Ready-to-hand Dean Eckles blogs on people and technology

The Wall Street Journal’s Venture Capital Dispatch reports on how Aardvark, the social question asking and answering service recently acquired by Google, used a Wizard of Oz prototype to learn about how their service concept would work without building all the tech before knowing if it was any good.

Aardvark employees would get the questions from beta test users and route them to users who were online and would have the answer to the question. This was done to test out the concept before the company spent the time and money to build it, said Damon Horowitz, co-founder of Aardvark, who spoke at Startup Lessons Learned, a conference in San Francisco on Friday.

“If people like this in super crappy form, then this is worth building, because they’ll like it even more,” Horowitz said of their initial idea.

At the same time it was testing a “fake” product powered by humans, the company started building the automated product to replace humans. While it used humans “behind the curtain,” it gained the benefit of learning from all the questions, including how to route the questions and the entire process with users.

This is a really good idea, as I’ve argued before on this blog and in a chapter for developers of mobile health interventions. What better way to (a) learn about how people will use and experience your service and (b) get training data for your machine learning system than to have humans-in-the-loop run the service?

My friend Chris Streeter wondered whether this was all done by Aardvark employees or whether workers on Amazon Mechanical Turk may have also been involved, especially in identifying the expertise of the early users of the service so that the employees could route the questions to the right place. I think this highlights how different parts of a service can draw on human and non-human intelligence in a variety of ways — via a micro-labor market, using skilled employees who will gain hands-on experience with customers, etc.

I also wonder what UIs the humans-in-the-loop used to accomplish this. It’d be great to get a peak. I’d expect that these were certainly rough around the edges, as was the Aardvark customer-facing UI.

Aardvark does a good job of being a quite sociable agent (e.g., when using it via instant messaging) that also gets out of the way of the human–human interaction between question askers and answers. I wonder how the language used by humans to coordinate and hand-off questions may have played into creating a positive para-social interaction with vark.

BusinessWeek has an exhuberant review of current industry research and product development related to understanding social networks using data from social network sites and other online communication such as email. It includes snippets from people doing very interesting social science research, like Duncan Watts, Cameron Marlow, and danah boyd. So it is worth checking out, even if you’re already familiar with the Facebook Data Team’s recent public reports (“Maintained Relationships”, “Gesundheit!”).

But I actually want to comment not on their comments, but on this section:

In an industry where the majority of ads go unclicked, even a small boost can make a big difference. One San Francisco advertising company, Rapleaf, carried out a friend-based campaign for a credit-card company that wanted to sell bank products to existing customers. Tailoring offers based on friends’ responses helped lift the average click rate from 0.9% to 2.7%. Although 97.3% of the people surfed past the ads, the click rate still tripled.

Rapleaf, which has harvested data from blogs, online forums, and social networks, says it follows the network behavior of 480 million people. It furnishes friendship data to help customers fine-tune their promotions. Its studies indicate borrowers are a better bet if their friends have higher credit ratings. This might mean a home buyer with a middling credit risk score of 550 should be treated as closer to 600 if most of his or her friends are in that range, says Rapleaf CEO Auren Hoffman.

The idea is that since you are more likely to behave like your friends, their behavior can be used to profile you and tailor some marketing to be more likely to result in compliance.

In the Persuasive Technology Lab at Stanford University, BJ Fogg has long emphasized how powerful and worrying personalization based on this kind of “persuasion profile” can be. Imagine that rather than just personalizing screens based on the books you are expected to like (a familiar idea), Amazon selects the kinds of influence strategies used based on a representation of what strategies work best against you: “Dean is a sucker for limited-time offers”, “Foot-in-the-door works really well against Domenico, especially when he is buying a gift.”

In 2006 two of our students, Fred Leach and Schuyler Kaye, created this goofy video illustrating approximately this concept:

My sense is that this kind of personalization is in wide use at places like Amazon, except that their “units of analysis/personalization” are individual tactics (e.g., Gold Box offers), rather than the social influence strategies that can be implemented in many ways and in combination with each other.

What’s interesting about the Rapleaf work described by BusinessWeek is that this enables persuasion profiling even before a service provider or marketer knows anything about you — except that you were referred by or are otherwise connected to a person. This gives them the ability to estimate your persuasion profile by using your social neighborhood, even if you haven’t disclosed this information about your social network.

While there has been some research on individual differences in responses to influence strategies (including when used by computers), as far as I know there isn’t much work on just how much the responses of friends covary. As a tool for influencers online, it doesn’t matter as much whether this variation explained by friends’ responses is also explained by other variables, as long as those variables aren’t available for the influencers to collect. But for us social scientists, it would be interesting to understand the mechanism by which there is this relationship: is it just that friends are likely to be similar in a bunch of ways and these predict our “persuasion profiles”, or are the processes of relationship creation that directly involve these similarities.

