Megaminds, abstract evolution and the consistency machine,

aka how to build simulated realities

Megaminds, abstract evolution and the consistency machine,

aka how to build simulated realities


Mind uploading and simulated reality are the holy grail of science and technology. Mind uploading is the key to eternal life and gives godlike power in a new, “self-made” (simulated) reality. If you are not familiar with mind uploading, you can find here a short introduction.

Mind uploading is a computer simulation of the human brain. According to the present state of science, our brain is a network, built from an inconceivably big amount of neurons, and our consciousness is the result of the operation of these neurons. There is no law of physics that prevents us to copy the operation of these neurons and build software that works the totally identical way as our brain does. Sounds easy, but it is a huge technological challenge. Maybe the biggest (but last) challenge in the history of the human kind. As a technological challenge, it is only a question of time when we will be able to solve it. Thanks to exponential technologies like nanotechnology, brain scan, computer science, etc. it can be solved in a not too distant future.

Will we ever be able to build human level intelligence? The human brain is too complex to understand. We will never be able to do it…
So, it is not needed to understand how the human brain works. We have to understand how the neurons and synapses work, and we need the technology to copy the structure of the human brain. Mind uploading is the key, how we can build human level intelligence without understanding of the human brain. This would be also the key for brain-machine interfaces. We don’t have to understand the human brain to decode the signals. Thanks to big data and machine learning techniques, we only need a huge set of sample brain signals, what we can use to train a machine to read our mind. We won’t understand how the algorithm works, and won’t know anything about how our brain works, but we will be able to decode the brain signals, and build brain-machine interfaces.

If our consciousness is the operation of our neurons, the simulation would be totally identical to us. Not a copy of us, it would be us!

If you are interested in mind uploading, you can read more about it in book of Ray Kurzweil: Singularity is near

Mind uploading could be the key to eternal life, for real interstellar space travel (we should send only our digital form) and simulated reality. I think this will be the very last invention of human kind because if we can do it, there won’t be any more problems to solve. No more poverty, climate problems, overpopulation, wars, scarce resources, etc. because we will be able to build our own realities where these problems do not exist. Let’s see how…

The very first idea of everyone is to simulate the reality particle by particle as we simulate the brain neuron by neuron. But I think it is wasteful and impracticable. Accurate simulation of some particles needs huge computing capacity (think the equations of the quantum mechanics), simulation of the whole world needs much more. So, the simulation of every particle seems unsolvable. The good news is we don’t need to do this. Our brain is connected to reality through a thin channel, the sensory system, and we have to simulate only the sensory stimuli. Just imagine an apple. For our brain, this apple is not a set of particles, but a set of stimuli (visual stimuli, touch stimuli, taste, smell, etc.). So, it is enough to simulate these stimuli instead of particle-level simulation. It solves two problems. The generation of these stimuli needs much less computing capacity like particle-level simulation, and things which are not observed by anyone are not needed to simulate. If we put the apple in a box, it is not needed to simulate any more, because we don’t see it, don’t taste it, etc. The apple does not exist for us until we don’t get it out. The things that are not observed do not exist. Some interpretations of quantum mechanics look at reality in the same way. This solution radically reduces the needed computing capacity, but there is a problem with it. We need particle-level simulation for generating the correct stimuli. In the case of an apple, the simulated apple will act like an apple only if we simulate every biological, chemical, and physical activity. After some week, the apple should be robbed, if I compress it, its levier runs out, if I bite it, it gets brown, etc. It is only possible if every physical activity is simulated which are processed in the apple. An easier example, if I bite the apple and close my eyes. The apple won’t be observed, it is not needed to simulate, and it doesn’t exist anymore. What will happen if I open my eyes and continue the observation? The apple should be there with the bite, otherwise, it will be realized, it isn’t real, but only a simulation. So, the system has to store the state of every object and continue the simulation of the non-observed things. We got back to the starting point: we need near-infinite storage and computing capacity (particle-level simulation) if we want to build a perfect simulated reality. The good news: there is a tricky way to solve this problem. In fact, physical laws are not relevant, but the expectations of the observer. When I open my eyes, I expect the apple to be there with the bite. We can generally say, every act is an experiment where we want to prove we are living in a computer simulation. The experiment failed if the result is different from what is expected. So, the simulator should generate stimuli that meet our expectations. In this case, we won't be able to prove we are living in a computer simulation, because every result of our experiments will show it is the reality. This is the point where simulated reality meets mind uploading because the simulator has to read our mind and the easiest way to do it if it is also simulated. So, if the simulator can read our mind, then nothing has to be stored and only observed things (set of stimuli) have to be simulated.

