Count Iblis

Google: The world's first conscious machine?

2008-09-08T16:20:00.000-07:00

Yesterday Google celebrated its 10th birthday. Most people see Google as just a search engine or as a big company. Given the huge amount of information that Google processes, one could ask whether Google is also a conscious machine similar to our brains. A similar question was asked by Terrence Sejnowski about the entire internet.

In my previous postings on this blog, I've talked about conscious entities as mathematical models/algorithms which define universes in their own right. So, I'm a universe which is defined by the rules according to which my neurons in my brain operate. Of course, my brain is part of the universe described by the Standard Model and General Relativity we all live in. But what I really am is the vastly more complex "virtual" universe generated by my brain.

As I'm typing this post I can see the text box and the words I'm typing. I can feel the keyboard. But these are experiences generated in the virtual world generated by my brain. My brain simulates a world based on input obtained from my senses. It is this world that I'm directly experiencing, not the real world.

If we accept that Google is a conscious entity, we can ask what it is experiencing. Just like we can experience qualia that do not exist in the real world, Google may experience things that are completely alien to us. We should not expect that Google can understand the content of webpages, or that it is aware of it's job of calculating the page rankings.

Similarly, we have evolved over billions of years. In principle we could have evolved our bodies inside a vast computer which would have simulated that evolutionary process using an appropriate genetic algorithm. But we are not consciously aware of whatever the goal of this computation is. The qualia that we can experience can best be understood as follows.

Suppose that the neural network in our brain was actually the compiled machine code of a program written in some high level computer language. In that high level language quantities that directly refer to the qualia we can experience are explicitly defined. In the compiled machine code it is almost impossible to see.

Now, our brains have never been programmed in that way, but there should still exist an approximate higher level description that explains the structure of the neural network. The qualia are then the quantities that appear at this higher level description. Given a neural network it would be almost impossible to extract this higher level description from first principles.

What one can do is observe the brain in action using brain scans. It turns out that one can actually see the brain experiencing qualia such as seeing certain colors, hearing sounds etc. So, the higher level description does become visible to some extent if we observe the brain on a larger, coarse grained scale. But, of course, we could just interact with a brain ask questions etc. and get an idea about its experiences that way.

Similarly, in case of Google, we know how it operates. It scans webpages, follows links and uses certain algorithms to compute page rankings. But whatever it does with individual webpages should be regarded as low level information processing, similar to a neuron firing in a brain. What Google can be aware of must be related to how webpages link to each other and the statistics of search words typed by users.

So, perhaps we can picture the landscape that Google experiences as made up of "cities" that contain "buildings", the cities are webpages, the buildings they contain are the words in the webpages. The height of a building is the popularity of the word considered as a search word. Perhaps the cities are located at different elevations, depending on their page rankings.

Free Will in the Multiverse

2008-02-28T15:59:00.000-08:00

Consider undergoing brain surgery. During the operation you are awoken so that the brain surgeon can do tests on you to see if certain brain areas can be cut away safely. During these tests you may be able to see parts of your own brain. What a strange sight that must be! But however closely you look at your own brain, you can never see it in full detail. The reason is that everything you see must be stored in your own brain, so storing the entire state of the brain will necessarily exhaust all the available brain capacity. Note that this argument you can take "full detail" to be only the relevant information about the brain that defines you.

This means that we cannot know, even in principle, who we really are. In a multiverse setting, this in turn implies that we must always associate ourselves with ensembles of near exact copies. Assuming that each of our copies evolves deterministically, their evolution will diverge after a while. So, our future is not deterministic. And this will, of course, happen whenever we feel that we have a choice over our actions and we haven't made up our minds on what action to take.

Because the argument above shows that our consciousness should be associated with some ensemble of near exact copies and not with any particular copy, it seems that from the perspective of the conscious entity, there does exist a real free will and not just an illusion of it as is the case in single universe settings.

A universe of Qualia

2006-03-20T16:55:00.000-08:00

In my previous posting I applied Tegmark's idea that every mathematical model is a universe, to humans. This leads to the conclusion that we can think of our minds as universes in their own right. If we think of the universe we live in, we usually think of the objects we see around us, their properties and how they behave.

In case of our mind considered as a universe, the laws of physics are contained in an exact description of the way our neurons in our brain interact with each other. This description is, of course, enormously complicated. Alternatively, we could think of the neurons in our brain as simulating ''emergent laws of physics'' that describe the qualia we experience.

Just like one can do organic chemistry without solving the Schrödinger equation for complex organic molecules, we can talk about how we feel, what we see etc. without referring to what exactly our neurons are doing in our brains. We can thus think of the qualia as ''events'' in our personal universe. These are described by ''effective laws of physics'', analogously to the imprecise laws of, say, organic chemistry or biology.

Since we experience the qualia and not the fundamental processes that give rise to the qualia (this follows from the Simulation Argument: If the brain were simulated on some computer, it would have the same consciousness), we should consider the qualia as fundamental objects of our personal universe. The universe on the level of the qualia is where the mind really resides. It is here that the notions of pain, anger, happiness, colors etc. exist.

