Southwind, I am merely pointing out that the evolutionary algorithm that you have described still has implicit selection criteria. If you tried to impliment such a system without any selection criteria, then the system would not work.
You came very close to falling victim to my first "Watch the Post #1556 video link!" automated response with this jimbob, but given the courteousness of your post, and the seemingly thoughtfullness of at least some parts of it I'll indulge you for the time being. Incidentally, have you watched the video link yet?
The instructions are to build what has sold, and make small changes. I woul dsay that you have an evolutionary algorithm devoted to "making stuff that sells".
To be precise, the instruction is to replicate what was last made and make small, random changes, such instruction being triggered by the receipt of proceeds from sales, so the sale itself could be said to be the trigger, if you prefer. I'm not sure whether that affects your conclusion. I suspect not. Either way, if it is an 'evolutionary algorithm' (I'm not sure what that is, exactly), then to my mind so is the process that governs the reproduction of a cheetah, for example, which is triggered by mating, which itself is triggered by, and only by, the male and female cheetahs' survival of their environment to the age of sexual maturity, which, over the whole species, is wholly determined by the inherent ability of their characteristics and features as compared to competing animals.
How exactly do you determine a failure?
A failure is simply a variant that fails to sell, i.e. it's a disadvantageous mutation.
Suppose, as is likely, the first random variant did not sell.
Without a process or an algorithm to determine that this is unsucessfull, the system would remain waiting.
That's right, the system would stop. But don't forget that you're only talking about an isolated variant to one production run of the device. In reality, as with nature, the AA should envisage an 'army' of automatons, all replicating and making different random variants. In a sense all the slightly different devices would be competing with each other, in addition to all the other different devices produced by competing automatons in the marketplace.
If you add an algorithm to determine the fialure criteria, then you have added the required intelligence.
As explained above, the algorithm, if that's what it is, determines the success criteria, not the failure criteria, but it's no different from the 'natural' algorithm. If there's no such thing as a 'natural' algorithm then I deny that an algorithm exists in the AA.
Of course there could be an interesting case, when an imperfect copy actually has randomly acquired the ability to self-replicate. This part woould then, without any further input make a further (imperfect) copy of itself, and its descendents would, from this point actually evolve.
I still maintain that
self-replication is a complete red herring.
Rather than talking about hypothetical cases, I think it would be better to talk about real case studies where evolutionary algorithms have been used. They have been used to produce circuits that human designers couldn't, and which are currently very hard (impossible with present modelling techniques) to model, and in only about 100 generations.
If the introduction of an algorithm is tantamount to not accepting the AA as completely valid then I'm not prepared to compromise, not until somebody proves to me that the AA has failed its purpose.
Your story would also work for this, but it is probably better to have real examples.
See above.
This shows the power of evolutionary algorithms when harnessed to an intelligently defined goal. One can then point out that organisms show "imperfect copying", and that those which manage teo reproduce are obvioulsy "fit enough" to reproduce. This is just an evolutionary algorithm where the selection criterion is the ability to reproduce. Unsurprisingly, this results in good reproducers.
It sounds like this falls way short of achieving what I hoped to set out to achieve with the OP, in which case I'm not interested. No offence jimbob.