This is an exciting and scary direction, and I want to learn more about it.

Is this etching a self-portrait by Leonardo da Vinci created hundreds of years ago? That’s what I was told by a Californian friend who had “gone native” in Florence. Another matter: is this, in fact, a commonly believed and shared legend, and what other variations are there on it?

I shared the story with some fellow visitors in Florence on a lunch-time return to the piazza. Ed Chi tried to verify the rumor using a Web search, but with no success.  At least in English, there didn’t seem to be much on this in the Web. (See my photo and comments on Flickr.)

I posted the photo on Flickr. I asked questions on LinkedIn and Yahoo! Answers, with no success. I also asked for help from workers on Mechanical Turk. Here’s part of how I asked for help:

There is a portrait etched in stone on the wall of Palazzo Vecchio in Piazza della Signoria in Florence (Firenza), Italy. It is close behind the copy of the David there. I have heard that there is a legend that this is a self-portrait by Leonardo da Vinci. I am looking for any information about this legend, alternate versions of the legend, or information about the real source of the portrait.

What results have been offered seem to suggest that this legend exists — though perhaps it is “actually” (at least as captured online, since perhaps the Leonardo theorists aren’t as active digital content creators) about Michelangelo:

• Palazzo Vecchio in Italian Wikipedia
• Florentine Legends: Fact or Fiction (in Italian)
• Curiosities in Florence

The best way of finding out seemed to actually be my Flickr photo itself, since that’s where Daniel Witting provided the first two links above — however, this was a few months after the photo was first posted to Flickr. Turkers provided a couple useful links also (”Curiosities” above) on a shorter schedule and with a higher price. (I should have also tried uClue — where many former Google Answers researchers now work. This was recommended by Max Harper, who has studied Q&A sites in detail.)

-

Question and answer services along the lines of Yahoo! Answers rose to global (and U.S.) significance only after success in Korea, where Naver Knowledge iN pioneered the use of an online community to power a Q&A site. A major motivation Korea was the limited amount of Korean content online. With Naver’s offering, Korea’s Internet saavy, English population made information newly available in Korean (and did plenty of other interesting work).

This is as significant a motivation for Q&A sites by English-speaking folks in the U.S., but the present case is an exception.

Some of the questions that made this case interesting to me:

• What culturally-shared beliefs get manifest online? During this whole process, I and others wondered whether perhaps this local legend was only shared orally. It seems that it is represented online after all — at least the Michelangelo variant, but it could have been otherwise.
• How does the pair of languages a task requires knowledge of determine the processes, structres, and communities that are optimal for completing the task? For example, it seems quite important whether the target or source language has many more speakers than the other. (One could think about this simplistically in terms of conditional probabilities of skills with language A given skill with language B and vice verse.)

Increasing valuable annotation behaviors was a practical end of a good deal of work at Yahoo! Research Berkeley. ZoneTag is a mobile application and service that suggests tags when users choose to upload a photo (to Flickr) based on their past tags, the relevant tags of others, and events and places nearby. Through social influence and removing barriers, these suggestions influence users to expand and consistently use their tagging vocabulary (Ahern et al. 2006).

Context-aware suggestion techniques such as those used in ZoneTag can increase tagging, but what about users’ motivations for considering tagging in the first place? And how can these motivations for annotation be considered in designing services that involve annotation? In this post, I consider existing work on motivations for tagging, and I use tagging on Facebook as an example of how multiple motivations can combine to increase desired annotation behaviors.

Using photo-elicitation interviews with ZoneTag users who tag, Ames & Naaman (2007) present a two factor taxonomy of motivations for tagging. First, they categorize tagging motivations by function: is the motivating function of the tagging organizational or communicative? Organizational functions include supporting search, presenting photos by event, etc., while communicative functions include when tags provide information about the photos, their content, or are otherwise part of a communication (e.g., telling a joke). Second, they categorize tagging motivations by intended audience (or sociality): are the tags intended for my future self, people known to me (friends, family, coworkers, online contacts), or the general public?

Taxonomy of motivations for tagging from Ames & Naaman

On Flickr the function dimension generally maps onto the distinction between functionality that enables and is prior to arriving at the given photo or photos (organization) and functionality applicable once one is viewing a photo (communication). For example, I can find a photo (by me or someone else) by searching for a person’s name, and then use other tags applied to that photo to jog my memory of what event the photo was taken at.

Some Flickr users subscribe to RSS feeds for public photos tagged with their name, making for a communication function of tagging — particularly tagging of people in media — that is prior to “arriving” at a specific media object. These are generally techie power users, but this can matter for others. Some less techie participants in our studies reported noticing that their friends did this — so they became aware of tagging those friends’ names as a communicative act that would result in the friends finding the tagged photos.

