
A few years ago (eleven?) I became interested in computer based modeling of behavior, and eventually started playing around with a bio-plausible “selectionist neural network”.
In the process of creating the BGL2015 Visualizer demo, I realized that such a simulation could be recorded, and/or could be stopped and started, and/or could be “branched”, at any point. That would be very useful for studying in detail the processes involved in “learning”, or more generally, involved in the processes that alter responses to environmental events.
But the “branch” possibility is particularly fascinating. One of the most powerful research methods in behavior analysis is using an experimental subject as their own control. We refer to research paradigms such as “ABAC” designs in which the subject is exposed to one set of contingencies in the “A” condition, and other environmental arrangements sequentially in the “B” and “C” conditions. It is exactly the same organism, so the resulting behavior changes cannot be ascribed to individual differences.
Except that they can, because life is the process of changes in the organism. When you learn something, or when your behavior changes, there are physical changes within your body. You gradually become a different person. But, still, the you of 15 minutes ago and the you of now are more alike than you compared to any other person at any time.
And of course, some behavior changes (and therefore organismic changes) are larger than others. Especially over long time spans. Which leads to the issue of non-reversability. There are research/diagnostic designs that account for this. But suppose you could pick a point in a person’s life and ask, “Suppose we imposed a different set of conditions here? What would have been different?”
In a simulation, you have unlimited opportunity to do so (given the correct software architecture). If you were a simulation, the researcher could go back to when you were some fraction of your current age, tweak some little environmental factor, click the “run” button. You, as a simulation would have no recall of your “previous life” after the branch point, but would live a new life from that point.
A more interesting scenario is where the subject is, say, 90 years old, and you want to test out the age old question regarding, “If I knew then what I know now.” Let’s take the relevant internal state (say, your neural net) of the 90 year old you, and pair it up with an otherwise 19 year old version of you back in the same environment in which you were previously 19. After all, this is a simulation and we have recorded or can otherwise accurately reproduce anything and everything. And we still have the recording of exactly what you did in the “previous life” as a comparison. Any number of previous lives, in fact.
The 19 year old “you” with the 90 year old neural net would be interacting with the previous environment of the 19 year old “you” (remember back when you were 19?). Your “neural net” would be generating responses very different from when your 19 year old body had your 19 year old neural net.
But in addition, some of the responses might induce some sense of deja vu. Might be weird memories of an earlier time when you were 90 years old those 71 years in the future that have not yet occurred.
Experiences that might lead us to talk about being reincarnated.
Suppose the “this is all a simulation” folks are correct? Suppose that the “simulation hypothesis” is correct?