This law according to Dennett is an extension of Schank’s Law:
Because people understand by finding in their memories the closest possible match to what they are hearing and use that match as the basis of comprehension, any new idea will be treated as a variant of something the listener has already thought of or heard. Agreement with a new idea means a listener has already had a similar thought and well appreciates that the speaker has recognized his idea. Disagreement means the opposite. Really new ideas are incomprehensible. The good news is that for some people, failure to comprehend is the beginning of understanding. For most, of course, it is the beginning of dismissal.
any new idea will be treated as a variant of something the listener has already thought of or heard.
From a Bayesian point of view, this is as it must be. People have priors and will assess anything new as a diff (of log-odds) from those priors. Even understanding what you are saying, before considering whether to update towards it, is subject to this. You will always be understood as saying whatever interpretation of your words is the least surprising to your audience.
BTW, this is standard in natural language processing (which is what a lot of Schank’s AI work was in). When a sentence is ambiguous, choose the least surprising interpretation, the one containing the least information relative to your current knowledge.
The narrower your audience’s priors, the more of a struggle it will be for them to hear you; the narrower your priors, the more you will struggle to hear them.
Having shown how Schank’s Law is but an instance of Bayesian inference, I trust you will all find it acceptably unsurprising. :)
When shaping behavior in animals, we start with something the animal does naturally and differentially reward natural variations. Evolution of biological systems also involves differential selection of naturally occurring variations on existing systems. So it’s certainly possible to get “something new” out of mere “variants of something [that already existed]”.
That said, many cognitive systems do also seem capable of insight, which seems to be a completely different kind of process. Dennett and Schank here seem to be dismissing the very possibility of insight, though I assume they are doing so rhetorically.
What has a baby which does not understand speech “heard before”, that it can form variations on? Evolution is fine, but you do need a theory of abiogenesis, or in this case aontogenesis—knowledge-from-nothing-ness, in the vernacular.
Babies are not clean slates; there exist innate behaviors. We can get into a theoretical discussion of where these behaviors came from if you like, but I don’t need a theoretical justification to observe that babies do in fact do things they haven’t been taught to do.
While I agree with TheOtherDave’s point, I’m not sure it’s necessary. A baby doesn’t understand new sounds the first time it hears them, but may understand them the hundredth time it’s heard them- at which point it does have quite a bit of experience, both of hearing those noises in some situations and not hearing those noises in other situations. Then, once they’ve learned the general skill of acquiring words, they can correctly learn words quickly, sometimes even after hearing a single use- but that’s drawing on their previous experience in learning thousands of words.
Naturally we go through a period of believing everything we’re told when we’re kids, and transition to comparing everything we hear to what we’ve already heard before as we grow up.
(This is an inexact approximation, but in my more cynical moments it strikes me as only very slightly inexact.)
This does raise the question of how anyone learns anything in the first place.
Perhaps most people learn like this: They already have an idea X. Then they hear a very similar idea Y, so they accept it, although they interpret it as X. But once they agreed that Y is their idea, and they hear it repeatedly, they gradualy become aware of Y as something slightly different from X. Thus they made another inferential step.
Perhaps many people are willing to learn only when it does not feel like learning.
A less cynical take on this is that people compare what they hear to their previous experience (stored in compressed form) and accept or reject it depending on how well it matches.
This law according to Dennett is an extension of Schank’s Law:
-Roger Schank
From a Bayesian point of view, this is as it must be. People have priors and will assess anything new as a diff (of log-odds) from those priors. Even understanding what you are saying, before considering whether to update towards it, is subject to this. You will always be understood as saying whatever interpretation of your words is the least surprising to your audience.
BTW, this is standard in natural language processing (which is what a lot of Schank’s AI work was in). When a sentence is ambiguous, choose the least surprising interpretation, the one containing the least information relative to your current knowledge.
The narrower your audience’s priors, the more of a struggle it will be for them to hear you; the narrower your priors, the more you will struggle to hear them.
Having shown how Schank’s Law is but an instance of Bayesian inference, I trust you will all find it acceptably unsurprising. :)
This does raise the question of how anyone learns anything in the first place. :)
Don’t underestimate the power of variations.
When shaping behavior in animals, we start with something the animal does naturally and differentially reward natural variations. Evolution of biological systems also involves differential selection of naturally occurring variations on existing systems. So it’s certainly possible to get “something new” out of mere “variants of something [that already existed]”.
That said, many cognitive systems do also seem capable of insight, which seems to be a completely different kind of process. Dennett and Schank here seem to be dismissing the very possibility of insight, though I assume they are doing so rhetorically.
What has a baby which does not understand speech “heard before”, that it can form variations on? Evolution is fine, but you do need a theory of abiogenesis, or in this case aontogenesis—knowledge-from-nothing-ness, in the vernacular.
Babies are not clean slates; there exist innate behaviors. We can get into a theoretical discussion of where these behaviors came from if you like, but I don’t need a theoretical justification to observe that babies do in fact do things they haven’t been taught to do.
Quite so, but this contradicts the original idea that everything is variants on something that has been heard before.
I interpret “heard before” to include “programmed in your genetics”.
This.
While I agree with TheOtherDave’s point, I’m not sure it’s necessary. A baby doesn’t understand new sounds the first time it hears them, but may understand them the hundredth time it’s heard them- at which point it does have quite a bit of experience, both of hearing those noises in some situations and not hearing those noises in other situations. Then, once they’ve learned the general skill of acquiring words, they can correctly learn words quickly, sometimes even after hearing a single use- but that’s drawing on their previous experience in learning thousands of words.
Naturally we go through a period of believing everything we’re told when we’re kids, and transition to comparing everything we hear to what we’ve already heard before as we grow up.
(This is an inexact approximation, but in my more cynical moments it strikes me as only very slightly inexact.)
Depends how great the variance is. Sounds better if you say that people benefit from having things they’re learning related to familiar topics.
Perhaps most people learn like this: They already have an idea X. Then they hear a very similar idea Y, so they accept it, although they interpret it as X. But once they agreed that Y is their idea, and they hear it repeatedly, they gradualy become aware of Y as something slightly different from X. Thus they made another inferential step.
Perhaps many people are willing to learn only when it does not feel like learning.
A less cynical take on this is that people compare what they hear to their previous experience (stored in compressed form) and accept or reject it depending on how well it matches.