Toward Hybrid Computing

Editor's Notes

• #5: http://interrelativity.com/joe/research.html
• #9: Our first generation proactive display was a shared aural rather than visual display, created when I was at Accenture Technology Labs, was called MusicFX. The selection of music playing in a fitness center is a perennial problem for the members and staff – trying to find music that everyone loves, or, at least, no one hates, is a challenging problem. MusicFX enabled members of a corporate fitness center to create a profile of how much they liked or didn’t like each of 91 different genres of music (shown in the middle image) and associate this with their employee badge. People would badge in when they entered the fitness center (shown in the right image), adding their preferences to a pool, and a weighted random selection algorithm would then find music that was most popular – or least objectionable – to the people working out at any given time. References: http://interrelativity.com/joe/projects/MusicFX.html http://www.youtube.com/watch?v=WLgXLl4uGYk
• #12: We installed the MusicFX system in the fitness center at Technology Park in Northbrook in November 1997. The system was an instant hit with both the fitness center staff and its members. We conducted a poll of members six weeks after the system was installed. Over 70% of the respondents told us they liked having MusicFX select music more than the old complaint-driven music selection scheme. Only 7% of those polled told us they liked the old scheme better (these were probably among the squeaky wheels mentioned earlier). The others either thought the music was about the same or had joined during the previous six weeks and therefore had no experience with the old scheme. We asked members what they liked best and what they liked least about the system (in free text). Half of the respondents told us they liked the increased variety of music played -- we were very happy to see this, because although Joe and I were both eager to increase the variety of music, we weren’t sure how this would be greeted by the rest of the members. Over a third of the people we polled told us they liked havingsome say in what music was being played -- the silent majority was given a voice, and they liked it. The least liked feature cited by members was abrupt station changes -- when the system determines it’s time to change, it just does it; although we considered trying to detect the end of song boundaries and delaying the channel change until the end of a song on the old channel, we would likely cut into the middle of a song on the new channel. We may implement a feature to fade out, or turn down the volume on the old channel before the change, and then turn up the volume to fade in to the new channel. Finally, a few people complained about occasional bad (or whacky) music -- a natural side effect of the increased variety.
• #13: I’ll start off with some charts that show the most popular and least popular stations, as well as a list of the stations that we have yet to hear. Then I’ll talk about human nature and group dynamics: what we’ve learned about people’s musical preferences, what happens when you try to take into account a large pool of preferences, and finally some interesting experiences we -- and others -- have had with the system.
• #14: One context in which a larger group of people who don’t know each other so well gathers for a period of time is a conference. When I moved from Accenture to Intel Research Seattle, I was General Chair of UbiComp 2003, and so we decided to experiment with the idea of proactive displays at the upcoming conference. We again created special-purpose web-based profiles for people (shown in the upper left), in which they could enter their name, affiliation, photo of themselves, a photo representing some area of interest, and their home page. These profiles were associated with RFID tags that could be inserted into their conference name badge sleeves (shown in the upper middle photo), and then RFID antennas mounted near large displays would enable those displays to sense and respond to people in three different ways. The AutoSpeakerID application (lower left photo) showed the name, affiliation and photo of someone detected in front of the microphone stand during the question and answer period after a conference presentation, enabling the audience to see who the person asking the question was (and figure out how to spell their name). Ticket2Talk showed the same information, plus the photo representing a person’s interest, near the coffee break table, giving each person moving through the line 5 seconds of fame, and offering them a “ticket to talk” with each other about the photo of interest. Interestingly, one of the benefits people reported was that, because the person’s name was also shown on the Ticket2Talk display, they could save face by being reminded of a person’s name without having to look at their name tag. The third application, Neighborhood Window, showed a network graph visualization with nodes listing the names and photos of people detected near the display, and a small number of connected nodes that showed words and phrases the people’s home pages had in common, or those that were unique across the entire population, figuring those were the two most likely sets of topics to spark conversations. References: http://interrelativity.com/proactivedisplays/
• #27: http://interrelativity.com/joe/research.html
• #28: http://interrelativity.com/joe http://gumption.typepad.com http://www.slideshare.net/gumption