thoughtcrumbs : january 2000

Electric cars: Battery caching, navel gazing, and community building

CMU robotics prof Illah Nourbakhsh unveiled ChargeCar today, a project to redesign electric cars for urban commutes. He prefaced it by recounting all the times he's been in his own electric car, a RAV4EV, driving 70mph by the airport when he'd hear an NPR interview lamenting that electric cars just aren't able to go highway speeds or be dependable for long trips. He's also never visited a mechanic; no scheduled maintenance because the system is so much simpler, and regenerative braking means the motor does most of the work, rather than the brake pads. In eight years, he's only replaced the tires. (Though he can't pass PA emissions testing: You need a tailpipe.)

ChargeCar is based on the idea that hilly, stop-and-go, traffic- and coffee-detour-ridden urban commute is ideal for electric motors (and sucky for gas ones), but not for electric batteries. Batteries can't handle the constant changes, and have trouble regulating thermal processes when they're under heavy load. So, he treats them like a database and adds a caching layer: a low-voltage supercapacitor, which changes current direction easily but has lower energy density. Then he tries to hit the capacitor rather than the battery whenever possibly (charging when braking, and discharging to accelerate).

Okay, so battery caching is awesome. But the project also focuses on harnessing the power of communities:

1. Communities of commuters submit .gpx data from their daily drives to the site. (Is there an iPhone app in the works?) In return, they get personal informatics about their commute efficiency, cost, and carbon footprint. And how much they would save (in real cash money) if they'd been driving an electric. Could be used to pit neighbors against each other, à la this California MUD's social comparison energy bills.
2. Communities of hackers. They've released the totally naive controller algorithm (always hit capacitor first, then battery) as open source in Ruby. Hackers are encouraged to submit better algorithms. One of them will win an electric car. And über bragging rights.
3. Communities of mechanics. Pittsburgh is full of good mechanics and metal shops. One chop shop owner gutted the engine, gas tank, and transmission from a Toyota Scion demo car for free (the parts were worth more to him than his time) so the researchers could fill it with their own electric engine and batteries. ChargeCar would foster a community of mechanics to recycle cars (like Cash for Clunkers, eco-style).

addendum: vegetarian zombie tattoo beats all

Per the last post, Seth figured out the winning combination:

vegetarian beats zombie

The online dating site OKCupid, which had me at "hello, here's how our match algorithm works," has been publishing troves of great stats about self-presentation and response rates. In their latest analysis of the impact of words in conversational overtures, they find that netspeak (ur, u, luv) and physical compliments (sexy, hot, cutie) are a big FAIL. And that niche interests (like "zombie" and "vegetarian") increase reply rates. Their presentation of the results is even great (but guys, where are the error bars? What's a stats nerd to think?):

It's remarkably similar to the research I did the first couple of years in grad school, using machine learning to identify phrases that predict whether a forum message will get a response. In controlled field experiments we found that saying things like "I've been lurking here" or "I've been following this thread" roughly doubled reply rates. Seriously. (PDF of our paper on rhetorical strategies). But their data set is waaay juicier.

Autism and social norms online

Just submitted the final version of this paper:

Burke, M., Kraut, R., and Williams, D. (to appear). Social use of computer-mediated communication by adults on the autism spectrum. Proceedings of the ACM Conference on Computer-Supported Cooperative Work (CSCW) 2010. PDF

It's going to be the foundation of my thesis, and I'm psyched to be applying a combination of ninja data-munging and ethnography on a really meaningful problem. The gist is that a lot of adults on the high-functioning end of the autism spectrum are using Facebook, fan forums, online dating sites, and IM to meet new people and support existing relationships in a relatively comfortable medium. CMC removes the need for eye contact, reduces pressure to respond quickly, and is fairly structured and predictable, all properties that mesh with characteristics common to individuals with autism. By removing intonation and nonverbal cues (occasional emoticon ;-) notwithstanding), CMC levels the playing field for interpretation (or misinterpretation).

But CMC isn't ideal. It removes some forms of feedback and is subject to emerging social norms (such as how quickly to reply to an email from a new acquaintance). It's hard to tell who to trust, especially if you don't have a strong network of mutual acquaintances to vouch for someone.

Last spring I interviewed 16 Pittsburgh-area adults on the autism spectrum, and they shared their email inboxes, fan fiction, and MySpace pages with me. We talked about face-to-face social interactions, especially with classmates, co-workers, family, and friends. Then we logged in together and they walked me through any online communities or CMC tools they regularly used, and explained how online interactions differed from face-to-face ones. About half of the adults in the study were actively seeking to build new relationships, so this paper documents their successes and challenges, and proposes technology designs to support their goals.

One of the biggest challenges for several interviewees was figuring out social norms online. What should I say in my online dating profile? How soon should I email a new friend? How many Facebook status updates is too many? This is a problem for everybody to some degree, so interventions focused on one population should generalize well. So, I'm thinking about modeling typical social interactions online (messaging frequency, status updating, txting, tweeting, etc.), plotting distributions of activity, and showing an individual where he or she fits on that distribution. ("You update your status more than 60% of your friends, and more than 71% of all Facebook users your age.") And showing outcomes for following/deviating from the norms ("People who post their Farmville stories *this* often are 3X more likely to get unfriended.") The things that are easiest to measure aren't usually the most important (e.g., good friends won't diss you for tweeting too much, but poor phone or txting manners can stress a fledgling relationship), so if you have ideas for good behaviors to measure (and how to do it), shoot them my way. We're also talking about creating social stories (vignettes about somewhat ambiguous social situations) specific to CMC.

Out Safir-ing

From Ammon Shea's piece in this week's On Language.

Those who have pet peeves about language have chosen their domain of umbrage well, since it is inevitable that people will continue to make errors in speech and writing, and so there will always be something to be offended by, which can be very comforting.

And because the most savory quotes from the piece were longer than 140 characters--the man makes a great, if verbose, argument--I'm posting it here rather than Twitter. If Shea keeps contributing to the Times, perhaps this blog won't languish.