Most known “systems” decay, rot, age, and die. We usually focus on the death part, but the more fundamental problem is decay (a.k.a. rotting, aging). Death is almost inevitable, as immortality is extremely difficult to achieve. Systems that don’t decay can still die; we sometimes see systems where the chance of death stays constant over time. But for most complex systems, the chance of death rises with time, due to decay.
You are, as usual, stunningly ignorant. "that writer"'s essay is one of the best known and most influential among system developers. And of what import is any failure of persuasion? This cutesy comment is a typical illustration of your stupidity, irrationality, and intellectual dishonesty.
You talk about generalist species not rotting, but I think evolutionists would dislike your phrasing, even while agreeing with the intention of your point.
Generalist species tend to evolve and branch in many directions, with no particular niche being intentional. Most evolve toward specialized niche, as the space of possibilities is largest there. But then over time conditions change, and the niche species tend to die out, leaving the more general forms left. Change prunes the specialists branches from the evolved tree. So perhaps a more congenial to evolutionists way to phrase this is to say that generalist species tend to be selected for over the longest time scales, as niche species tend to be too inflexible against change.
The canonical evolutionary case of this (I've seen this example cited often) is the continual evolution of asexual species. Probably you know this example, but if not, let me dash off a few paragraphs since to me it is a very core concept on why life doesn't root.
First, the "twofold cost of sex" as tactically bad (inflexible but local optimal) but strategically good (generalist enhancing):https://en.wikipedia.org/wi...
Thensee nature paper here on asexual evolution always being at nodes of tree. Make sure to scroll down for the phylogenetic tree chart, which is how I always visualize this idea:https://www.nature.com/arti...
Now to the point. If you want an AGI that doesn't rot, you may need to replicate the idea of sexual reproduction for AGIs in some fashion. Maybe it can be lamarckian evolution, where the AGI self edits itself or it's children to maintain variance (which then can be selected against). Or maybe AGIs in fact do have a version of sex, where the software routines from two (or more) parent AGIs mix and match their core algorithms into a child. Then darwianian evolution could select from the generalist/non-rotting variants out of that mix. That is to say, since AGI forms will rot, we need variation coupled to a darwinian sythe to remove the rotting AGIs from the pool, leaving those lucky few more flexible forms to survive.
For more than a decade David Brin, the sci-fi author, blogged on Blogger. Not too long ago I found him on Medium critiquing capitalism. I use both platforms as well. For me the word "rot" doesn't really describe what happened to Blogger. Really the main issue is that it has stayed pretty much exactly the same. Now it seems especially outdated because we can compare it to Medium.
As a plant hobbyist, the word "rot" has a very specific meaning to me. The plant does not stay the same, it drastically deteriorates. Stems that were firm and green become mushy and brown. Perhaps this might be applicable to some products and firms, but I think most are well-anchored boats. It's the status quo bias... staying the same is the safest bet.
The status quo bias will also ensnare Medium. The problem is really simple... the Visible Hand is at the helm. If it was replaced by the Invisible Hand then the organization would be optimally adaptable. The big decisions would be made by the market instead of by a committee. Committees only made sense when it wasn't practical to make decisions by harnessing the incredible intelligence of the market's collective brain.
Now, thanks to the internet, market decisions are practical. For example last year the Libertarian Party used the market (donors) to choose its convention theme. This is the kind of experiment that all economists should be conducting.
From some reason that I can't fathom you're far more interested in prediction markets than decision markets. It's like you're straining to reach a tiny fruit that's way out of reach instead of simply picking the huge fruit that's right in front you.
Possibly of interest here as a strategy to try to limit "rot": the Urbit project, which is trying to rebuild a non-UNIX inspired computing stack, use strict separation of layers of software, with smaller, more highly verified and stable layers being expected to be very rarely changed. Instead of regular software versioning schemes, they use "Kelvin versioning", where a given layer starts out very "hot", and every time a change is made, the version / temperature gets lowered with 1K.
This means that very low temperature layers get effectively frozen in place, which if done responsibly should imply that they are very highly stable.
I don't know what the best resource is to find out about the project, but this google result seems to describe the basic principle:
Also your argument seems heavily dependent on how one chooses to individuate systems. I mean yes, we often choose to redo parts of systems but we also keep many pieces for a long time. Modern linux operating systems still have utilities with code from early unix systems in them, companies go bankrupt but often pieces of them are purchased by competitors.
