Making Minds: Some Notes

Here are some notes related to my "Making minds" series. I use the phrase "a mind" throughout rather than "a brain", because I view a brain as just one kind of mind. For example, I believe that companies such as NASA also have a mind.

1. A mind contains representations of many things, at varying degrees of detail. For example, if you're looking at a cat, your mind will contain a highly detailed representation of the cat because your senses are supplying your mind with a lot of details. If you're imagining a fish, then it will be less detailed than the cat (unless it's parts of a lucid dream).

2. Some of the representations will include a temporal component. For example, you could imagine how your cat looked when it was a kitten, or imagine what your cat will look like when it's old.

3. Your mind can associate a value with some representations. For example, you might find a kitten to be cute and a cat in pain to be very distressful.

4. The emotion associated with a representation increases with its level of detail. If you see an animal in pain in real life, it will usually cause more distress than if you simply imagine what an animal in pain would look like. This is why you'll typically look away from a distressing image - in order to decrease the level of detail of its mental representation, and thus the level of emotion that it causes.

5. A mind prefers positive emotions over negative ones, and the stronger, the better. Since a highly detailed representation creates a more intense emotion than a vague one, a mind will want to contain representations that are detailed and positive.

6. A mind experiences a negative emotion when representations are inconsistent, and therefore prefers them to be consistent. For example, if you try and imagine a fish, you'll probably start by creating a basic representation with primary features such as a body, head and fins. However, a fish does not normally exist in a vacuum, so in order to maintain consistency, you start to fill in related details such as the water that surrounds it. This in turn causes you to imagine its fins moving since fish don't normally hang motionless in water.

7. Details about a representation can also involve time. For example, if you see a fish swimming towards you, then additional details can include where the fish was in the past and where the fish will be in the future.

8. A mind uses rules to fill in details. A mind can generate new hypotheses and will keep more copies of the rules that are accurate and less copies of the rules that are inaccurate. I use the 'copy' metaphor instead of the 'strengthen' metaphor because it's more appropriate for parallel distributed systems.

9. Rules can be of many types, including temporal and spatial. For example, a rule could be of the form "if X happens at time T, then Y usually happens at time T+1". Similarly, a rule could be of the form "if X is on the left, then Y it usually on the right". Note that both of these rules are relative; the temporal rule could apply to events in the past, present or future.

10. Rules can create multiple possibilities, some of which might be mutually inconsistent. For example, if we see a fish swimming towards us, there will be many possible future positions for the fish. But of course the fish can't be in all those positions at the same time.

Comments

[David]

Hi, one of the Davids from philosophy club here. (The quieter one.) First thought: What are some predictions you'd make based on these notes? It seems like some of them ought to be testable through informal experiments.

Some small things:
Rather than reducing the detail, it seems like looking away is more an instance of totally avoiding the input. I think we also observe the opposite, eg, people reading up (and becoming upset) about crimes, perhaps particularly relevant here, crimes that happened in places distant in time and place so seemingly not pursued out of any more overriding reason -- not to learn avoid immediate danger, etc.

Why prefer copying over strength as metaphor? I think neural networks are one of the main examples of architecture for and example of parallel processing, and they work in terms of weights, I think.

Well, I'll stop here for now, but I'm happy to have found this and look forward to reading more. I'm particularly curious about your business.

I also thought I would point out this: http://www.lpsf.org/ I believe their next meeting will be on June 12th.

[Graham Glass]

Hi David,

Nice to hear from you!

There are indeed some testable predictions; I will enumerate some of them soon.

Re: reducing the detail. When you turn away, it does indeed remove the sensory input, but there's still plenty of representation of the event that remains in a mind. Sensory input is really only just one way of creating representations in a mind. If you really removed the detail, then you would instantly forget that the event even existed.

Re: reading about crimes. There is indeed a value to reading about the crime. However, the gory details of the crime (a graphic feature of the decapitated victim) are usually not published because they would be truly upsetting. I think that some people get a positive emotion from reading about a crime, especially if it happens to a celebrity, because they think "Hah, they might be rich & famous, but they can still be a victim of a crime". (see any national enquirer for examples of this).

In a distributed system, you don't want just one copy of anything, even if it's strengthened, because it's a single point of failure. If that copy is destroyed (and neurons die all the time), then you would forget that rule. Instead, if you copy things you want to remember, then the most useful things become more fault tolerant as well as stronger (based on the combined "voting" of all the copies).

Thanks for the libertarian party link - I must go to one of those events soon!

Cheers,
Graham