[Interoperability] Feedback on section 3.3 of the Interop Report.

[FabienGandon]

A few comments/suggestions about the current version of section 3.3 of the Interop Report.

I was surprised by "Value Alignment : The agent acts in ways that are consistent with the goals, preferences, and interests of its end-user, and that respect human values, ethical principles, societal norms, and fundamental human rights." : I understand the first part is directly link to the User Agent values of the Web Arch. While I agree with the second part I wonder how " respect human values, ethical principles, societal norms, and fundamental human rights." works in a Web that spans several societies, countries and cultures that, for instance, do not share the same "societal norms" or "ethical principles".

Concerning "Figure 3 shows a set of architectural patterns (...)" I did not see the patterns immediately in Fig. 3. May be you want to mention in the legend of Fig. 3 that the patterns are the boxes and that there are five of them interconnected in that figure. There is also a small typo "Figure 3 A pattern langu a ge for Web-based agents and MAS."

Fig.4 : I am not sure I understand what is the difference between a provided interface and a shared interface and why it matters.

"The trade-off is decreased efficiency: components must translate (...)" Here I would say "The trade-off is decreased local/individual efficiency"

[smnmyr]

On the value alignment, I agree and would further shorten the statement from

The agent acts in ways that are consistent with the goals, preferences, and interests of its end-user, and that respect human values and norms.

to

The agent acts in ways that are consistent with the goals, preferences, and interests of its user.

...and then directly link to https://w3ctag.github.io/user-agents/ as soon as that is done.

[andreiciortea]

Thanks @FabienGandon and @smnmyr for the feedback.

We discussed in today's regular meeting (with @smnmyr as well), and we reached consensus on the following definition:

The agent acts in ways that are consistent with the goals, preferences, and interests of its end-user, and that respect stakeholder values and norms.

We reached rough consensus on the following points (cc @smnmyr @bassetthound @scranefield @habtom @asorici and Önder):

• we agreed it is important to keep value alignment as a design goal and to acknowledge all the issues surrounding it, but the group will not directly address those issues; we will discuss ethical implications (for this point and others) in a dedicated section
• we agreed to keep both parts of the definition to capture the values of both the end-user and the society (I'm using "society" loosely here)
• we agreed on "stakeholder values" for narrowing down the discussion; we will explain in the text what we mean
• value alignment might provide further input to a discussion on transparency; there is also a large amount of work on policies and norms in MAS that seems related (incl. monitoring and enforcing norms)

@FabienGandon I've added a legend to Figure 3 and fixed the typo in the caption. We did not get to discuss shared vs. provided interfaces. A shared interface is just a way to show visually that it is, in fact, the same interface provided to multiple components (they are not different interfaces). We carry this over from Danny Weyns’ original diagrams (referenced in the report). However, given that these interfaces are internal to the agent, whether they are defined as shared or not should not have much impact on the rest of the report. I will raise this point in the next meeting for more input.