Representation of AI applications and predictions

Derived from the discussion in saref4watr#1 (closed).

Here is a use case:

I have a sensor that observes the temperature of a room. On the other hand, I have an AI application that based on previous data, predicts the future temperature of the room.

The question is how the AI application and its prediction can be represented in SAREF. Would it make sense to model prediction as a saref:Observation, as a subclass of saref:Observation, or as a subclass of saref:ProcedureExecution? In any case, the procedure is executed by an AI application, not by something tangible like a device. An approach could be a generalization based on https://w3id.org/pep/ProcedureExecutor.

mentioned in issue saref4watr#1 (closed)

One thing to take into account here is that right now in SAREF the concept of Device (as a tangible object) is a central one. If we want to include other things apart from devices we will need to allow other concepts to be involved in the measurements as well as to relax some of the axioms that we currently have.

I'm not against such a relaxation, and would happily implement what's required for v4. Any other opinion @all ?

my own timeline for v4 is: final draft 26/11 for validation during SmartM2M#72, and publication in december (hopefully).

Interesting point. I'm also not against such a relaxation. I think it would be in line with the broadening of the A from SAREF which originally meant "Appliances" but already for quite some time means "Application". A relaxation also would be in line with SSN such that the sensor may also be a human.

Still, I don't think we should see such an (AI) application that has made some forecasts as something special. It's just an application/device like many others that produces a sequence of values, the semantics of which in this case means that it's forecasted. So, we should definitely be using the same structure.

Another important consideration is to keep the structure clear and concrete, instead of abstract, such that it's clear that the device remains central to the ontology architecture.

Although the current documentation states:

OP saref:madeBy links a procedure execution to the device that made it.

in the ontology, saref:madeBy and saref:madeExecution can already be applied to other things than devices (it applies to features of interest in general)

also, saref:hasFunction applies to feature kinds and features of interest in general, not just to devices.

To simplify things, I suggest we add a sentence in the specification, that explains that other things can make executions and have functions too, and give as example a software application (incl. AI applications or simulation models)

resolution found during call 2025-01-17:

TODO: update the TS as follows:

OP saref:madeBy links a procedure execution to the device that made it.

EXAMPLE 1: <>

NOTE: in general, OP saref:madeBy can link a procedure execution to any entity, not necessarily a tangible device.

EXAMPLE 2: example with a software application

EXAMPLE 3: example with a person

changed milestone to %V4.1.1

added 2 - Needs implementation label

implemented 2025-02-11. propose closing

added 5 - Propose closing label and removed 2 - Needs implementation label