Re: Clean Object Class Design -- What is it?

>www.media.mit.edu/people/minsky/papers/Frames/frames.html
>
>
>>If you are looking for a formal model of
>frame systems perhaps someone else can provide a link.
>
>Well, that is a priceless remark—seeing as many ideas in theat paper
>have never been formalized at all, e.g., the basic idea of
>'frame-systems' (renamed 'frame-arrays" in Society of Mind). As a
>mathematician, I like to say that "Mathematics should be on tap, not
>on top"—because any particular formalization is likely to miss other
>important points. So most people think of frames as nothing more
>than property-lists, and miss the idea of introducing inheritance by
>re-assigning the values of default-assignments. Also, I think this
>may be the first paper that complains that conventional logical
>formalisms are too "monotonic."

I skimmed the article and bookmarked it for later. I hope you can forgive me for not fully digesting every detail right now. Please correct any inaccuracies I present below:

It seems to me that your paper addresses the issue of representing knowledge in a way similar to the human mind -- both for the purpose of better understanding the human mind and for the purpose of artificially reproducing the feats of the human mind. The human mind does not deal with fact and memory in a deterministic and entirely reproducible manner, and I don't think you impose any restrictions that a frame-based memory should be deterministic and entirely reproducible either.

For instance, the paper recognizes the need to limit what moves from short-term to long-term memory.

Is it not possible, then, that a frame-based memory could exhibit artifacts similar to those described in the article at http://schwab-news.excite.com/news/abc/010627/12/bugs-bunny-study ?

Humans have long recognized the need to augment their memory with additional (deterministic) technologies such as writing. The databases we have been discussing until now in this thread certainly fall into the same category of technology. As such, I have treated determinism, reproducibility, accuracy etc. as axiomatic goals. If we choose fundamentally different goals, I expect to arrive at fundamentally different conclusions.

That said: I saw nothing in the paper that would preclude the use of a relational database as the basis of "long-term memory". It is only a matter of how one chooses to represent a frame, a framework, a node, a relation (in the sense used in your paper) etc.

Even if one rejects the use of a relational database as the basis of "long-term memory", I suspect that artificially intelligent systems will have to resort to other (deterministic) technologies for recording certain facts. I think an RDBMS would fill a vital role in this aspect just as ink and paper would.

(It is fun to think of the chaos the legal profession would create if computer memories were as reliable as human memories. Isn't it?)

I notice that you published the paper in 1974 and that you only "pretended" to have a theory at that time. How has the theory held up over time?

>> > >> Unfortunately, they made some huge fundamental mistakes in their
implementation of inheritance. Inheritance applies to domains and not to relations.
>
>The above "framework" paper also contains Scott Fahlman's original
>ideas about inheritance and his early discussions about the problems
>of deciding when (for example) an object should remember the history
>of its inherited properties, etc.
>
>==
>
>Umm, on second thought, I take most of that back. Steve Wart is
>right: if I hadn't been so imprecise, I might have forseen other good
>ideas about both OO and database-backed programming. So one always
>should try to think both formally and informally—and not let either
>approach get out of control.

A set is a set. What does it inherit? And from what? What does a set of propositions inherit from anything else? What are the "kinds" of propositions among which we can inherit properties?

We can have statements of fact about different things, and those different things can inherit properties from other things. For instance, your paper mentions a statue of a cow. The statements: "A statue of a cow has the shape of a cow", and "A statue of a cow moves like a statue" are both statements of fact. What do they inherit from anything else? Aren't the different kinds of inheritance merely statements of fact or propositions? Can we not represent them as tuples?

Certainly, inheritance in the OO sense applies to domains as object classes where one object class can inherit properties, representation and/or behavior from other object classes. The relational data model allows one to establish further inheritance relationships among things (ie. beyond the scope of inheritance as understood by the OO crowd) using sets of propositions (ie. fact statements).

What benefit does one gain by adding another object to the logical data model specifically to describe inheritance? Received on Sat Aug 04 2001 - 19:37:17 CEST