W3C

– DRAFT –
Voice Interaction

28 August 2024

Attendees

Present
debbie, dirk, gerard
Regrets
-
Chair
debbie
Scribe
ddahl

Meeting minutes

dirk: need to talk about different types of digital assistants and not everything applies to all of them

gerard: personal assistant in the cloud is problematic

debbie: we should actually say why personal assistants in the cloud are problematic

debbie: there is the personal assistant that knows about you and doesn't represent a company

dirk: we could point out that it would be good if it were not in the cloud

debbie: enterprise assistant represents, e.g. your bank, your school, neighborhood association

dirk: data is shared between you and the entity in a trust relationship
… what about something like Siri or Google? Data is shared but there's no trust relationship, generic cloud-based assistant

debbie: public assistant that you use, but don't share data, like a smart city assistant

debbie: embedded personal assistant, like Mycroft

dirk: it could share data with your bank, so there's a transition

debbie: not completely clear-cut

dirk: state these three types

debbie: there could be some in-between cases

debbie: what about IoT devices

dirk: they would be like public assistants?
… just authentication

debbie: it just has to know that you're authorized to turn on the lights

dirk: they might also start collecting patterns about how you use them?

debbie: data could be gathered by the assistant or explicitly shared data
… there's data that the assistant knows about you but you don't, for example, your bank balance or your grades
… or medical

dirk: there must be a taxonomy of private data types

debbie: we should look up different types of assistants
… we should wait to modify documents until we get more information about assistant types

dirk: we should send this question to the mailing list

ACTION: dirk to send question to the mailing list

dirk: made changes to one drawing
… tried to resketch one image

debbie: in the diagram we have LLMs, that includes both foundation models and adapted models

debbie: we should send some kind of announcement about the implementation at some point

dirk: it can be used as a basis, but needs some tweaking

(looking at GitHub issues)

dirk: after 54, 55, and 56 we can go for 0.1

dirk: how to publish? source code, doxygen
… should we share binaries?

debbie: yes, that's pretty common

dirk: doxygen is like javadoc but not language-specific
… usually used for C++ code

debbie: something higher level than doxygen?

dirk: do we need interfaces document?

debbie: generate and see what's missing

dirk: the interfaces document should be updated to be in line with the current implementation

ACTION: dirk to update issues with issues from today's discussion

ACTION: debbie to check on issue 48 (license)

Summary of action items

  1. dirk to send question to the mailing list
  2. dirk to update issues with issues from today's discussion
  3. debbie to check on issue 48 (license)
Minutes manually created (not a transcript), formatted by scribe.perl version 229 (Thu Jul 25 08:38:54 2024 UTC).

Diagnostics

Succeeded: s/assitant/assistant

All speakers: debbie, dirk, gerard

Active on IRC: ddahl