Even culture itself is constantly being partially replaced. We decide one generation that old fashion sexual morals need to go and replace that part. Another generation decides to revamp our attitudes about racial equality etc..
It's only at the biological level of individual humans that there is a non-arbitrary way individuate systems and replacement is favored over repair. But that's easily dismissed as a consequence of the evolutionary need to produce offspring in the first place.
Is it really true there are generalist species that are particularly generalist species which retain some greater ability to adapt in the future, i.e., behave as if they had more modular DNA?
I mean of course animals that become highly dependent on a narrow evolutionary niche tend to have less chance to evolve into something else because they tend to go extinct more quickly and it's a longer genetic distance to the next survivable form. But that's not enough to show their genes are rotting. They might be just as capable of evolving as the next animal if they didn't have to make a great leap to find the next survivable form.
So how would we even test such a theory against the alternative that it's almost all a matter of luck? I mean obviously having the luck to produce a whole bunch of evolutionary offspring that occupy different niches increases the chance that one of them will get lucky again.
Maybe it means something like this. If an animal has an evolutionary sibling that gives rise to a successful evolutionary lineage we should expect it to be more likely to do so as well even controlling for advantages common to the two species. That seems like a tricky analysis to do and I don't see how you could possibly know it was true.
Though I don't really think evolution is a good example for your argument because it generally does throw more resources than are close to practical at the problem, e.g., millions and billions of years.
I suspect you're conflating two different things in the word "rot".
Biological aging is the imperfect maintenance of a nominally ideal *static* form (say, what an individual is like at age 25). This kind of rot, modularity can help with.
The "rot" of firms and empires is, at least in part, failure to adapt to a *changing* environment. It's failure to change adequately, not failure to stay the same.
So, not the same thing.
Biological systems don't rot because they sacrifice individual replicators in the service of Darwinian evolution. This is indeed an immense cost - consider that fish produce many thousands of offspring to replace themselves once. Not efficient.
And I'll argue that markets don't rot, for the same reason. Individual traders in markets come and go over time, and this is precisely why markets retain the ability to adapt - as in an ecosystem, those participants who can't adapt well are selected out (to the advantage of the better-adapted).
Many people have pointed out this similarity between economies and ecosystems; I remember reviews of Michael Rothschild's "Bionomics" when it came out (never read it). And many of those reviews pointed out that he'd independently discovered things that Hayek had talked about decades earlier.
Re: "WHY it is easier to make new things than to repair old things" I would say we have a raft of theory about that under the general topics of "theories of senescence" and "advantages of sexual recombination". The short answer is: many reasons. Rejuvenation is technically challenging, expensive and often unnecessary. Old things accumulate parasites adapted to their weaknesses which are challenging to remove. New things often have new genotypes, which helps their lineage to outrun its parasites, in a type of "red queen" race. Sex also concentrates deleterious mutations in individuals that fail to reproduce, taking their deleterious mutations with them. Living for a long time is evolutionarily favored infrequently - but we do have *some* creatures which exhibit "negligible senescence" which are a natural "proving" ground for our theories about the causes of senescence.
Why isnt the answer obvious?There seems to be an extreme bias in how powerful selection is with respect to making copies vs repairing a rotting entity.
A single dividing microorganism doubling every 30 minutes will produce 2^48 offspring in a day, and it will achieve species level avoidance of rot if >1 of those offspring survive, and if it is in relatively deep local maxima of the fitness landscape. Such exponential doubling is, of course, oversimplified, but in my work, I find that yeast cells remain phenotypically stable over 50-100 doublings, beyond which I typically go back to a cryogenically frozen stock. Yeast in the wild are certainly phenotypically stable for much longer; a recent study estimated >75,000 generations (http://dx.doi.org/10.1016/j....The census population of yeast cells today is obviously nowhere near 2^75000.
Compare to the repair strategy, where we can assume that a very simple single celled organism decays to the point of inviability after ~20 doubling times (experimentally determined for a yeast cell- and ignoring for now that that cell is indeed actively repairing itself). If each "repair" attempt has a P probability of failure, then avoiding rot over 100 doubling times has probability (1-P)^5, getting exponentially worse over longer time periods.
I imagine this difference is profound. For making new things, you only need one success out of exponentially many tries; for repairing a rotting thing, it is instead the failures that compound exponentially.
You are, as usual, stunningly ignorant. "that writer"'s essay is one of the best known and most influential among system developers. And of what import is any failure of persuasion? This cutesy comment is a typical illustration of your stupidity, irrationality, and intellectual dishonesty.
