Re: CogAI vs. sparql or CogAI + Sparql/GraphQL?

Hi Timothy,

SPARQL is a query language for RDF, whereas cognitive AI needs to deal with knowledge that is imperfect and imprecise. Natural language is an obvious way to interact with cognitive agents, but it is also possible to define a structured query language, e.g. “few ?x where color of ?x includes yellow from ?x kind-of rose”. That is an example from my web-based demo.  This is based on the plausible knowledge notation (PKN), and supports a variety of fuzzy quantifiers. PKN supports properties, relationships and implications, together with qualitative metadata as a coarse model of prior knowledge.

Cognitive AI lends itself to distributed systems with an analogy to HTTP’s request/response patterns.  My previous work on chunks and rules featured asynchronous operations on chunk databases, along with the analogy to HTTP.  More generally,  machine learning could be speeded up via a form of hive minds in which cognitive agents share cognitive databases, so that learning by one is accessible to all. A more sophisticated approach involves a diversity of agents with different personalities, skills and values.  This would give substantial advantages when it comes to addressing complex challenges. Of course, we are a long way from realising that, and there are many research challenges ahead.

Cheers,
Dave

> On 6 Jan 2023, at 17:27, Timothy Holborn <timothy.holborn@gmail.com> wrote:
> 
> Happy new year all; I hope we're able to make significantly positive progress during 2023.
> 
> I've been writing documentation for an ecosystem solution. The documentation is being split into two parts and it's not really ready yet - indeed, the more notable area of work, on 'permissive commons technology' has only just been started.  
> 
> So, as I'm going through the process of defining this ecosystem's related methods; i was led to ponder what the future relationship might be between a query standard like sparql vs. CogAI. 
> 
> I think the easy answer is that 'they're different things', but I wasn't confident that the conclusion was 100% correct.  So, I thought I'd ask. 
> 
> - Might it be the case that CogAI could provide what is in-effect an alternative (or major update) to Sparql?
> - Or are they entirely different things - that compliment each-other, in a broader ecosystem? 
> 
> Therein - my intended meaning is not that it is feasible to create a useful system (at least not like what i'm working to create) that doesn't support sparql (and graphql fwiw); but that, when making entirely new systems (ie: permissive commons) I wonder to what extent a SPARQL like agent is required at all, if the work on CogAI is sufficiently supported to advance to some sort of future alternative, etc. 
> 
> Thereafter the inference being; how might CogAI provide a request language, for sending queries across a decentralised network of records stored in semweb related formats, etc...? 
> 
> perhaps alternatively also - how are CogAI related queries performed via sparql-family interfaces? or for that matter also, graphql?  
> 
> apologies if my questions are foolish...  
> 
> some other notes below - might simply be tl;dr,
> 
> Cheers, 
> 
> Timothy Holborn.
> 
> Few other notes....
> 
> _______________________________________________________________________
> ChatAI (chat.openai.com <http://chat.openai.com/>) 
> 
> I provided ChatAI: https://github.com/w3c/cogai/blob/master/demos/decision-tree/rules.chk
> 
> this was its response, 
> 
> It appears that this is an example of a chunk file written in the procedural knowledge representation language used by the CogAI system. This chunk file appears to contain a series of rules that describe how to make a decision about whether to play golf based on the weather conditions.
> 
> The first rule states that if the outlook is sunny, the next state to consider is the humidity. The second and third rules describe how to make a decision based on the humidity: if the humidity is less than or equal to 75, the decision is to play golf, but if the humidity is greater than 75, the decision is not to play.
> 
> The fourth rule states that if the outlook is cloudy, the decision is to play golf. The fifth rule states that if the outlook is rainy, the next state to consider is the wind. The final two rules describe how to make a decision based on the wind: if the wind is true, the decision is not to play, but if the wind is false, the decision is to play.
> 
> I then asked it: "Can you write me an example of a chunk file that describes the procedural knowledge representation for how to make a decision about whether or not an action breached the UDHR as described by an ontology 
> 
> and it provided this response,
> 
> udhr {state start} =>
>     udhr {state action}
> 
> udhr {state action; action ?a} =>
>     udhr {state rights},
>     ontology {@query rights violated by ?a}
> 
> udhr {state rights; rights ?r} =>
>     udhr {state done},
>     console {@do show; value "Action breached UDHR: ?r"}
> 
> udhr {state rights; rights none} =>
>     udhr {state done},
>     console {@do show; value "Action did not breach UDHR"}
> 
> This chunk file contains four rules. The first rule states that when the system is in the "start" state, it should move to the "action" state. The second rule states that when the system is in the "action" state, it should query the ontology to determine which rights are violated by the given action. The third rule states that if any rights are returned by the ontology, the decision is that the action breached the UDHR, and the fourth rule states that if no rights are returned, the decision is that the action did not breach the UDHR.
> 
> _______________________________________________________________________
> 
> noting - it appears; that whilst ChatAI knows about rww/solid (and is helpful when working on related software); it doesn't know about CogAi.  although maybe, its just an area of technology that its got rules that restrict the usefulness of it for this type of purpose.  idk.  
> 
> I'm looking forward to doing more experimentation soon.... and finding alternatives (whilst seeking to build tooling for decentralised alternatives).
> 
> FWIW: the website connected to DID:Health was an interesting find, 
> https://github.com/w3c/did-spec-registries/commit/347608e6903fd2c9cc867e6949a86f615f1db266
> 
> (FWIW: I'm still not sure whether DIDs will end-up being useful for the ecosystem I've been working on, but that's a tangent...  )
> 
> cheers,
> 
> Tim.h.
> 

Dave Raggett <dsr@w3.org>