What is Intelligence: Alien Intelligence


This work is about intelligence so it is necessary to contemplate what alien intelligence might be like even though we haven't encountered any till date. There are those that say humans are the only intelligent beings in the universe, but I think that is not correct and that the size of the universe in comparison with the size of the earth makes the possibility of conscious (using this word in the way it is understood regularly) beings might inhabit other worlds or places.

There are many possible answers for why we have not yet contacted other beings but I am not going to go into that now because what I am focused on is a scenario where there are intelligent beings out there and what their intelligence might be like.

I do not think that intelligence is limited to human neural architecture. I think that human neural architecture is just a hard implementation on one kind of substrate that runs software that makes the human act with intelligence.

Intelligence itself is just algorithms and data structures and can be realized in a variety of hardware and software systems. So if we were to encounter intelligent beings out there or if they come here I don’t think we should expect them to possess hardware that looks like our brain in all cases.

If we encounter humanoid aliens then it is possible that they might be using similar kind of architecture like us and are probably using this kind of brain we have, and if they are carbon-based then the probability of them having brains like ours increases but if they are not humanoid then we could have them doing their cognition on any kind of substrate.

The core algorithm/data structure that is AGI I have proposed somewhere in this work is the network. I even subscribe to Stephen Wolfram assertion that space might be a network. To me, everything in this our universe can be represented as a network, as nodes and edges and the operations between them. It is the most universal form of representation at least in this our universe and thus it is the intelligent principle itself.

Even DNA sequences can be represented as gene-networks. Anything can be viewed as a thing and the connection between the thing and another thing. Our current Neuron-Weighted-edge system for doing AI is just the beginning of an exploration of the power of network representation. We should note that a tree, in computing, is just a specialization of the much more general framework of networks.

So indeed alien intelligence can be implemented on a crystal, it might be implemented in clouds, it can be implemented as a whole planet and all the ecosystem could just be computational nodes that perform some kind of computation that contributes into the cognitive capacity of the whole planet.

We might not be able to recognize such intelligence if we limit our understanding of intelligence to the kind of task that humans perform, but if we take a broader perspective and realize that intelligence can be done by computation and many systems are computationally equivalent and this means even systems in nature are capable of sophisticated computation, then it is not far fetched that our planet or any other planet, star or even galaxy is an intelligent entity performing computation that is as sophisticated as what a human would perform but having goals that are quite alien from human goals.

The goal system might be the seed initiator for the development of any kind of physical mechanism for performing computation. We might have evolved this way because of our seed goal that guided the evolution of our intelligence from primitive beginnings up to the point we are now.

The goal of an alien being might have evolved the corporeal identification in such a way as to suite such goals and as such we might encounter alien entities that do not speak or even see but can act in ways that disclose intelligence to us.

If we want to think about intelligence it will do us good to distinguish human based smarts, which is not what I am talking about, from the raw algorithms of intelligence.

There is nothing intelligent in a deep learning network like a convolutional neural network. We supply labelled data into the network as numbers, and it transforms those numbers through certain kinds of computations, in layers until it arrives at the same label as what was initially labelled. When we arrive at our original label through training, we have gotten a state that represents the image in a higher representational form that generalizes well for all images of the class of which the original input image belongs. With this generalization when we query the trained network with data it has not seen before it is able to say with a high degree of confidence what the correct label is.

The effect or the output of the network is something that we call an intelligent action or effect but then there is no intelligence encoded into the network itself. When programming intelligent action in a high level language we usually use conditional statements like if and else to make decisions. When using this high-level constructs we are able to encode that which we may view superficially as intelligent behaviour.

If we are writing a program to control the navigation of some robot in a room, we can write some instructions like:

If inFrontOfWall(): turn left;

If atDoor(): turn knob;

Assuming that the low-level functions inFrontOfWall() and atDoor() have been fully defined in the larger program our robot running this code will display behaviour that is “intelligent” to a cursory observer that has not seen the code or doesn’t even understand programming.

The robot will appear to have some kind of intelligence as it is seen making decisions about what to do when encountering different scenarios. If the programmer loads so much of this kind of code into the robot, it will appear to display behaviour that is considered for every reason as intelligence in the domain of application that it is involved in. It is this kind of power the brought about the enthusiasm about AI in the GOFAI days and that made people make bold claims about achieving intelligence with computers as far back as the 60s, but from experience, we have seen that this is not the right path to take.

If you look at the code snippet above you will see a function: inFrontOfWall(), how would the robot know that it is in front of a wall or any other obstacle? It would do this with some kind of range finding sensors. The power of today’s deep learning is that with a cheap camera and some deep learning software, the robot will be able to see that what is ahead is a wall because it can now see pictures. This summarizes what is known as AI in the current times, the other part of AI that is still perplexing for researchers is that of generating meaningful information.

GOFAI systems were the kinds that did AI with this kind of conditional statements. If you read the code you can see some kind of intelligent decision making encoded by some programmer. But if you see the layout of some deep neural network there is no “intelligence” encoded there! Its just numbers passed from one function to another till some final result is arrived at and read out by some human.

Intelligence in generality is more like the architecture of a deep neural network and less like the If-Else systems of the past. It is with these kinds of networks that we can arrive at some general description of intelligence that can perform all kinds of sophisticated tasks which may gradually approach what we may call AGI or the cooler term StrongAI.

Hardware like Google TPUs (Tensor Processing Units), tries to capture much of these network features in hardware giving rise to a new kind of CPU. It is on these kinds of architectures like the Google TPU that we will write the kind of high-level languages that will enable us to describe cognitive architectures which can in most cases pass for human intelligence or even go beyond it.

Alien intelligence, whenever we might encounter it will be equivalent to human intelligence. It might be specialized to perform certain tasks better than us no different from the way a pure mathematician is much better at math than a beach bum, but it will be equivalent to human intelligence nonetheless. That is because computational equivalence is a law of nature.

I am not assuming that any extraterrestrial being we encounter will be more intelligent than us, but if we should see a spaceship in the sky right now from another planet the probability that they will be smarter than us in many domains will be higher just because they were able to cross the vast chasms of space and reach us here. But the possibility that the only thing they know is physics and engineering which enabled them to find a trick to bypass the limit of the speed of light is very real.

They might be as ignorant of literature or arts like a caveman so therefore in those domains we might be superior to them.


Popular posts from this blog

Its not just about learning how to code

Nigeria and the Computational Future

This powerful CEO works remotely, visits the office only a few times a year