In the rest of the article, I will write about the simulator, and how can it work. I’m not an expert in neuroscience, so what I will write about is built upon what I’ve read on the topic. As I’ve written, I think mind uploading (and mind-reading) will be necessary to build a “reality equivalent” simulated reality. Don’t forget, that we are building worlds for neural matrices (uploaded minds), and the goal is to build a network of them by keeping some rules. If we look at these minds as matrices, and the simulations as an algorithm that is running on them, then it will be easier to accept things like non-linear time, what I will write about later.

As I’ve written, in a perfect simulated reality, it is enough to simulate things that are observed if the simulation meets our expectations. In this case, it is not possible anyway to prove it is a simulation. It sounds like the brain should work in a reverse way. When we observe the world, our brain converts sensory information to abstract things, and then memories. When the simulator builds the expected reality, this should be processed backward. Remember to the sample experiment with the apple. When I bite the apple, my brain creates a memory about the bite in the apple. When I close my eyes, the apple won’t exist anymore. When I open my eyes, the simulator has to render the apple (a set of stimuli) based on our memories. It is something like a reversed perception. I think, our brain can do this because when we are dreaming, it does something like this. This is also the base of Google’s DeepDream algorithm, which was used to generate that famous hallucination like images. These images are generated by backward processed convolutional neural networks. What we see is the visualized knowledge (memories), which are stored in the network. So, I think, DeepDream is a good name for this algorithm, because the process is something like our brain do when we are dreaming. So, this solution sounds good for building simulated realities which meet the expectations of the observer. But it is not enough to build the whole reality because there are many details in which the brain has no information or any memories. Every detail which was not memorized, every place which was never visited are unknown things. These details can be built by the simulator from freely generated stimuli and details which are read from the observer’s mind.

If you are interested in how our brain projects the reality, watch this TED talk of Anil Seth (Cognitive neuroscientist):

Of course, a world like this would be really lonely if there wouldn’t be some other mind besides us. To keep the world consistent, it has to meet the expectations of every mind. If I left my apple somewhere, the other people should find it there. Every observer has to experience the same reality. The simulator has to operate every mind parallel on a backward way to build a perfect, consistent reality. It is something like the reality would be a mosaic, and every mind would add a part of it. If we are thinking in “subsystems”, our simulated reality looks something like this: Every mind are connected to an upper, common layer through the sensory system. The minds are getting stimuli from this layer (observe the reality), and also send stimuli which are generated from memories by backward operation of the mind (reality projection). This common layer is responsible for building a valid intersection of these stimuli, and a reality (set of stimuli) which are the same for everyone. This layer is responsible for keeping the reality consistent, this is why I call it: “consistency machine”. So, the consistency machine is the simulator itself, but it simulates nothing, only creates a valid intersection of the set of stimuli, this is why I think the consistency machine is a better name for it.

In many brain model, the brain is a hierarchical and modular neural structure. Small modules are built from neurons, and other modules are built from these modules, some bigger modules are built from these modules, etc. There is two-way communication between the modules, and conscious information is the result of these. When any stimuli are coming from the sensory system, it activates low-level modules, which are activating higher-level modules. These higher-level modules activate higher-level modules, which respond back, etc. This system is massively parallel, and this gives its huge power. If we see an apple, many low-level modules are activating, which are assigned to curves, colors, etc. These modules activate higher-level modules, like the “module of a sphere”. The “sphere module” activates higher-level modules, like “planet module”, “ball module”, etc. and the “apple module”. But higher-level modules respond back, it cannot be a planet, because it is smaller, it cannot be a ball, etc. After a short time, the “apple module” wins, and the concept of an apple will come up to the conscious level with every connected property like taste, touch, smell, etc. So, our brain is continuous “rustling”, signals are flashes up and down between modules, and after a short time, a concept is manifesting on the conscious level. Thinking is something similar process. From a thought, our brain associates the other thought, and new concepts will be generated through the process.

If you are interested in the hierarchical pattern recognizer model of the human brain, you can read more about it in the book of Ray Kurzweil: How to create a mind

In the case of an apple, we can imagine, how this process works, but can this simple method operates such a complex thing like the human brain? It is hard to imagine how this works, but it is not the model’s fault. We cannot really imagine how these huge, massively parallel networks work.

