Simulating Minds inside a Simulated Reality
From Simulism
Preamble
This is work in progress, and represents ideas and arguments I have been pursuing for some time. Apart from the initial reference, I have stripped out all other references to research, theory and opinion of others, and therefore what is left is pure 'idea' , pure 'speculation', and is therefore open to challenge. Please feel free to use the talk page to do just that. I would welcome debate.
Contents |
[edit] Abstract
In David Deutsch's "Fabric of Reality", it is argued that we already are "living in a simulation", but one of our own making. Every one of us constructs reality in his or her own mind, and renders it as a virtual world. Whatever is 'out there', if anything, is nothing like the world as we perceive it through our senses, and our interpretations of those perceptions are simply a way of making the best sense we can of those current sensory inputs in terms of our past experiences and our past interpretations of those experiences. We are not so much 'simulating' reality as constructing it moment by moment on the basis of evidence and hypothesis, and imbuing those constructions with meaning. This has profound implications for the nature of reality per se.
[edit] Philosophical and Psychological Considerations
The idea that reality as we perceive it does not actually exist, and that we are the constructor of our own reality is not a new concept; phenomenology as a philosphical position and radical constructivism as a psychological theory of mind posits that each person is unique and alone, interpreting and constructing the world anew at each moment in time. In constructivism schemas are created which hypothesise how the world works and behaves; evidence is gathered, and if it fits with the theory, the evidence is assimilated, and as such can be "understood" in terms of the schema; if evidence contradicts it may be ignored, sidelined or deemed to be irrelevant. Occasionally, the weight or the nature of evidence is too overwhelming to be ignored, and the situation demands a complete rethink of the schema; it needs to change, in order to accommodate the new information. This is tantamount to a 'paradigm shift' in an individual's thinking. Old ideas are no longer sufficient to explain the world as we perceive it, we need a new vision, a new rendering in order to make sense of our sensory data.
In constructing our reality therefore, on a moment-by-moment basis we are making assumptions and forming tentative hypotheses about what is actually 'out there'; we are testing those hypothesis by prodding, peering and challenging our ideas. To illustrate this, imagine walking down the road on a dark night, when you can just about make out some shapes in the distance on the footpath. As you walk down the road, you are not sure whether a particular shape is a tramp lying in the gutter, whether it is a bundle of rags, or whether it is a large dog waiting to pounce. The mere fact that you do not consider it to be a three-toed sloth or a giant jelly baby means that even when you cannot perceive exactly what something is, already you have ruled out a lot of things that you think it can't be, and are looking for the most likely interpretations of visual and auditory stimuli. In order to make decisions about what the object is, something about the situation would need to change - and it is most likely us that would be the agents of that change; we would go nearer, we might shine a torch at it, we might even be bold enough to prod it, if we were not sure. These actions provide us with quasi-experimental data which allows us to decide between competing hypotheses.
[edit] The Use of the Scientific Method in Constructing Reality
Rendering reality in our minds is therefore not as straightforward as it seems. It is based on the application an semi-intuitive version of the scientific notion, incorporating repeated iterations of observing, hypothesing and testing. If we are to create a simulated world, then the minds observing that world will probably need to be able to interact with the world and form their understandings of what that world is via the same process as that described above - essentially via some suitably modified version of the scientific method.
However, before we proceed further, given the above description, an important question should be asked: is it that I am personally interpreting or modelling the visual rendering process via an already-understood and well-documented schema (namely the scientific method), or was the scientific method itself created as a formal abstraction from observing how individuals go about the process of making sense of the world from their sensory inputs? If it is the former, then the description that I offer below will appear insightful to me but offer little to others about how to construct a simulated reality; on the other hand, if the interpretation and modelling offered above is an essential truism concerning the way that individuals choose to interpret and understand their world, then this will provide us with the basis of a a methodology under which we can devise principle for simulating minds.
[edit] How do minds Agree?
Before we embark on this, we must now address a small problem. Radical constructivism teaches us that each person, in theory, will simply make sense of the world as he or she sees fit. There does not have to be a consistent viewpoint and each person can hold one or more contradictory interpretations simultaneously. Even if an individual creates what appears to be a consistent world-view, there is no guarantee that it is correct; a new piece of information can come along at any moment and cause us to do a radical rethink, and overturn our previously-held notions. Each person is doing this at an individual level, and there is no way to reliably communicate to another person those interpretations; any attempt to do this is subject to the same process - the person trying to understand the viewpoint of another will merely form their own interpretations of what the speaker is saying, according to their own preconceptions, prejudices, schemas and models. Their interpretation of the other's explanations may or may not make sense to them, but there is no guarantee that it will be an accurate rendering of what the initial speaker had in his or her own mind.
This is somewhat stark, and we are in danger here of reducing the simulation enterprise to a very depressing nihilistic solipsism, but that does not have to be. Although we cannot guarantee that each person will view the same objects in the same way, the process will guarantee that two or more people will have shared understanding of a process; separately, they will each form their own models which will provide them with predictions on what will happen in particular circumstances. If these models are empirically effective and internally consistent (i.e. they work for the individuals), then they will provide accurate predictions which will agree with the 'reality out there', and therefore will agree with each other at least in terms of the predictions. Within the narrow confines of one particular context, the 'shared understanding' between a group of people means that empirically they will agree, but conceptually they may disagree. That is, while they may agree on what has happened in the past, and what will happen in the future, they may have differing opinions of why it happens.
