Consider three design problems:
- Case #1: First, you are asked to modify a stock car, making it into a truck to haul stuff. After you do that, you are asked to create a race car. Which would you rather start from for this second task, another stock car, or the stock car that you turned into a truck?
- Case #2: A species of beetle lives in a varied and changing environment, and so has a rather simple and basic design. Some of these beetles invade a different and more stable environment, and acquire adaptations specific to that environment. A third rather different but also stable environment opens up adjacent to both previous environments. Which beetle type’s descendants will likely fill this third environment?
- Case #3: Over the last decade a group wrote software to do a certain task (e.g., print driver, web server, spreadsheet, etc.) This design of this software was matched to certain features of the problem environment, such as hardware, network speeds, etc. Today there is a need for software to do a similar task, except that the problem environment has changed. To write this new software, would you have your team modify this previous software, or start a new system mostly from scratch?
In all these cases, one makes a system to function in a given environment, and can either modify a complex system adapted to a different environment, or “start over” via modifying a simpler system less adapted to any specific environment. In general, the more different is the new environment from the old, the better it is to start over. Old systems tend to be rigid, which makes them fragile, in that they break if you bend them too far.
This suggests that designed systems tend to get irreversibly fragile as they adapt to specific environments. When context changes greatly, it is usually easier to build new systems from “scratch,” than to un-adapt systems designed for other contexts. Software tends to “rot“, for example.
An empirical prediction here is that species occupying highly variable environments tend to have more descendant species in other environments, compared to species occupying less variable environments. I don’t know if this has been tested. It fits with the Innovator’s Dilemma though, where firms who serve the low end of a product line with simpler techs tend to creep up and displace those serving the high end; high end products tend to be more complex.
Today I’m focused on this being bad news for the feasibility of immortality, at least for human-like creatures. You see, our minds seem designed to adapt to the environment in which we grow up, via youthful plasticity transitioning to elderly rigidity. For example, we are great at learning languages when young, and terrible when old. We are similarly receptive when young to new ways to categorize and conceive of things, but once we have often used particular ways, we find it harder to understand and use alternatives.
The brains of most animals peak in functionality during their key reproductive years, and do worse both before and after. Short lived animals peak sooner than long lived animals. Some of the early rise is due to learning, and some of later decline is due to the decline of individual cells and connections. Some of this pattern may even be due to an explicit plan to turn up some dials on plasticity early on, and then turn down those dials later. But I think another important part of this rise and fall is due to a general robust tendency for adapted systems to slide from plasticity to rigidity.
Thus even if we succeed in creating emulations of whole human brains, “ems” which can use backups, body swaps, etc. to avoid bodily death and decay, we should expect such ems to decay by getting mentally rigid with subjective age. Even if we do not emulate any decline in individual cell and connection performance, nor any age-specific general plasticity dial settings, the mind itself may well decay with subjective experience, because such decay is just intrinsic to mind design.
Now in software design one can often slow a slide to rigidity by refactoring code, such as by looking for better abstractions to achieve modularity. But the brain probably already has some analogues to refractoring, such as in its ways to reorganize concepts. And even with large refactoring efforts, most designed software eventually gets rigid, so that when environments change enough such software is replaced wholesale by new systems built from scratch.
Similarly, em workers who start out subjectively young, and then learn how to work in a stable environment, may become increasingly productive in that environment, even after thousands of years of subjective experience. But when a new quite different work environment appears, one can probably gain more work productivity by training subjectively young ems for it, rather than trying to change ems who had spend thousands of subjective years adapting to a very different environment.
Today most houses and cars are in principle immortal, in the sense that enough maintenance can keep them functioning indefinitely. Yet most houses and cars are not immortal in practice, because those maintenance costs keep rising to the point where it is cheaper to build new houses and cars. Similarly it might be possible to keep very old ems around, even when they have become much less productive because relevant environments have changed. Someone, however, would have to pay that cost, relative to the option of using more productive younger ems. And as with houses and cars today, maybe few will pay.
If you personally hope to become an em with an especially long productive subjective life, it is probably important to stay general and flexible for as long as you can. Prefer to acquire habits and insights that are widely applicable, and whose value is likely to long continue. Prefer to write, deal with people, and manage complexity, rather than learning the detailed layout of a city or how best to write in a particular new programming language.
Eventually we may find mind designs with a much weaker tendency toward rigidity with age. And we may find ways to transfer some important elements of once-human minds, such as their memory and personality, into this alternative framework. But even then there should be some aging. And it gets even less clear if you’d want to think of such a changed creature as you.
Even more eventually, the universe should get a lot more stable, and with it the environments where minds function. Then there will be a lot more scope for very long lived human-like minds. If there are any human-like minds left at that point.
Added: Stem cells fit this; bodies usually make cells designed for specific places from general simpler stem cells, not by changing other specific cells.