How much do we know about reasoning about subjective concepts? Bayes’ law tells you how probable you should consider any given black-and-white no-room-for-interpretation statement, but it doesn’t tell you when you should come up with a new subjective concept, nor (I think) what to do once you’ve got one.
You may be interested in the literature on “concept learning”, a topic in computational cognitive science. Researchers in this field have sought to formalize the notion of a concept, and to develop methods for learning these concepts from data. (The concepts learned will depend on which specific data the agent encounters, and so this captures the some of the subjectivity you are looking for.)
In this literature, concepts are usually treated as probability distributions over objects in the world. If you google “concept learning” you should find some stuff.
Well, I guess that by “subjective concepts”, I mean every concept that doesn’t have a formal mathematical definition. So stuff like “simple”, “similar”, “beautiful”, “alive”, “dead”, “feline”, and so on through the entire dictionary.
The only theory-of-subjective-concepts I’ve come across is the example of bleggs and rubes. Suppose that, among a class of objects, five binary variables are strongly correlated with each other; then it is useful to postulate a latent variable stating which of two types the object is. This latent variable is the “subjective concept” in this case.
Think of subjective concepts as heuristics that help you describe models of the world. Evaluate those models based on their predictions. (Grounding everything in terms of predictions is a great way to keep your thinking focused. Otherwise it’s too easy to go on and on about beauty or whatever without ever saying anything that actually controls your anticipations.)
How much do we know about reasoning about subjective concepts? Bayes’ law tells you how probable you should consider any given black-and-white no-room-for-interpretation statement, but it doesn’t tell you when you should come up with a new subjective concept, nor (I think) what to do once you’ve got one.
You may be interested in the literature on “concept learning”, a topic in computational cognitive science. Researchers in this field have sought to formalize the notion of a concept, and to develop methods for learning these concepts from data. (The concepts learned will depend on which specific data the agent encounters, and so this captures the some of the subjectivity you are looking for.)
In this literature, concepts are usually treated as probability distributions over objects in the world. If you google “concept learning” you should find some stuff.
“Subjective” seems uselessly broad. Can you give a more specific example?
Well, I guess that by “subjective concepts”, I mean every concept that doesn’t have a formal mathematical definition. So stuff like “simple”, “similar”, “beautiful”, “alive”, “dead”, “feline”, and so on through the entire dictionary.
The only theory-of-subjective-concepts I’ve come across is the example of bleggs and rubes. Suppose that, among a class of objects, five binary variables are strongly correlated with each other; then it is useful to postulate a latent variable stating which of two types the object is. This latent variable is the “subjective concept” in this case.
Think of subjective concepts as heuristics that help you describe models of the world. Evaluate those models based on their predictions. (Grounding everything in terms of predictions is a great way to keep your thinking focused. Otherwise it’s too easy to go on and on about beauty or whatever without ever saying anything that actually controls your anticipations.)
Have you read the rest of 37 Ways That Words Can Be Wrong?