This kind of function of tagging people is executed more generally — and for more than just techie power users — by Facebook. In tagging of photos, videos, and blog posts, tagging a person notifies them they have been tagged, and can add that they have been tagged to their friends’ News Feeds. This function has received a lot of attention from a privacy perspective (and it should). But I think it hints at the promise of making annotation behavior fulfill more of these functions simultaneously. When specifying content can also be used to specify recipients, annotation becomes an important trigger for communication.

—————

See some interesting comments (from Twitter) about tagging on Facebook:

• noticing people tagging to gain eyeballs
• exhorting others not to tag bad photos (and thanks)
• collapsing time by tagging photos from long ago
• tagging by parents

(Also see Facebook’s growing use and testing of autotagging [1, 2].)

References

Every person who logs into Facebook is met with the same interface but with personalized content. This interface is News Feed, which lists “news stories” generated by users’ Facebook friend. These news stories include the breaking news that Andrew was just tagged in a photo, that Neema declared he is a fan of a particular corporation, that Ellen joined a group expressing support for a charity, and that Alan says, “currently enjoying an iced coffee… anyone want to see a movie tonight?”

News Feed is an example of a particular design pattern that has recently become quite common – the activity stream. An activity stream aggregates actions of a set of individuals – such as a person’s egocentric social network – and displays the recent and/or interesting ones.

I’ve previously analysed, in a more fine-grained analysis of a particular (and now changed) interface element for setting one’s Facebook status message, how activity streams bias our beliefs about the frequency of others’ participation on social network services (SNSs). It works like this:

• We use availability to mind as a heuristic for estimating probability and frequency (Kahneman & Tversky, 1973). So if it is easier to think of a possibility, we judge it to be more likely or frequent. This heuristic is often helpful, but it also leads to bias due to, e.g., recent experience, search strategy (compare thinking of words starting with ‘r’ versus words with ‘r’ as the third letter).
• Activity streams show a recent subset of the activity available (think for now of a simple activity stream, like that on one’s Twitter home page).
• Activity streams show activity that is more likely to be interesting and is more likely to have comments on it.

Through the availability heuristic (and other mechanisms) this leads to one to estimate that (1) people in one’s egocentric network are generating activity on Facebook more frequently than they actually are and (2) stories with particular characteristics (e.g., comments on them) are more (or less) common in one’s egocentric network than they actually are.

Personalized cultivation

When thinking about this in the larger picture, one can see this as a kind of cultivation effect of algorithmic selection processes in interpersonal media. According to cultivation theory (see Williams, 2006, for an application to MMORGs), our long-term exposure to media makes leads us to see the real world through the lens of the media world; this exposure gradually results in beliefs about the world based on the systematic distortions of the media world (Gerbner et al., 1980). For example, heavy television viewing predicts giving more “television world” answers to questions — overestimating the frequency of men working in law enforcement and the probability of experiencing violent acts. A critical difference here is that with activity streams, similar cultivation can occur with regard to our local social and cultural neighborhood.

Aims of personalization

Automated personalization has traditionally focused on optimizing for relevance – keep users looking, get them clicking for more information, and make them participate related to this relevant content. But the considerations here highlight another goal of personalization: personalization for strategic influence on attitudes that matter for participation. These goals can be in tension. For example, should the system present…

The most interesting and relevant photos to a user?

Showing photographs from a user’s network that have many views and comments may result in showing photos that are very interesting to the user. However, seeing these photos can lead to inaccurate beliefs about how common different kinds of photos are (for example, overestimating the frequency of high-quality, artistic photos and underestimating the frequency of “poor-quality” cameraphone photos). This can discourage participation through perceptions of the norms for the network or the community.

On the other hand, seeing photos with so many comments or views may lead to overestimating how many comments one is likely to get on one’s own photo; this can result in disappointment following participation.

Activity from a user’s closest friends?

Assume that activity from close friends is more likely to be relevant and interesting. It might even be more likely to prompt participation, particularly in the form of comments and replies. But it can also bias judgments of likely audience: all those people I don’t know so well are harder to bring to mind as is, but if they don’t appear much in the activity stream for my network, I’m less likely to consider them when creating my content. This could lead to greater self-disclosure, bad privacy experiences, poor identity management, and eventual reduction in participation.

References

Gerbner, G., Gross, L., Morgan, M., & Signorielli, N. (1980). The “Mainstreaming” of America: Violence Profile No. 11. Journal of Communication, 30(3), 10-29.

Kahneman, D., & Tversky, A. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5, 207-232.

Williams, D. (2006). Virtual Cultivation: Online Worlds, Offline Perceptions. Journal of Communication, 56, 69-87.