[edit] Requirements for Simulating Minds
Simulating a group of minds will therefore need to have this feature 'built-in', that there is not one 'correct interpretation', but that each mind will come to its own conclusions about the way that the universe functions. This is in essence an active, not to say interactive process. Any idea that a mind can merely be 'shown pictures' or 'fed reality' is mistaken. Similarly, any idea that the simulation could possibly "fool" the recipient of a "reality feed" is naive to say the least. Any version of our current reality worthy of the name would need the simulant to be programmed with the ability to make up his or her own mind. They would be able to test, to check and to prod and poke the simulation, and would be doing this continually at a variety of levels: subconsciously, as part of the sensory data rendering process, semi-consciously, when situations are unclear or ambiguous, and consciously, when new situations or events are encountered, and new ideas need to be formed and schema need to be modified.
To begin with, simulated minds would need some basic autonomous processes, which would allow such things as head-turning, attention-focussing etc., which do not in themselves require thought to carry out, but which could ultimately be invoked at the behest of thoughts. This is a fine distinction, and it may be that many such actions begin as conscious processes, but after a while become so learned, so ingrained that an automaton-like state is reached, and thinking is concerned with processing the results of carrying out the actions rather than conducting the act itself. The parallel here is in learning to drive a car. To begin with one is highly conscious of hand and foot co-ordination. Eventually, these are carried out automatically, when one thinks "slow down" or "accelerate". Ultimately, there is no need even to think whether the car is slowing down or acclerating, the driver and the vehicle act as one, and the driver's thought might be about taking this turn or that, coming to a roundabout, or even on something else entirely.
This means that in order for learning to take place, a mind would either have to be programmed with a methodology of how to learn, or an intuitive process which would allow it to learn how to learn by interacting with its environment. However this takes place, some basic process describing this would need to be 'hard-wired' into the code right from the start, as essentially this is a given. To explore this further - it is conceivable, for example, that a baby can learn to control movements via a random process which starts off by involuntary head movements, but which result in the baby viewing an interesting object, and giving pleasure. Associations begin which tie in particular movements with a particular viewpoint, and the pleasure factor provides a motivation to attempt unconsciously or semi-consciously to control movements so as to repeat the viewings. Learning here is at three distinct levels: learning to control physical actions; learning that control can be achieved by mental effort, and learning that there is a process of learning which provides a method for tackling other situations.
The 'hard-wiring' therefore does not have to be as rigid or as intrusive as it first appears, but it does have to be at an extremely general, yet fundamental level which encompasses all the principles of learning, and expressed in clear and unambiguous relationships between tangible entities. As can be seen in the example above, in order for this to be viable it will be necessary to have an apparent separation of mind and environment; for in order for the mind to hypothesise about something, there will need to be an external component of mind, even if this is totally illusory and a fabrication of the programmer. This is because in order to learn, to interact, the mind will need to interact with something, which, for the sake of argument we might call an environment. In order to act on the environment, the mind will also need an extension of itself which can effect changes, even if these actions were purely psychokinetic in nature.
All of this implies that there would be a need to distinguish non-mind from mind, and thus the mind would need some sort of boundary at the interface. The mind, in interpreting the world it perceives would model both itself and the environment. By necessity it would construct some sort of holding vessel for itself, in order to contain mind as distinct from non-mind, and in doing so would have invented a body. When expressed in these terms, it becomes clear that in constructing such a reality, it would not be necessary to programme in everything; following the reasoning above, given a suitable learning algorithm, much of what we take to be reality would be automatically constructed by the simulants. What we as programmers would need to do is to ensure that the simulants are equipped with sufficient tools which allow them apparently to manipulate their environment, and that the environment responds in consistent and predictable ways to particular actions.
[edit] The Emergence of Consciousness
Under these circumstances, it is conceviable that conscious minds could arise spontaneously out of this situation, in the following manner: A mind, in trying to render reality internally, posits an interpretation of the universe in which it is situated. It models the universe and everything in it, including its own place in the cosmos, giving itself both form and function. In attempting to create a self-consistent model of itself, it needs to ask reflective questions about the nature of being, the nature of self and of existence, and in trying to answer these questions, it needs to perform quasi-experiments on the object, which in this case it itself, and the attempts to interpret the results and provide answers to the questions initially posed, cause the entity to raise its own level of consciousness from simply rendering sensory inputs, and possibly attempting to imbue those inputs with meaning, to a the much higher level of self-awareness and consciouness of being.
However, if the above is all that there is, then the question needs to be asked whether or not some level of consciouness is achieved by animals, as it is clear that the description given above does in some measure apply to all beings, whether animal or human. The subtle distinction may be that while animals may perform quasi-experimental renderings of their sensory inputs, and do attempt to find meaning in their environment, it is not clear that they have a model of self, nor do they appear to take the final step and ask reflective questions about the model that they hold of themselves. This may be because animals appear to have much more 'intuition' hard wired in than humans. Many animals can function as beings at a much earlier stage than humans - for example some animals can walk almost from birth, and can detect predators immediately. It may be that such 'hard-wired intuition' acts as a barrier to further development. There would be no need to develop sophisticated and dynamic models of predators if this is already hard-wired at birth.
The trick therefore in attempting to ensure that simulated minds are challenged in some way to make the final step to consciousness, is to provide them with sufficient motivation in order to raise their level of mental activity to that which incorporates self-awareness and consideration of their modelling of themselves. One way of doing this might indeed have been highlighted in the pargraph above. If we simulate human beings with fewer 'instinctive' processes than animals, this means that the human beings would not only need to learn more things, but they would also need to learn how to learn. This initial lack of knowledge appears to be a disadvantage, but would lead to a questioning of everything, and taking nothing for granted, and so, paradoxically it may be the lack of programming into minds that causes them to achieve thinking at the highest levels. An unfortunate consequence of this approach would be the time needed to set up such a simulation; in effect all simulated minds would start out as babies, and would need extended periods of time in which to mature and grow ar individuals.
Tony Fleet May 2007 (Work Ongoing)
Category: Essays