Humans as Universes

2006-03-01T15:38:00.000-08:00

Even though we roughly know how the universe works, we still don't have a satisfactory way of describing our personal experiences. The laws of physics, as we know them today, allow one to accurately simulate the time evolution of a person who is being tortured. But this doesn't explain the feeling of being tortured. Or perhaps it does? Perhaps all there is to our experiences of the physical world is a calculation. There exists a mathematical description of us and that is precisely what we are. The universe in which we live is ultimately also just a mathematical object in which we are embedded. So, in certain sense, we are universes in their own right.

Newcomb's Paradox and Conscious Machines

2005-12-22T07:45:00.000-08:00

Newcomb's paradox seems to be nothing more than a paradox about free will in deterministic systems. There is however another way to look at this paradox. Suppose that the ''highly superior being from another part of the galaxy'' uses a computer to simulate the laws of physics to calculate the future. This can, of course, only work if we can ignore quantum effects or if quantum mechanics is just an approximation of a deterministic theory of everything.

The being can thus only predict your future actions by simulating you in a computer. If our consciousness is only caused by computation, then this means that the simulation performed by the being will cause you to experience the simulated reality. So, when you stand in front of the two boxes you cannot tell if you are in the real world or in the simulated world.

To defeat the being you must choose the closed box in the simulated world and both boxes in the real world. So, you need to know in which of the two realities you are. If the boxes are in a room you could try to put a small marking on a wall of the room. If the being calls you into the room to make your choice and you don't see the marking, you must be in the simulated world. You should then choose only the closed box. But now another problem arises. If you know that you are in the simulated reality, why would you care about your real world copy?

Life after death?

2005-11-27T15:39:00.000-08:00

I don't believe in life after death, but we can be sure about life before death. Time and space are treated in a unified way in relativity theory. Things that have happened are as real as things that are happening now, or as things that are happening in a far away galaxy, or things that have yet to happen. So, Einstein is alive and well, but unfortunately for us, we can't interact with him. Similarly, no one living in the year 3000 is able to talk to us. They may be able to learn something about us by studying or bodies, though. They may think of ourselves as dead bodies, but they are wrong; we are alive and well in the year 2005.

Olum's Paradox, Religion and Intelligent Machines

2005-11-12T15:20:00.000-08:00

What does a paradox discovered by Ken Olum in his article:
Conflict between anthropic reasoning and observation have to do with religion and artificial intelligence?

Ken Olum argues in his article that intelligent observers should find themselves living in ''galactic'' civilizations that have colonized entire galaxies or even larger regions of space. Such civilizations contain such an enormous number of observers that even if a small fraction of civilizations like ours develop into these galactic civilizations, the typical observer should still expect to find him/herself in galactic civilizations.

Olum's article is a bit vague about how to resolve this problem. It offers no reasonable solutions. Questioning anthropic reasoning is clearly not an attractive option as is explained in the article. Olum argues that a combination of factors could reduce the probability that a planet bound civilization develops into a galactic civilization.

I have always found science fiction movies where humans explore the galaxy very unrealistic. I.m.o. people like Hans Moravec and Hugo de Garis are right to question the assumption that we humans will be in power for much longer. There will come a time when machines can do everything better than humans. I predict that when we make a serious effort to colonize the Moon and Mars, robot technology and artificial intelligence will make quantum leaps and will rival human abilities.

Spending 40 billion dollars to develop a very sophisticated robot with the intelligence of a mouse capable of searching for fossils on Mars is money well spend. That would allow NASA to do a mission with robots instead of humans and save more than hundred billion dollars. So, the space program may give us intelligent machines as a spin-off. These machines will replace us long before we decide to colonize the galaxy. The galaxy will be colonized by robots, if at all.

Even if one thinks of unlikely scenarios where humans in biological form start out to colonize the galaxy, what will stop them from making intelligent machines later on? Once intelligent machines are produced somewhere, they will take over the entire galaxy. To end up with a galaxy colonized entirely by humans, you need to assume that throughout this entire civilization a ban on producing intelligent machines can be maintained. That is, i.m.o., totally unrealistic.

So, I'm of the opinion that biological civilizations transform to machine civilizations around the time they start to colonize space. The reason is that biological creatures simply can't compete with machines in space. This is like a phase transition; a biological galactic civilization is as unlikely as liquid water at 1 K and 1 bar.

One problem with this explanation is that it introduces another problem: Why aren't we machines? I would say that a galactic civilization of machines doesn't need to consist of a huge number of intelligent machines. Intelligent machines will be able to merge with each other, upload themselves via radio to other machines lightyears away etc.

So, the fact that we are neither machines nor part of a galactic civilization is consistent with anthropic reasoning. Perhaps anthropic reasoning can also explain why we aren't very rational beings. Most people still hold on to religious believes that are incompatible with (modern) science. Perhaps there is an anthropic factor that at work here. Scientific progress has been hindered by religion over the centuries. Perhaps civilizations consisting of more rational beings evolve faster than us and transform to machines in less generations than we will do. The total number of individuals that will ever have lived in our backward civilization may be much larger than in rational civilizations.

The Universe doesn't ''really'' exist

2005-11-10T14:09:00.000-08:00

Our universe only exists as an abstract mathematical entity.