to making us completely different from any other species. But, what if this way of thinking was completely wrong? Let's go deep on this, let's try to see if some historical observation can shed light on this.
How come we are intelligent?
We believe that our innate creativity is a trait that demonstrates how humans are unique and different from other species. We can dream, fantasize, explore mental states and perceptions that bring us to create artistic works which are able to inspire and move our fellow men and to communicate feelings that in no other way could be conveyed. After all, no other creature on our planet has ever made a poem, or a song, or even deployed any inspiring painting (until recently, as we will see).
Of the many investigations that could be made around the evidence that we are creative creatures, one of them immediately stands out from all the rest: we humans started making creative activities tens of thousands of years before achieving any, albeit rudimentary, technology advancement. Even more, tens of thousands of years before even trying to embark in any kind of technology endeavor. For example the first ever trace of a human investigating the phenomenon of electricity dates back to 600 BC when Thales of Miletus made a series of observations on static electricity from which he believed that friction rendered amber magnetic, in contrast to minerals such as magnetite, which needed no rubbing. And yet some historians believe that the oldest known paintings we made date back to 32,000 years ago, which is 14,000 years before we invented the bow and arrow, and an outstanding 22,000 years before we started developing agriculture. Oh, ehm, and by the way, all this while never giving up some good pieces of sweet music: the oldest flute discovered dates back to 41,000 years ago.
If history teaches us something, then intelligence is the creativity's next step, not the other way around. It seems clear that creativity must necessarily manifest itself beforehand, for any kind of intelligence to have the opportunity to attempt to emerge. Creativity is not some superior quality we gained after having developed a strong intelligence, neither it is something unrelated to intelligence or parallel to that. It is, instead, a primordial kind of intelligence, a precursor of intelligence, a necessary ingredient without which intelligence can't be cooked. If we want to see our Narrow Artificial Intelligence evolve into Artificial General Intelligence we must first give it the opportunity to develop creativity.
This is the central thesis of this article.
Albert Einstein was a strong believer in the importance of creativity and immagination in thinking.
It's really intriguing that we've always thought the contrary. Mind you, even in science fiction the question has always been something along the lines of: once intelligent, will machines dream too? Will they also be able to feel emotions after becoming intelligent? Once superintelligent, will machines also become self-aware or creative? There is not even a single example of a well known novel, movie or comic in which the exact opposite is wondered, that is: once the machines become able of dreaming, will they also become intelligent? Once they become self-aware, will they also be intelligent? Once they got to be creative, will this drive them to become also intelligent? And yet this fact is incredibly intriguing because this path, i.e. first self-awareness, first creativity, first feelings and emotions and fantasy, and only later a strong intelligence, has been under our eyes practically since ever, exactly for us. Raise your hand the first one who wondered, seeing a newborn: who knows if once grown up, after having developed the intelligence of an adult, he will also develop emotions, and will finally become aware of himself or will even be creative or have fantasies? Sounds strange? Of course yes, we are not born to first become intelligent and only later also develop the ability to dream, the exact contrary is true, but for some strange reason we've always imagined that for machines it would have been the other way around. This is clearly a huge mistake from our part and to be honest nowaday I truly believe that this is one of the main reasons we still haven't reached AGI so far. We focused too much on trying to understand as soon as possible how to achieve true intelligence forgetting that to learn to run you must first learn to stand and then to walk, that to get to develop the necessary intelligence to graduate from university you must have first spent some decades not only studying but also and especially dreaming, fantasizing, playing games, living in general and also fucking up big-time from time to time.
There is additional growing evidence from many fields of human knowledge that the ability to create, generate, invent, largely precedes other intellectual abilities. One of the most striking examples of this comes from the way we humans see the world around us, that is: from our visual perception system. As early as 1867, Hermann von Helmholtz, a prominent visual perception scientist of his time, already concluded that the human eye is incapable of producing a high quality image. Insufficient information coming in our brains through the eyes make vision impossible. He therefore concluded that vision could only be the result of some form of "unconscious inference", which is a term he coined at the time to say what contemporary neuroscientist Anil Seth describe with: "We don't just passively perceive the world, we actively generate it".
Square A seems darker than square B but it's not true! It's just your brain that decides to see that way.
Cognitive psychologist Donald Hoffman, professor in the Department of Cognitive Sciences at the University of California, Irvine, goes even further. According to his theory we totally generate in our brain representations of the world that not only are completely invented, but have also nothing to do with reality itself, that means: it's not even a goal for our brain to create internal representations that really resembles the real world around us, we just make representations of what interests us and in ways that are more useful for us, which, by the way, for those passionate about science fiction, is in great accord to the Douglas Adams's cloaking device based on his famous "Somebody else's problem field".
To explain what he means by representations completely disconnected from reality, Hoffman makes a very effective example. Nowaday we are used to working with files and folders on our computers, still we just see and interact with the icon representations of those entities, we don't directly manipulate the diodes and resistors or the megabytes of software involved in the editing of a picture or a text document. The table under your notebook you're seeing now while reading, as well as any other thing you're seeing right now, have absolutely nothing to do with reality itself, they are just the result of your brain generating a more convenient representation of those things.
These theories are not any new, it just happens that they are starting to be verifiable. Neuroscientist Henry Markram, professor at the École Polytechnique Fédérale de Lausanne in Switzerland and director of the Blue Brain Project and founder of the Human Brain Project, has been working on realistic simulations of biological brains on supercomputers for many years. Since his first simulations of the brain neocortex a decade ago, Markram coined his "perception bubble" theory directly derived from his observations. According to this theory, as per his own words: "The brain creates, builds, a version of the universe, and projects this version of the universe, like a bubble, all around us." Even when simulating just a few tens of thousands of neurons the experiments clearly showed how a neocortex column of the brain builds on its own an internal representation of the reality presented to it, confirming that in fact this is exactly what happens in our brains: there is a creative process that uses the very few information we perceive from the external world to invent an internal representation of what our brain decides that the external world is, and then we live only inside this bubble totally invented by our brains. "99% of what you see is not what comes in through the eyes", concludes Markram.
But how does all this relate to artificial intelligence? Is there any irrefutable evidence that in order to be intelligent a machine should really first become creative?
For now let's say that at least for us humans, and we are the most intelligent species in the universe we've found so far, this seems exactly the path evolution has chosen for us: we needed to develop creativity first, even just to try to make sense of the world around us. It could surely be that this is not the only possible way to develop intelligence, of course there could be other ways, and still there are intriguing developments in recent AI progress that tell us it could be not just a coincidence that we first developed creativity and then become intelligent.
That's what I'm going to write about in the third and final part of this article, which you find here.