The human brain contains ~90bn neurons. 90bn small processors which are connected on many ways. But how much is 90bn? Let’s count it! If we can say a number in 1 second then the counting would be >170 000 years. Near the whole human history. It is an inconceivably big number. But really can the operation of these simple neurons results the consciousness? There are many cases when use of a very simple rule results a very complex thing. Let’s think of DNA. Only four base (adenine, thymine, guanine, cytosine) can describe every living form on the Earth. Or in the computer science. A simple logic function (NAND) is enough to build every logic circuit, every processor or algorithm, every program which are ever existed or will exist.

So, I think, it is an acceptable model of the human brain. I’ve written about the hierarchical model of the human brain to show, how it fits into our simulated reality model. The consistency machine can be simply one or more layer upon the uploaded minds, which are themselves also a set of hierarchical modules. Uploaded minds communicate to these modules through the sensory system like the modules communicate with each other inside them. In this system, the sensory system is not a “window to outside”, only a level of the organization. So, maybe, it is possible to add the consistency machine as new modules to the uploaded minds and build a “megamind” from them which is able to simulate reality without any other component, using the natural behavior of the mind. In the system of interconnected minds, individuality is on the middle layer, not on the top, and reality is something like a collective, endless dream. A megamind solves the consistency problem in the following way: when individual minds project some components of the reality, the upper layers (what I’ve called ‘consistency machine’) merge them and send back the generated reality. If this reality is inconsistent with a mind, it sends back the ‘error’, and the ‘consistency machine’ modifies the reality. After the end of this communication, there will be generated a consistent concept of reality. This is exactly the same way, how an individual mind creates a conscious concept from the inputs which came from the sensory system. This system can solve the consistency problem not only in space but also in time. As I’ve told, the reality is generated from memories of the minds. If something is not stored in any individual mind, then it does not exist. If nobody has any memory of an event, it has not happened. This is why a megamind can keep consistency in time. If something happened in the past which results in consistency in the future, the contradiction can go down in the layers of modules and can change the memories. In this system, memories are the past. So, if memories are changed then the past is changed! It is something like ‘time-traveling’. Something bad happened? No problem! Go back in time, and change it! This method can keep the consistency, and also protect the reality because if any mind finds any proof of simulation, the ‘consistency machine travels back in time’ and make it hasn't happened. Time is non-linear in this world. Linear time is only the way how individual minds experience happenings. Time doesn’t exist globally, but only on the level of individual minds.

So, what is a megamind? It is a very special living form. One mind, one living thing for an outside observer. It is something like a god in its own universe because it creates and operates it. But it has no individuality because inside, this superintelligence is split into many individual minds. One huge mind dreams itself as many minds which are living in a simulated reality. This state of consciousness is something that we cannot imagine. It is over the singularity. But what would be the physical form of it? A giant biology-based brain? A huge computer, like a Matrioshka brain? Maybe. But for the end of the article, I have a more extreme variation. How good the universe is a megamind?

In the current state of the art, our universe has been created from nothing many billion years ago by the big bang. It is built from particles which are based on one simple structure: the superstring. Every particle is a form of these superstrings. There are some physical rules what operate the system and change the state of these particles. Why these rules? Why superstrings? etc. Nobody knows. These are the axioms of our universe. From the perspective of math (which is the best perspective, I think), the Universe is a big state matrix (state of the strings), and state-changing operations (physical rules) are operating on it. If we look at our Universe from this perspective, a megamind is not really different. A megamind is built from neurons instead of strings (or particles), and there are some rules (like physical rules) that operate the system. Every neuron has a state, and interact with each other by the ‘neural rules’ (this is the ‘physics’ of the megamind) like particles interact with each other by the physical rules. I think, if you leave it alone for some billion years, life can be formed spontaneously like it spontaneously formed in our universe. I think evolution can also work in this universe, but here the result is not life in the traditional meaning, but the consciousness and the individual minds. This is why I call it ‘abstract evolution’. So, I think, a megamind at least as the good universe like ours. Can we say anything about the other universe if we know only our universe? Not really. But just one thing! If we were living in a megamind universe, can we find any proof of it, or we find it exactly the same as our current universe? I wish you good thinking…

If you are interested in the neural network theory of the Universe, read my article on the topic: