https://futureofcoding.org/ logo
Join Slack
Powered by
<https://youtu.be/C8AkqwS_dJQ> (5m) Early-stage de...
# share-your-work
j

https://youtu.be/C8AkqwS_dJQ▾

(5m) Early-stage demo of the new auto-chatbot feature I'm working on in Blawx. Interested to know whether it's a compelling motivation for "rules as code", and why or why not.
a
Neat! Is this based on a standardized state machine encoding?
and/or datalog/prolog implementation 😉
j
Specifically, stable model constraint answer set programming, implemented in the s(CASP) library for SWI-Prolog.
a
oh, it looks like it might be working anyway, it’s just that the right sidebar wasn’t visible
j
Yeah, don't look too hard, or you will see my shame. And jsyk the live demo doesn't have the chatbot yet.
a
heh
http://dev.blawx.com/6/test/does_anything_fly/ says “No models” even when I introduce a penguin and a non-penguin
j
The sidebar should really pop open when people click run. Long list of stuff like that to fix, but I'm focused on some proof of concept issues first.
I don't see a penguin on that link. Did you save first?
a
No, I didn’t want to overwrite anything 🙂
Saved now
aha, saving is required, it doesn’t use transient states
j
Another of the issues on the backlog, yeah. :)
a
Looks pretty neat anyway! The chatbot interface seems potentially useful but the underlying ideas are cool.
Does the CLEAN markup get automatically translated to rules in any way, or just serve as something more akin to machine-readable doc?
j
It's just "markdown for LegalDocML".
I'm skeptical of attempts to auto-generate code from legal text. DataLex had had limited success with a weaker semantics. No one else has shown any progress. Eventually, Blawx might generate a dataset that would be useful for training a ML approach that would maybe get you started?
a
yeah, that sounds potentially fruitful
j
There are also people in Finland working on tech for legal drafters that would annotate the relationship between terms in legislation and ontological concepts say the source, which is a possible approach, but hard to get adopted.
a
Verification of the learned model will be a pain though :)
💯 1
j
And there is LegalRuleML, which is a logical target for attempts to extract rules from legal text. So some of the pieces exist. Just no stack that actually works, yet.
a
I think this concept might be useful for state machine modeling, not just legal stuff 🙂
j
I bet you're right. The visual code interface has wide applicability. Constraint answer set programming has wide applicability, but less so. I don't know about state machine modeling specifically, but formal verification tasks generally have a tonne of overlap with what s(CASP) is good at. I have often thought one potential use-case would be to generate a set of inputs and corresponding outputs that could be used to verify software built in other languages, for example.
If you create an example of state machine modeling in Blawx, send me a link or a pull request. Blawx doesn't have an "export" feature, in this version, yet. (Did I mention the list is long? It's a long-ass list.)
a
#startuplife
j
Seriously. If you have a better use case, let me know. Startup life is hard enough if your target clients are NOT technologically self-defeating governments that take a year to do security audits of bug fixes. 😡
2 Views
Open in Slack
PreviousNext
Powered by