OWL, JSON, Atomese and friends

[kabirkbr]

This is a discussion transferred from another repo (nunet-io/ai-dsl-ontology#13) in the process of merging them. Things may have changed since then, but i copying everything in its original form, at least for the purposes or keeping proper history...

-- original text follows --

These are raw ideas, or rather list of references with comments, so beware.

1. Often a chosen format for expressing metadata is JSON. We use it at nunet (https://gitlab.com/nunet/nunet-infra/-/tree/master/metadata-descriptions) and by snet (https://dev.singularitynet.io/docs/ai-developers/service-setup/#metadata-overview). json is good for many reasons, but it is less good for all kinds of formal reasoning and validation purposes. First and most simple, we cannot prove the correctness of json structures without some additional specifications.
2. In decentralized machine-to-machine communication systems this becomes a problem -- we want a sound way to check and validate data streams consumed and produced by services (inputs and outputs) in order to stitch them together. The issue of verifying json data against ontologies or schemas is an active r&d field:
   • translating json schema to owl axioms;
   • JSON Schema (https://json-schema.org/understanding-json-schema/index.html);
3. There seems to be a way to encode Haskell types in json and to the converstion (which possibly means that we can do the same in Idris):
   • https://stackoverflow.com/questions/26694619/haskell-way-of-modeling-a-type-with-dynamic-json-fields/26697459
   • https://hackage.haskell.org/package/aeson-0.6.1.0/docs/Data-Aeson.html#g:8
4. It seems that there is (or at least was) some interest in converting json to atomese (some potentially useful code.  opencog/agi-bio#84) which may (?) allow to pass data (and programs) to / fro OpenCog engine and would be very much into the spirit of AI-DSL and Singnet in general. In any case, it does not seem too difficult to make such conversion.
5. For AI-DSL data ontologies we will most probably need to enable conversions between /understanding of different ontology standards. For example in biomedical domain, OBO is widely used for describing data ontology, while OWL is considered a gold standard for ontologies (http://www.obofoundry.org/). It seems that OBO can be expressed in/converted to OWL, however (https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-019-3002-3).
6. All above points me to the direction of trying to define json-schemas (or owl ontologies) for workflow, service and machine - metadata (https://gitlab.com/nunet/nunet-infra/-/tree/master/metadata-descriptions) with the distant idea that parts of these shemas (pertaining to input or output types) could be converted to Idris for type checking. However, the very usage of Idris here becomes questionable, since its purity would be seriously abused...
7. Whether Idris is involved or not for 'probabilistic type checking' or evaluating a complex decision function required for matching inputs and outputs of services, expressing metadata / input / outputs in popular and widely used languages may allow us to use several approaches at once (e.g. type-checking by Idris and decision function by other languages).
8. I assume that the fact that we use grpc is not an issue, since protobuf files could be easily converted at least to json...

[kabirkbr]

It seems that we could try using the following package for that https://github.com/arhs/owl2json-schema;
So the idea then would be to define owl ontology, convert it to json schema and verity actual json metadata files against that schema.