Language variants for agents

[kach]

I'd like the ability to create agents who have restricted subsets of the language. For example, I would like to be able to say that Alice thinks Bob can't use the "thinks" statement.

[iancero]

I like the idea that this could lead to some very readable code, depending on the setup. For example...

alice: thinks[
    bob: lacks(imagine)
]

... reads almost exactly as "Alice thinks that Bob lacks imagination".

You could also imagine being able to restrictions on an agent's recursion depth, as a way of modeling some features of level-k reasoning. To be fair, the language already has some ways of doing this with parameters, but it's interesting to explore the possibility of doing it in the actual syntax.

[iancero]

Or!

For modeling psychosis (in which people often have the mistaken sense that others are both reading their minds and inserting thoughts in their minds) you could imagine language variants that are supersets of memo:

# mind reading
alice: thinks[
    bob: knows[
        alice: thinks[
            ...        
        ]
]

# thought insertion
alice: thinks[
    bob: inserts[
        alice: thinks[
            ...  
        ]
    ]
]

Again, I think some of these concepts are already expressible with the current verbs, but it is interesting to think that modification of the syntax of memo is equivalent to both (a) developing minds of young children who have TOO FEW ToM constructs and people with thought disorders who maybe utilize TOO MANY ToM constructs.

From an engineering perspective, I'm intrigued by the possibility that modifying "knows" in this way, where Alice thinks Bob always has constant up-to-date knowledge of her internal models might be expressible with a continuation, similar to how wants is implemented now.