I wrote this as a heads-up email after yesterday’s AgentSwitch catch up in London, but I think it might be useful for posterity and general interest. This is not intended as objective ‘Minutes of the Meeting’, rather, I’m discussing some points from my point of view.
Ongoing development
Soton hired a new full time developer (Michele) who is exclusively working on AgentSwitch.
Besides code improvements, the major change in the architecture/paradigm of the app, is a shift from a stateless RESTful app that only stores and processes data for the duration of a user session to a background service/DAEMON-enabled model, where data is kept and processed offline (i.e., beyond the duration of the user session). Reasons are that a) the Matlab-driven prediction is computationally expensive and takes too long to return for real-time interaction on the web, and b) to enable advanced provenance tracking. Output by the Matlab module is also required by the new modules that will drive group buying. I will talk more about group buying further down. First, I wanted to highlight that this changed model is closer to the ‘agent’ vision that provides an opportunity to think about interactional arrangements. E.g., this offline service means the user could (and probably should) be notified when the prediction is ‘ready’, etc. This raises questions how and when the user might be notified.
To enable notifications, I have suggested an event-module should be implemented that interested parties can subscribe to, similar to https://code.google.com/p/pubsubhubbub/ or http://urbanairship.com/ or https://developers.google.com/android/c2dm/ - and there’s perhaps more/better fitting solutions depending what the notification channel of choice should be, e.g., mobile notifications vs. email.
Another major change to highlight is planned support for multiple data sources such as CurrentCost and AlertMe (and Pachube was mentioned) data stores via their API, but also for a future MyJoulo type logger to be developed (nb., this is what we hope to use for the GreenVision/WesterPower deployments).
Regarding deployments, to enable UIs in which users are likely to view groups of homes, the backend needs to support user interfaces beyond the one-use-one-dataset model. This is a requirement for scenarios like the GreenVision/WesterPower deployments: an ‘energy advisor’ might want to upload, process, analyse and compare multiple datasets.
Group buying
Technically, Soton has a major interest in moving this forward, there is clear value and potential to publish around this aspect in the MAS and (applied) machine learning community. I have stated my reservations about this from an HCI perspective, mainly based on the fact that – in contrast to the rest of the system – group buying relies on fictional tariffs. To the user, there is no obvious transfer or impact of the (potential) experience with group buying in AgentSwitch to the user’s everyday life.
Having said that, the development of group buying is being pushed and it is likely this ‘feature’ will be the Next Big Thing in AgentSwitch. Gopal is keen to do a pilot user study of some sort around February, and perhaps aim for a submission to UbiComp. Perhaps a more technically focussed short paper or a long paper with a ‘proof of concept’ style user study of sorts would be possible. Something similar to the IUI paper maybe. Something to think about. An advantage over the deployments below is that this is much less contingent on other parties’ availability, willingness to co-operate etc.
Deployments and user testing
I have outlined our desire to re-skin and re-purpose AgentSwitch meets MyJoulo as a tool to support energy advisors in their work. Opportunities with GreenVision and WesternPower are currently being pursued by Alex and me, but are still uncertain. The ‘Centre for Sustainable Energy’, a charity and collaborator of WesternPower also do this kind of work: http://www.cse.org.uk/projects/view/1195 . This kind of work has a lot of potential to publish studies of work practice of energy advisors, participatory design sessions with energy advisors to co-design the tools to support them, and studies of the technology in use, in support of the real-world work of energy advisors.
I am very keen on this line of work and hope we can do something in this space, we will hopefully have some more concrete info whether and how this will be moved forward before Christmas.