[HS]

With similes, one thing is compared to another to generate a flexible optimization concept. Like animal inspired martial arts. This could be a good default thought process for bots. Mimicry is how we do a lot of our learning anyways, mirror neurons have proved quite an advancement in our evolution. This way we can set goals which don't have to be meticulously worded like wishes for a devious genie. What'd ya think, this could be a good format for all sorts of processes within a flexible intelligence, not necessarily a super intelligence, but definitely a broad and adaptable one.

[LOCKSUIT]

Remember I said in my big movie that text/vision is universal language/data. So your word2vec similarity net web can store relational connections to cat to image of cat to sound of cat to the shape of its ear. Humans collect data so they can develop new data in the pink brain and make/adjust relational connections. Some of that data can be the answer directly, and hence the speedy learning. If you want a Fast Learning Rate without the answers directly, you collect Big Diverse Data to cover all grounds of domains, all words explain each other - small world network.

[goaty]

Meticulous worded rammifications for the robots search seems unavoidable to me,  and its actually not a problem to me either,   I don't mind the robot doing exactly what I say, I don't think itll be that difficult as a form of programming. (not that ive done much of it yet. ), making the robot to begin with seems like the hard bit to me.

If its a very powerful search,  then its more important to make sure you lock it into the task properly,  especially if it was some kind of security droid,  it has to make sure it incapacitates without harm,  but im pretty sure that could be under 10 lines,  its pretty lazy business im imagining once you've got the rest of it.

But im not sure what you mean by mimicry helping this situation,  but definitely animals move beautifully and they are worth noting at least if you are some kind of acrobat.  id guess.