Of course ... that's because "technical debt" is a real thing whereas "software rot" or "bit rot" are stupid phrases with no real referent.
Ok, I have to admit to being a little unsure of how to fully cash out your claim then.
You talk about generalist species not rotting, but I think evolutionists would dislike your phrasing, even while agreeing with the intention of your point.
Generalist species tend to evolve and branch in many directions, with no particular niche being intentional. Most evolve toward specialized niche, as the space of possibilities is largest there. But then over time conditions change, and the niche species tend to die out, leaving the more general forms left. Change prunes the specialists branches from the evolved tree. So perhaps a more congenial to evolutionists way to phrase this is to say that generalist species tend to be selected for over the longest time scales, as niche species tend to be too inflexible against change.
The canonical evolutionary case of this (I've seen this example cited often) is the continual evolution of asexual species. Probably you know this example, but if not, let me dash off a few paragraphs since to me it is a very core concept on why life doesn't root.
First, the "twofold cost of sex" as tactically bad (inflexible but local optimal) but strategically good (generalist enhancing):https://en.wikipedia.org/wi...
Thensee nature paper here on asexual evolution always being at nodes of tree. Make sure to scroll down for the phylogenetic tree chart, which is how I always visualize this idea:https://www.nature.com/arti...
Now to the point. If you want an AGI that doesn't rot, you may need to replicate the idea of sexual reproduction for AGIs in some fashion. Maybe it can be lamarckian evolution, where the AGI self edits itself or it's children to maintain variance (which then can be selected against). Or maybe AGIs in fact do have a version of sex, where the software routines from two (or more) parent AGIs mix and match their core algorithms into a child. Then darwianian evolution could select from the generalist/non-rotting variants out of that mix. That is to say, since AGI forms will rot, we need variation coupled to a darwinian sythe to remove the rotting AGIs from the pool, leaving those lucky few more flexible forms to survive.
For more than a decade David Brin, the sci-fi author, blogged on Blogger. Not too long ago I found him on Medium critiquing capitalism. I use both platforms as well. For me the word "rot" doesn't really describe what happened to Blogger. Really the main issue is that it has stayed pretty much exactly the same. Now it seems especially outdated because we can compare it to Medium.
As a plant hobbyist, the word "rot" has a very specific meaning to me. The plant does not stay the same, it drastically deteriorates. Stems that were firm and green become mushy and brown. Perhaps this might be applicable to some products and firms, but I think most are well-anchored boats. It's the status quo bias... staying the same is the safest bet.
The status quo bias will also ensnare Medium. The problem is really simple... the Visible Hand is at the helm. If it was replaced by the Invisible Hand then the organization would be optimally adaptable. The big decisions would be made by the market instead of by a committee. Committees only made sense when it wasn't practical to make decisions by harnessing the incredible intelligence of the market's collective brain.
Now, thanks to the internet, market decisions are practical. For example last year the Libertarian Party used the market (donors) to choose its convention theme. This is the kind of experiment that all economists should be conducting.
From some reason that I can't fathom you're far more interested in prediction markets than decision markets. It's like you're straining to reach a tiny fruit that's way out of reach instead of simply picking the huge fruit that's right in front you.
Possibly of interest here as a strategy to try to limit "rot": the Urbit project, which is trying to rebuild a non-UNIX inspired computing stack, use strict separation of layers of software, with smaller, more highly verified and stable layers being expected to be very rarely changed. Instead of regular software versioning schemes, they use "Kelvin versioning", where a given layer starts out very "hot", and every time a change is made, the version / temperature gets lowered with 1K.
This means that very low temperature layers get effectively frozen in place, which if done responsibly should imply that they are very highly stable.
I don't know what the best resource is to find out about the project, but this google result seems to describe the basic principle:
http://www.anthonyarroyodot...
Not quite the same concept.
I didn't claim that specialist species are rotting; I claimed that at least some generalist ones are not.
Also your argument seems heavily dependent on how one chooses to individuate systems. I mean yes, we often choose to redo parts of systems but we also keep many pieces for a long time. Modern linux operating systems still have utilities with code from early unix systems in them, companies go bankrupt but often pieces of them are purchased by competitors.
Even culture itself is constantly being partially replaced. We decide one generation that old fashion sexual morals need to go and replace that part. Another generation decides to revamp our attitudes about racial equality etc..
