The “imitation game” is not only the name of the famous movie about Alan Turing’s great and tormented genius . It is our first swiss-knife for knowledge.
That’s what every little baby (or small monkey) does to learn to do something. Perhaps at the dawn of civilization, it was probably the case that shown to our distant ancestors something that they imitated (of course without their understanding) and used it in the battle of survival that has led us to this day.
Tools and memes
An imitation game creates a “tool” that beyond than a physical tool, is a meme that is written in our brain. From physical and mental first, then only mental , it’s a series of instructions that allow us to do something. Similar to a small program or a set of interactions with enviroment. It’s a Dawking software version of genes.
Very important, a tool can be used without knowing it in all its features and capabilities; only what we need in that moment……we can make the example of a baboon using a iphone to break a walnut.
tools and understanding
Use and comprehension of tools/memes are not automatically tied up at the beginning and this relationship can have different levels.
And as long as the environment around baboon does not need anything more, our ape will always use the iphone to break the nuts with Steve Jobs’s peace.
On the contrary example if we have an incredible powerfull tool in the form of a magic Henry Potter’s wand, capable of solving all of our questions and problems, than can cure every illness, or take us into space and know all the secrets of the universe, we would not be tempted to find a different tool and we would not again be very concerned about how it is done. We’ll just use it.
So use comes before understanding in the early stages and is that’s what matters ultimately.
So the key is enviroment request.
From nuts to general relativity
Our learning itself is at the beginning an imitation game, starting from scratch, often through the example..
Every day we all , (adults), learn many skills, from bobbing, or the principles of physiology, or medicine etc …
Let’s learn how to program by writing down list pieces that make the mythical “hello world” appearing on screen. And if that allowed us to write any program we would stop there.
Obviously then we go on and learn by adding new programming rules. But why ? Because in that environment, writing the magic beginner phrase on the computer screen, does not take us much more into it. So we have to go on until we can build and use something new and original that allows us more.
Understanding at many levels comes later, if needed; like evolution of our knowledge.
But at the base there is always that first step of the imitation game. And for us it is enough until the environment tells us the opposite.
Using a tools is a long and complex way. And using/understanding/evolve them is even longer.
Explanation and digression : At first we only begin to use something in a way that I would call passive and primitive, without worrying about how that is done. We are like the pilot who “uses” the aircraft without much concern about the rules of aerodynamics, or the programmer who uses a library in its most elemental and primordial form to format the printing of a function.
This is similar to deep learning. we connect inputs to outputs without worrying about what’s inside mechanism.
We can only remark that right now in deep learning, for example, we use the inhuman brute force of the computer to see thousands and thousands of examples to get to know more than doctor if a tissue is malignant or not. But what did we do in reality? We have simply exploited a particular piece of the machine, that is, its inhuman capacity to “eat” a lot of data to get a result that seems stunning. So we can only observe that a human doctor need lesser examples to train his skills, also if less successfull(?). And if we want to make a comparison we will have to give the doctor thousands of years to train.
Deep learning seems…deep… imitation game ONLY . Of course we cannot actually know if this approach is another way to intelligence with a low profile approach of abstraction.
So learned meme from deep learning is very trivial (seen by an human point of view ) the meme is extremely different from human one and probably primitive. It cannot be use in other contexts. It seems not able to evolve and join with other memories to go further. UNTIL NOW .
We don’t really know. May be a good starting point, but UNTIL NOW has little true evolutionary capacity, and it seems similar to a biological counterpart of a monocellular algae that remains as it is for four billion years.
end of digression. I’d write down a few thoughts on the subject.
The difference
The human brain is capable of doing something more than the artificial neural network, which for the time is scarcely imitated by the artificial counterpart, although there are some insights in the right direction.
NB…..GANs from a certain point of view this network build a mini environment of two competing memes: an internal simulation space.
But let’s talk about our ape…..
We left the baby, the monkey, our ancestor (and the deep neural network), using for example, a stalk to open the walnut by hitting it.
But a miracle happens. Not all nuts break, so they must devise something to solve the problem. We might say we have to imagine something different, but the term “imagining” already has the smell of evolution and knowledge.
Be careful again: the crappy walnut that does not break is generated by the environment that writes a series of slightly different stories. In a theoretical world all nuts break and . But in reality it do not happens . It’s our Karma and our progress engine of evolution.
So ape needs to find a greater branch and put the nut on a rock to prevent rolls. Perhaps by chance, then more consciously, over time improves its understanding of the tool, and evolves. He has to do it because the tool he has used until now is not the most suitable: it is the environment that asks for it. At first monkey does not want to take the Nobel Prize for nuts breaking or creates the “great unified theory of nuts”.
What’s happen now? Baboon probably “hooking” in the space of other experiences memorized in different contexts, mix and remix (say in abstract) the various memories born from other individual experiences (also collective) for example changing the scope of their application.
Anything simple like a banal recombination? No of course, because the number of possible solutions would soon overwhelm its processing capabilities. And here we look at the second miracle that is the elimination game.
Here the child (or the monkey) is guided by some empirically learned higher memes and rules (that they are also tools), or better yet, they are meta-tools. The simplest of all is the similarity, (different and partially equal) that can be recombined with more success. And then learn the most powerful mental weapon that allows him to control the combinatorial explosion.
The elimination game
Call it a selection, let’s call it meme’s death, call it destruction. However it is the abandonment of those knowledge before and then of strategies that have not worked in the past that helps him to overthrow the forest of solutions.
It works like a self improvement computer. He learn to learn. And selection and distruction lead us to knowledge. Like death in nature take evolution. We simulate evolution.
So these tools become meta-tools, or more general tools. The enviroment leads him to have tools are infinitely more powerful than before. Always applying the two rules. Change and death.
In the example of the first, the pilot now for infinite ways knows a little more aerodynamics and therefore knows what to do if the aircraft has a malfunction or loses an engine.
Then summarizing what has been said so far:
Mind learning is a continuous scale of increasingly complex and general skills that can be assembled as a lego.
At beginning we use without understanding, but only by imitation.
Understanding is “on request” by enviroment.
After having accumulated a discrete number, you can use some tools that will allow you to level up your previous tools: we can define the higher-level meta-tools in self construcion pyramid of knowledge.
Third point combine these tools with different strategies trying to keep them in a computable number that can be selected, forgetting or killing those that are not useful.
If the environment requires it, we seek to find winning combinations that allow building and understanding at a more in-depth and complex level.
Throughout this process, the protagonists are obviously the environment and the selection and the case for us that shakes the nuts of our knowledge and our future.
Final personal thoughts
A strategic question. Artficial intelligence? I would say in spite of the spectacular interest and achievements, backwards enough in this evolution of tools. The results arise more from the brute force of calculation than from the refinement of the instruments. Will we do better in the future? who knows?
It is extremely difficult to figure out whether the deep learning network will create an internal competition and selection environment unless it has been implemented a priori from external programming, and this leaves us from the hypothesis of self-evolution. A little as it happened in the biological counterpart where in our test tubes the primordial broth did not lead to anything significant. But all can happens.
But It is the environment that generates our knowledge. If we take the famous example of “brain in the vat” if we remove what is innate in DNA, how does it evolve?
Another nodal point and how our brains must be educated …… and how the web needs to be made.