INFERENTIAL LEARNING What is inferential learning? It is learning by inference, by logical deduction or induction, from assumptions and observations to conclusions. Take for example the following deductive sylogism, All dogs are animals. (Assumption) Joey is a dog. (Observation) Therefore, Joey is an Animal. (Conclusion.) The consclusion is true if both the assumption and the observation are true and the logical form is valid. The conclusion may yet be true even if the assumptions are false or the logical form is invalid. But if the assumptions are true, and the logical for is valid, then the conclusion MUST be true. When using logic to infer the truth of conclusions, certainty of conclusions depends on certainty of the assumptions and certainty of the observations, and of course certainty of the validity of the logical form. The assumption in the above case is true by definition, so there isn't much to proving it true. But in many other cases, the underlying assumptions to conclusions are much harder to prove true. Sometimes one can prove that the assupmtions CAN'T be proven true and thus the truth of the conclusions can be held in abeyance forever. The machine certainty theorem is very simple, it says basically: A machine can't be certain of anything. A machine can't be certain it exists, it can't be certain of changes in itself, and thus can't be certain of time, and it can't be certain of cause and effect and thus can't be certain that anything else exists either, let alone what its properties may be. Here is the proof. A machine learns about the world around it by lookikng at effects received in itself allegedly caused by the world around it. From these effects the machine, hypothesizes, theorizes, and INFERS about the various causes that may have created the effects. The following sylogism is used exclusively in such inferences. All effects are caused. (ASSUMPTION) I have received an effect. (OBSERVATION) Therefore there must be a cause. (CONCLUSION) From above we know that the only way the machine can be certain of the conclusion, is to be certain of its assumptions and observations. Well let's grant for the moment that the machine is certain of its observation, namely that it received an effect. Notice that the word 'observation' *MEANS* 'received and effect' and 'received an effect' *MEANS* 'observation.' We have already said that a machine can't know it has changed state with certainty because present state does not prove prior different state, but let's grant the machine the truth of its observation that it received an effect anyway. All that is left then is for the machine to attain certainty that 'All effects are caused'. This means that if the machine changed state there must have been something else out there that must have changed state earlier which caused the effect the machine received. *THIS* something else is the object of its learning. Ok, well we all believe that every effect is caused, that changes in state in an object in present time flow from OTHER changes in state in other objects in prior times. This flow of changes in state is called a causal pathway. Each object that changes state later in the causal pathway is called a 'symbol' for the prior object that changed state before it which we call the 'referent.' The later event symbolizes and refers to the earlier event that caused the later event. The first event or object that changed state to start it all is the original referent. We are not looking for a First Cause here, but an earliest one of significance in the causal pathway we are interested in. So if C is the effect of B which is the effect of A, then A is the original referent because A 'started it', and B is a symbol for A, and C is a symbol for A and B. A -> B -> C Referent -> Symbol For example we receive a photon at our light sensor in the lab, that's a change in state. Was it caused? If so then that event is a symbol for some prior referent. Well of course it was caused, it didn't just happen out of nowhere right? If we follow that photon back we come to the sun where two hydrogen atoms fused to form helium and a whole mess of photons one of which came to us. So there is your referent, the prior event that was cause of the photon that hit our detector. Simple? No. Just because we take for granted that every change in state is preceeded by a prior change in state elsewhere, doesn't mean it is true in any absolute sense. Just because everyone knows that all effects are caused, again doesn't make it true. It may be an assumed given, but a given is not a proven truth. It is not scientific to take something that 'everyone knows' and postulate it as truth just because everyone knows it. We may grant it as a self evident truth, but one doubts that anything could be self evident to a machine. Therefore machines have to held to a higher standard of certainty than humans do lest they assume some very odd things. So we have to ask the following question. Can we build a machine, that can prove to its OWN satisfaction, and to ours, that EVERY EFFECT IS CAUSED? That every change in state here, is preceeded by a prior change in state there? If such a machine can be built, then that machine can attain certainty of the assumption that every effect is caused, and thus along with the certain observation that it has received an effect, the machine can conclude with further certainty that there must have been a prior cause of that effect. It might even with certainty then be able to tell you about the nature of that cause deduced from the nature of the effect it received. But all this depends on being able to prove with perfect certainty that all effects are in fact caused. Now listen very closely. The only way a machine can learn about ANYTHING is by looking at effects! Namely effects in itself. So is there any way to prove that all effects are caused merely by looking at the effects which are caused? Does effect prove cause? Does change in state here, prove prior change in state there? Maybe our poor machine can't prove that ALL effects are caused, because God knows it might have to experience all possible effects to know that. So how about something less demanding, Can a machine be built that can prove that THIS effect is caused merely by looking at THIS effect? Now this is a very deep problem in mechanics and may not come easy. The answer is no. Effects in and of them selves do not contain the information that there was a cause. There is NOTHING that can be observed about any change in state that proves or even indicates that there was a