It's only at the biological level of individual humans that there is a non-arbitrary way individuate systems and replacement is favored over repair. But that's easily dismissed as a consequence of the evolutionary need to produce offspring in the first place.
Is it really true there are generalist species that are particularly generalist species which retain some greater ability to adapt in the future, i.e., behave as if they had more modular DNA?
I mean of course animals that become highly dependent on a narrow evolutionary niche tend to have less chance to evolve into something else because they tend to go extinct more quickly and it's a longer genetic distance to the next survivable form. But that's not enough to show their genes are rotting. They might be just as capable of evolving as the next animal if they didn't have to make a great leap to find the next survivable form.
So how would we even test such a theory against the alternative that it's almost all a matter of luck? I mean obviously having the luck to produce a whole bunch of evolutionary offspring that occupy different niches increases the chance that one of them will get lucky again.
Maybe it means something like this. If an animal has an evolutionary sibling that gives rise to a successful evolutionary lineage we should expect it to be more likely to do so as well even controlling for advantages common to the two species. That seems like a tricky analysis to do and I don't see how you could possibly know it was true.
Though I don't really think evolution is a good example for your argument because it generally does throw more resources than are close to practical at the problem, e.g., millions and billions of years.
FWIW, "tech debt" was about 5.7 times more popular than "software rot" in my Google duel.
No, there really is a third kind of "rot", which is to become fragile due to repeated attempts to adapt to changing circumstances.
Is there a more common phrase that means "software rot"?
I suspect you're conflating two different things in the word "rot".
Biological aging is the imperfect maintenance of a nominally ideal *static* form (say, what an individual is like at age 25). This kind of rot, modularity can help with.
The "rot" of firms and empires is, at least in part, failure to adapt to a *changing* environment. It's failure to change adequately, not failure to stay the same.
So, not the same thing.
Biological systems don't rot because they sacrifice individual replicators in the service of Darwinian evolution. This is indeed an immense cost - consider that fish produce many thousands of offspring to replace themselves once. Not efficient.
And I'll argue that markets don't rot, for the same reason. Individual traders in markets come and go over time, and this is precisely why markets retain the ability to adapt - as in an ecosystem, those participants who can't adapt well are selected out (to the advantage of the better-adapted).
Many people have pointed out this similarity between economies and ecosystems; I remember reviews of Michael Rothschild's "Bionomics" when it came out (never read it). And many of those reviews pointed out that he'd independently discovered things that Hayek had talked about decades earlier.
Re: "WHY it is easier to make new things than to repair old things" I would say we have a raft of theory about that under the general topics of "theories of senescence" and "advantages of sexual recombination". The short answer is: many reasons. Rejuvenation is technically challenging, expensive and often unnecessary. Old things accumulate parasites adapted to their weaknesses which are challenging to remove. New things often have new genotypes, which helps their lineage to outrun its parasites, in a type of "red queen" race. Sex also concentrates deleterious mutations in individuals that fail to reproduce, taking their deleterious mutations with them. Living for a long time is evolutionarily favored infrequently - but we do have *some* creatures which exhibit "negligible senescence" which are a natural "proving" ground for our theories about the causes of senescence.
Why isnt the answer obvious?There seems to be an extreme bias in how powerful selection is with respect to making copies vs repairing a rotting entity.
A single dividing microorganism doubling every 30 minutes will produce 2^48 offspring in a day, and it will achieve species level avoidance of rot if >1 of those offspring survive, and if it is in relatively deep local maxima of the fitness landscape. Such exponential doubling is, of course, oversimplified, but in my work, I find that yeast cells remain phenotypically stable over 50-100 doublings, beyond which I typically go back to a cryogenically frozen stock. Yeast in the wild are certainly phenotypically stable for much longer; a recent study estimated >75,000 generations (http://dx.doi.org/10.1016/j....The census population of yeast cells today is obviously nowhere near 2^75000.
Compare to the repair strategy, where we can assume that a very simple single celled organism decays to the point of inviability after ~20 doubling times (experimentally determined for a yeast cell- and ignoring for now that that cell is indeed actively repairing itself). If each "repair" attempt has a P probability of failure, then avoiding rot over 100 doubling times has probability (1-P)^5, getting exponentially worse over longer time periods.
I imagine this difference is profound. For making new things, you only need one success out of exponentially many tries; for repairing a rotting thing, it is instead the failures that compound exponentially.