Implications of mind evolution vs idea evolution
Agents have the advantage over humans that they are able to employ evolution to improve their minds and not just their ideas. This is relevant to whether the impact of agi will be larger than brain count increase suggests.
Why would the ability for evolution to occur at the level of minds have advantages over the evolution at the level of ideas, as is already possible within and between human brains?
It seems that all of the following factors contribute to effective evolution:
- ease with which improved strains can be identified
- ease with which improvements can be replicated
- accuracy of improvements replication
- efficiency of variation, i.e. how targeted variations are towards the goal
-
- ease with which improved strains can be identified
- Ease of identification of idea
- Useful ideas are often hard for their user to identify or even realise the existence of, whereas effective minds are comparatively easy to identify due to being distinct pieces of software.
- Ease of running controlled experiments - it is easier to run a controlled experiment to determine the usefulness of a mind than it is to do so with an idea
- this is because a mind can be easily compared to another version of itself or another mind
- whereas the same can not be said for ideas, as it is not possible for everything about two (human) minds to be the same with the exception of one idea.
- because this would require duplicating a human mind, which is not possible.
- Ease of evaluation
- whereas the same can not be said for ideas, as it is not possible for everything about two (human) minds to be the same with the exception of one idea.
- this is because a mind can be easily compared to another version of itself or another mind
- It is much easier to evaluate a mind than an idea, unless it is possible to have two minds that are identical w/ exception of the single idea in question (so same issue as mentioned above)
- ease with which improved strains can be identified
-
- ease with which improvements can be replicated
- cost of replication relative to number of recipients
- idea replication often requires separate contextualisation/discussion for each mind the idea is replicated to, although can be more efficienct once economies of scale are reached (e.g. via books/forums etc.)
- mind replication is lossless and requires only non-interactive transfer of bits
- cost of replication relative to complexity of idea - the more complex an idea/dissimilar from existing ideas, the more it costs people to learn it (in terms of time + effort) and hence the higher the transmission costs are
- mind replication cost scales linearly with size of mind, but is very low (bandwidth is cheap) and the similarity of an idea relative to existing ideas doesn’t necessarily imply that for a mind to understand it, a mind must be larger
- ease with which improvements can be replicated
-
- accuracy of improvements replication
- lossyness of replication
- Minds can be replicated with zero errors
- idea replication is always lossy
- accuracy of improvements replication
-
- efficiency of variation, i.e. how targeted variations are towards a goal
- it seems that similar strategies can be used by both minds and ideas to generate variations, given that variations in a mind are the same as variations of the ideas within the mind
- However due to more efficienct replication and selection (points 1) and 2) above)
- variation of minds should be more targeted than variations of ideas, due to better minds being much more likely to proliferate, and due to better minds having better strategies for varying their ideas.
- efficiency of variation, i.e. how targeted variations are towards a goal
If you found this interesting, have feedback or are working on something related, let’s chat: twitter (@0xdist) or schedule a 20 min call