Greg Detre
Thursday, November 07, 2002
speech
understanding problem
v large search space, errorful/incomplete input data,
imprecise/incomplete problem-solving knowledge
need to be able to develop solutions incrementally, apply diverse types
of knowledge, adapt strategies to particular problem situations
blackboards allow problem solving (search) in which the system
dynamically switches among different levels of abstraction
difficult to effectively utilise the features are the model that are the source of its flexibility os they complicate the process of selecting actions wt he maximum expected value
interesting definition of the value of an action in terms of how much it contributes to progress in problem solving relative to the computational costs of the action � where progress is judged by how much the action reduces the remaining amount of computational effort required to meet the system goals
opportunistic incremental construction of solutions at different levels of abstraction, the ability to concurrently pursue multiple lines of reasoning, the possibility of multiple derivation paths for solutions,a dn the existence of multiple methods for accomplishing goals
how are the issues of search here specific to blackboards, rather than constraint satisfaction/search in general???
once you have multiple panels in a hierarchy, how is this different from a high-level semantic net???
blackboards are an architecture. how does the architecture reflect/direct preconceptions???
does the blackboard system help with the frame problem???
I kind of suspect that it might be relatively easy for a blackboard to self-organise, but the modularity that might ensue might well not be very human-transparent
I don�t understand how the blackboard system is any better/different for working on different levels of abstraction
I suppose it�s because all of the information, at all the different levels, is placed in the same space
hmm, but that�s not true. it seems almost as though the blackboard metaphor is misleading � what work does it really do??? after all, all the hard computational work gets down by these KSs which work in different domains (defined by their preconditions)
all the blackboard does is implement a universal communication protocol for sharing/showing information � perhaps that actually makes a big difference to the way the system works/is designed though???
are we supposed to be reading this with regard to Open Mind implementations???
what other CS stuff is going on in the Media Lab besides Open Mind???
how successful has Open Mind been??? is there a predicted critical mass???
have you tried focusing in on a domain to see whether the approach could work when you drill down???