Approaches to building artificial general intelligence have so far been unsuccessful because all of them fail to identify how they would creatively select goal functions. How does an AGI identify which problems to solve? That’s the question. This is why David Deutsch has been saying we are all waiting for a philosophical innovation in AI, and that once that philosophical solution has been found, the technical part of building AGI probably will not be that difficult. I took him seriously, and went on a search for such an idea.
The Darwinian theory of life has the same problem. Both mysteries hinge on explaining purposive action. There is an assumption that evolution began with bare replicators that later developed the ability to stay organized and respond sensitively to their environment. But this is only an assumption; it’s not based on anything other than our inability to point to exactly how this happens, and our discomfort with admitting that our theory requires another major component to be complete. So we have construed the theory in a way that waves away and diminishes that component. We have no examples of these bare replicators now and the explanation is that they were outcompeted and disappeared. But that’s just a guess.
Evolution as we now know it can only begin once a set of molecules is trying to stay organized. There is no reason to insist that replication had to come first, and self-organization second. It could just as well be that self-organization came first. Whichever came first it is obvious that life now requires self-organization, and that is also required for the only AGI we can all agree on existing now, which is human. Something in every reproducing cell carries the heritable templates to the next generation, but something also carries this self-organizing process to the next generation as well.
We can say now that it is relatively certain that there is a static thing in all organisms that carries heritable templates: DNA. But in every cell we know of, there is also a dynamic process that consists of oscillators (repeating chemical reactions) that is changed by its environment and that change has something to do with use and disuse. This dynamic process is a critical part of every life form. And every life form also inherits an initial cell with this process already intact and working.
My guess with epistolution is that these oscillators are that critical missing piece in both explanations. Something Lamarkian happens in all organisms…they heal themselves from damage, and with exercise they grow stronger. That is a critical difference between life and non-life. My guess is that all it takes to set purposive action in motion is that you need a large set of oscillators that respond sensitively to their environment by reinforcing themselves where they are used, and mutating where they are unused. This could explain how organisms can anticipate their environments. Of course these Lamarckian oscillators will be embedded in non-Lamarckian organic structures. Part of an organism is dynamic and self-healing, like neural connections, and part is static and non-healing, like hair or antlers. The test I’m proposing for epistolution is to create the static part out of artificial hardware and let it be a platform for a Lamarckian dynamic set of oscillators built in software which controls their behavior. This would test epistolution by separating it from evolution.
In order for evolution to have occurred, I am conjecturing that there must have been this self-organizing process present all along. In this sense purposive action was primary, and evolution secondary.
Current evolutionary theory avoids this problem by insisting that no purposive motive other than replication exists, but it hasn’t established any test to prove that that is the case. There is an obvious apparent motive in every living being…it tries to stay organized and alive. This motive hasn’t been refuted or explained. Of course this behavior must be consistent with Darwinian evolution; but since Darwinian evolution as we know it requires it to begin, Darwinism does not establish that no motive is really there. Darwinism is just a philosophical point of view that excludes God and includes heredity, and it still requires a self-organizing mechanism if it’s going to to apply to the organisms we see today.
Environmental epigenetics, symbiogenesis, and recent microbiome research makes this crisis worse. Now we have to explain how, without any common evolutionary history, organisms can join together into functional super-organisms. We also have to explain how different lineages have apparently in many cases fused into new forms of organism. We also have to explain how, in some cases, genetic change is not blind, but rather information acquired during this (possibly purposive) living behavior can bleed into future generations by affecting the expression of DNA and the rate and sites of mutation. This makes the non-purposive explanation considerably weaker.
Epistolution forms a much better explanation for these phenomena, because it gives the environment a (non-magical) mechanism to drive behavior in all organisms, which explains why they behave appropriately in a given context, and how this behavior can look and feel like general intelligence especially at higher levels. It is a really simple idea.
I have a slide deck now I’m working on that walks through the exact steps to build a set of oscillators and mutate them based on use/disuse and give them input and output. I now am persuaded that its easiest to start off testing this explanation by putting these oscillators into a virtual environment. The slide deck is intended to be a set of instructions for a software engineer to follow. I would love your feedback as I finish them. If you could critique whether you think these are specific enough for an engineer to follow I would be deeply grateful. And if any of you help me carry them out and actually build this thing I will share the Nobel prize with you one day