Re: [dispatch] [hrpc] [art] [Secdispatch] Open Ethics Transparency Protocol

[[e-mail]]

Thanks a lot John, Gina, Larry, Michael.
All your comments are very valuable here I'll try to put my reflections 
as a response in one single email to continue the conversation.

@Michael,
> Are you talking about security disclosures?
> Or something else?
The approach/lens that we're taking here is to allow describing every 
autonomous system with the help of the ethical "vector". We are aiming 
to bring at least a very basic formalism to what is now becoming a 
buzzword. https://openethics.ai/vector/

Security part in the disclosure which forms these vectors is a 
component, but not everything. In some applications security is key, 
while in others users are ready to compromise security for other 
"values"

> I don't really understand the semantics of the content
When we have a disclosure, we transform it in the machine-readable form 
(the one you've seen in the section 5).

Then we present this disclosure in the machine-readable form - to other 
machines
And to humans - in the form of icons. The basic implementation of the 
label is available here https://openethics.ai/label/

The machine-readable file and the visual label could be generated here 
https://openethics.ai/label/generate/

> in some ways it is similiar to securitytxt in syntax
Indeed, the securitytxt is similar to the extent that they provide 
mechanism to the disclosure. In Open Ethics I've decided to bring the 
infrastructure for the signatures (PGP in the securitytxt 
implementation). PGP doesn't cover the case of disclosure withdrawal 
which is critical for the applications that

>  think you need a more detailed, more well worked set of examples
Thank you! This is something that I should focus on, 100%


@Gina,
> Is this referring to Datasheets for Datasets?
Yes, this is very close and linked. Let me explain.
1)
Datasheets for Datasets, Google Model Cards, are integral part to 
describe the Training Datasets. We're augmenting it with Data Passport. 
The Data Passport has a purpose at depicting the origins of the training 
datasets by bringing a standardized approach to convey information about 
data annotation processes, data labelers profiles, and correct scoping 
of the labeler’s job.
https://github.com/OpenEthicsAI/OEDP

2) We insist that the information about training datasets/decision logic 
is key in the disclosure, however should be augmented by to other 
pillars to call systems "disclosure-complete". Together they are: (A) 
Code libraries/components, including the approach to data processing and 
data transfer. (B) Decision Space, restricted vs unrestricted, plus 
information about failure modes of the system. (C) Training data and 
heuristics origins.

3) Transparency Protocol is not only about the process of the disclosure 
formation, but importantly about how it should be exchanged. The goal is 
so that transparency will not be only on the "surface", but will also 
tell us something about the components with their approaches, practices, 
processes, code put in place.

@Larry, @John
Thanks a lot, we've been looking at this work at W3C .
This is close with the overall mission, but different in implementation.

We're guided by a very similar set of assumptions. 
https://openethics.ai/manifesto/

What we believe is that similar statements should become widespread. 
However, it is not sufficient for a regulator or a community of 
"ethical" developers to start using them. They should be communicated in 
a standard manner. What is crucial is to unlock the bottom-up regulatory 
mechanisms, and to make sure that every user/consumer with no knowledge 
of security, privacy, fairness, bias, safety, architecture, etc, will 
start learning more about labels on the digital products. This has 
happened in other industries like food, construction, pharma, but not 
yet in the IT.


Looking forward to your feedback and thank you a lot for your 
contributions and thoughts.

Nikita Lukianets

---













On 2022-02-01 14:34, John Curran wrote:
>> On 31 Jan 2022, at 8:22 PM, Larry Masinter <[e-mail]> wrote:
>> 
>> Check out
>> https://github.com/w3ctag/ethical-web-principles
>> which seems to have some momentum and some similar goals.
> 
> Larry -
> 
>> A very interesting initiative - thanks for sharing.
>> 
>> After brief review, it would appear that the web folks have a
>> signficiant advantage in their efforts to protect human rights on
>> the web, as they are successful in affirmatively stating their
>> desired outcome for its users – e.g. "The web must make it
>> possible for people to verify the information they see” - this
>> speaks directly to the functionality that must be provided to the
>> users rather than the human rights implications of an application
>> failing to do so…
> 
>> 
> 
>> An interesting thought exercise lies in considering why the
>> IETF/IRTF work in this area differs in this regard - I suspect it is
>> not lack of desire, but inherent to dealing with many protocols and
>> implied application contexts rather than one predominant model (i.e.
>> the user/web browser context).   It is quite reasonable that the
>> abundance of protocols and user contexts results in pragmatic limits
>> in postulating desired user functionality for an ethical Internet,
>> but it does raise the question of whether more focused efforts for
>> the most popular applications (e.g. email, messaging) could be
>> undertaken in other contexts or for some reason are simply
>> unachievable outside of the web context...
> 
> Thanks again for the pointer to this work!
> /John
> Disclaimer:  my views alone - may cause drowsiness - do not read while
> operating heavy machinery.