Emergence as the starting point to Artificial General Intelligence
Humans’ zealous pursuit of knowledge led them to decipher lots of phenomena. Yet, we still dazzled by many others, as elementary as they may appear. Fasten your seat belts and let’s begin our journey with one of the most charming mysteries of our universe: Emergence.
As you may have heard before, emergence is complexity derived from simplicity. It is usually summarized through the famous statement: “The whole is greater than the sum of its parts”.
Now, let’s leave the poetic language aside for a moment. More formally, Emergence refers to a rich and global behavior only appearing when a large number of entities begin to interact with each other. But how this kind of systems choose a coordinator to define and align their complex duty? It is simple: They don’t.
Actually, one of the leading characteristics and strengths of emergence-based systems is the exemption from a “top-down” hierarchy.
That being said, emergence is not just a fascinating quirk of science: it is nature. Take a look into these atoms forming molecules, then proteins, cells, organs to finally make up individuals. How a set of atoms are able to form a human being! And who said “finally”! You could go further if you want to see how individuals generate cities or nations. One extra step, have you ever thought about the Internet from an “emergence point of view”? It gave billions of humans the access to the entire human legacy. And what is emergence but big numbers × exchange.
What makes US unique ?
What do you think about a quick time travel? About 50,000 years ago? You’re already wearing your seat belt, so let’s go!
Back then, Human’s impact on the world was insignificant. They were not the biggest, the tallest, the fastest, the most powerful, nor those who live the longest. One might even wonder how this species survived so far. Ultimately, if we compare Human’s physical capacities with those of the other inhabitants of this planet, there are not many categories in which humans could hope to take advantage.
How about Intelligence? Here is the area where Human can feel proudly unique … ooor not 😳
First of all, despite their outstanding intelligence, humans have not yet been able to define the term “intelligence”. Anyway, let’s dig into what it commonly encircles.
If by intelligence we mean the ability to solve complex problems, then we should think twice before comparing our intelligence to the crows’. If it refers to the use of tools; monkeys, birds or octopuses do not let humans reign alone. What remains for humans to prove their uniqueness? Self-awareness? Long-term memory? Empathy? Games or humor? These criteria have all been observed repeatedly in many species. Even the human creativity should be considered with humility; back then ants had already invented agriculture, farming, social classes and chain work. And to be honest, many human inventions are basically replicas of the living world around them.
When each human was –literally- minding his own business, an important event came out to change the history of (at least) the whole planet: We started talking to each other. We discovered Language.
Henceforth, each individual could use Humans’ collective intelligence as if he produced it all himself.
Language gave humans the ability to communicate and consequently to cooperate flexibly in very large numbers. There are other species on earth that cooperate in relatively large numbers like bees or ants, but they are kind of rigid in their “thoughts”, in the sense that they can do nothing but executing the same actions again and again. And yet, relatively to their size, they have made miracles as we’ve seen before, but humans do more. Of course there are lots of other social creatures that cooperate flexibly with each other like hyenas or chimpanzees but just in small numbers, so nothing impressive emerges.
So yeah! What made humans unique and therefore let them triumph over this planet, is their ability to add flexibility to the common ingredients of emergence: number and cooperation.
Humans can cooperate in a very flexible and sophisticated way, in large scale, even with strangers.
So that’s the situation: Bring a lot of entities specialized in doing some basic stuff, provide them with the ability to cooperate, you will already be able to see emerging phenomena that are not “programmed” in each entity. If you succeed in adding an extra feature of flexibility, i.e. your beasts are not just obeying your hard-coded rules utterly, then you are in the right way for creating almost human-like societies, a.k.a. Artificial General Intelligence or AGI.
Game of Life
Just one more, maybe technical, lesson to take before removing your seat belt. A lesson I did not learn from nature phenomena, but from artificial ones.
Have you ever heard about the game of life? The following video shows fascinating manifestations of this concept. Basically, it’s a simulation of a very basic two dimensional universe portrayed like a grid of square cells. These cells have two possible states (dead or alive) that are determined by a very few elementary fixed set of rules.
Can you see the amount of complexity generated by 4 simple rules? Can see how “societies” can thrive or destroy themselves? The first time I discovered this concept I was like:
The thing is, before seeing the game of life I was fascinated by the concept of emergence, but I had my doubts. I was thinking that perhaps ants are not completely stupid creatures, maybe there is something hidden in them that triggers this fascinating behavior called emergence. After all we do not know exactly how ant brain works. Idem for bees, birds and so on. Here on the other hand, I know PRECISELY and EXHAUSTIVELY to the exact semi-column, what is there in the “brain” of these creatures out there in the screen; and yet I was watching some surprisingly bizarre manifestations that are ONLY the result of the interaction between these tiny units.
Wait but where is the lesson here? Is it the indisputable emergence’s proof of the existence? Well I would say yes back then; but the point I wanted to share with you here is the importance of the initial states.
As you might have already noticed, the initial state of the environment is a key element in conceiving the behavior of the whole system. And we can see this clearly in the game of life because we have the ability to see the big picture, and the other parameters are invariant.
So what if we add some initial knowledge/intelligence to the systems we are creating and stop starting from “tabula rasa” scenarios? Needless to say that we, humans, and lots of other animals are born with some inner data, or if you want we are set to an initial state 😎. Actually, a significant part of the learning process of our children is done before we can tell them “this is a cat”.
With that in mind, the way we are doing Machine Learning/Deep Learning (ML/DL) today (i.e. full focus on Supervised Learning techniques) does not appear to be the best path to AGI. Maybe we should throw a wider net.
Put it all together
Now that we have all the ingredients of the recipe, let’s put it all together using a practical and engineering mindset.
Statistics affirm that in the next few years, we’re going to have more than 50B connected devices, which is to say around 7 devices per person. Remember! Each of which is equipped with connectivity. And you can deduce for yourself what is the result of big number × exchange.
This is what everyone refers to as the Internet of Things (IoT).
Now imagine that the rules leading the behavior of each device are not hard-coded, but can improve over time. These “machines” can now “learn” from past experiences, either theirs or others’.
Put this way, basically the magic wand Machine Learning tools provided us with is: Flexibility, a key concept in mimicking humans behavior. In fact, before ML all we were able to do was to “program” precisely what a machine will do. It has no choice to do unpredictable moves. Now on the other hand, even the experts are in awe of the unprecedented choices AI can make. “Move 37” is now a slogan and an admirable illustration. It refers to the outstanding move that Google’s ML algorithm made, and which no human would ever think of doing.
Geoffrey Hinton, the Godfather of ML/DL, says that “maybe the ways we’re doing things today –in AI- aren’t the best way of doing things, we should think about alternatives to some of the basic assumptions we’re making”.
The way we’re doing things today is based upon supervised learning. Maybe the answer is finding the balance between how much we have to learn versus how much innate knowledge we need at the beginning, in the initial state.