prior change in state in another object elsewhere. When a photon comes in and hits our machine, all our machine can do is change state, all the machine 'knows' is its new state. At that moment, while basking in its new state, the photon is already gone, its energy having been absorbed into the machine. THERE IS NO EVIDENCE LEFT THAT THE PHOTON HIT IN THE FIRST PLACE LET ALONE THAT IT WAS GENERATED BY SOME OTHER EVENT ELSEWHERE. We only think about cause because we know about dependable followingness between two different events. We throw the light switch and the photons hit our sensor. We can't actually see the cause between the two events but we like to anthropomorphize existence in human terms, and we know for sure we are cause of some things, so we like to think there is cause out there also. That's fine, there very well may be, but WE CAN'T SEE IT DIRECTLY. DEPENDABLE FOLLOWINGNESS IS SUFFICIENT TO WITNESS DEPENDABLE FOLLOWINGNESS BUT IS NOT SUFFICIENT TO WITNESS CAUSATION. CAUSATION IS NOT SUFFICIENT TO WITNESS CAUSATION, because all we can observe are the two events that follow each other dependably. LOOKING AT EFFECTS IS NOT SUFFICIENT TO WITNESS CAUSE. LOOKING AT TWO EVENTS IS NOT SUFFICIENT TO SEE THE CAUSE BETWEEN THEM. If there is cause between two different events the only way we can be alerted to this is by observing the dependable followingness between the two events. And yes a machine can do this also. This gives reason to SUSPECT the possibility of causation between them, but not absolute certainty. Every time I throw the light switch, the light comes on and the photons hit my light sensor. Does the light come on because I throw the switch? Well if there is no cause between switch and light coming on, how come they follow each other every time, huh? Ok that's a fair question, but let's reverse it around. Can we be sure that just because two events follow each other all the time that there is cause between them? Clearly the answer is no. It could be coincidental. It could be independent synchronicity. It could be a third party is affecting both to make it look like there is cause between them. Maybe someone is watching me throw the light switch, and when I do, he throws his own switch which then causes the light to come on. Any computer game is an example of this, the computer is the third party making it look like the bullets on the TV screen are blowing up the bad guys on the same screen. The little light images on the screen certainly are not doing anything to the other light images on the screen, right? The third party is 'virtualizing' cause on the screen so we can believe cause is between the objects on the screen when really its between the computer and each object individually. These may seem like silly arguments, surely there isn't any third party making the physical universe behave AS IF there was real cause inside it. Hydrogen becomes helium, photon is released, that's the end of it. OK, fine, that is a perfectly good philosophical stand to take, but it is not PROOF, namely perfect certainty. Thus we come back to the machine certainty theorem. Can a machine be built that learns only by looking at effects that can prove that those effects are caused? The answer is no. Can it even get the *IDEA* that they are caused, even if it is witnessing endless examples of dependable followingness. The answer is no. All the machine can do is record the fact of dependable followingness between two events, and produce a running record of how dependable that followingness was and is. Cause is not an observable event, so the machine will never come up with the idea. The machine drops the apple 2 million times and it falls down to the ground. Must be cause around here somewhere, right? No, not to the machine, all it knows is that as of now the apple has fallen 2 million out of 2 million times. Can it derive from that fact that the apple will PROBABLY CONTINUE to FALL THE SAME WAY FOREVER. No, it wouldn't even concieve the idea. That idea has to be programmed into the machine by a human as an unprovable assumption. You see it is the human that believes that dependable followingness implies causation, i.e. NECESSARY CONTINUED DEPENDABLE FOLLOWINGNESS. The machine will only roll its eyes at this assumption, because all the machine knows is OBSERVABLE FACTS, and since cause can not be observed directly, cause can never become a fact to the machine. Dependable followingness can be a fact to a machine. But NECESSARY dependable followingness can not. Since the only way a machine can learn about anything is to be the effect of it, and since it can't prove with certainty that any effect is caused, it can not know with certainty anything about the cause, including whether it exists or not. Now include the fact that a machine can't even be certain it has changed state, and thus can't know for sure it has received an effect at all, let alone that it was caused, and we have proven that a machine can't be certain of anything. Because consciousness CAN be certain of some things, including its own existence, change in state, time, and agency (cause!), we can safely conclude that consciousness is not a machine, i.e. not a system of parts, learning about cause by looking at effects, across a space/time distance, So how then? If consciousness can learn about and verify two different colors in its own mockups with perfect certainty, and it can't be doing this by mechanical learning, how then is it kearning? It is learning by looking at CAUSE, not by looking at effects! It is quite possible the mind can only understand mechanical learning, since the mind was designed to understand the physical universe of space and time. Thus although consciousness can observe itself learning with perfect certainty if there are two different colors out there, the mind may not be able to understand the process even while observing it going on! "What? Something I can see but not understand, and KNOW I am seeing it, and KNOW I am not understanding it?" Yes. No one ever promised that God was understandable from the mechanical point of view. That one can see God in operation via perfect certainty is not lessened by the non comprehension of the mind at this time. The non comprehension is yet another perfect certainty. Homer ------------------------------------------------------------------------ Homer Wilson Smith The Paths of Lovers Art Matrix - Lightlink (607) 277-0959 KC2ITF Cross Internet Access, Ithaca NY homer@lightlink.com In the Line of Duty http://www.lightlink.com Mon Jan 2 15:38:38 EST 2006