EXAMPLE: EHARMONY

Consider eHarmony, the online dating service that uses some highly sophisticated statistical methods for matching people up, with the express goal of long-term compatibility.

From The Atlantic: How do I love thee?

“We’re using science in an area most people think of as inherently unscientific,” Gonzaga said. So far, the data are promising: a recent Harris Interactive poll found that between September of 2004 and September of 2005, eHarmony facilitated the marriages of more than 33,000 members—an average of forty-six marriages a day. And a 2004 in-house study of nearly 300 married couples showed that people who met through eHarmony report more marital satisfaction than those who met by other means. The company is now replicating that study in a larger sample.

“We have massive amounts of data!” Warren said. “Twelve thousand new people a day taking a 436-item questionnaire! Ultimately, our dream is to have the biggest group of relationship psychologists in the country. It’s so easy to get people excited about coming here. We’ve got more data than they could collect in a thousand years.”

The stength of eHarmony, and what makes it so popular and apparently successful, is the sheer amount of data they have collected, and their theoretical models of relationships that can mine the data for compatibility results. They claim to be using science to build relationships (contrast with chemistry.com, which basically uses a suped up Myers-Briggs test).

Question: who is responsible for the resulting pairs suggested by the system?

Consider: The statistical models are the result of lots of r&d from some rather prominent academics and experts in this field of psychology.

None of the scientists responsible for building those models (or, for that matter, any of the programmers and engineers responsible for implementing the model) directly influence the resulting suggestion from the statistical analysis.

It seems at least intuitively plausible to say that the machine (the models and so forth implemented in actual computer systems) is the one making the suggestion of compatibility, questions of responsibility aside. That is, it is the machine who produces matches from the data, and nothing but the machine could have produced those matches.

Common practice suggests that the one who makes a suggestion is often blamed (if not held ‘responsible’ in a robust sense) for suggestions that go awry. “Let’s go to the Esquire!” (20 minutes later) “Who suggested we come here?”

I would guess that this common practice flows over into the eHarmony situation. On a blind date going poorly: “they picked a bad match”. Note ‘they’ here is ambigious between eHarmony qua company, or the scientists modeling the relationship data, or the system that actually produced the match. That the first two possibilities represent different kinds of agents (corporate vs institutional) suggests it is at least an open possibility that other deviant agents (like the computational systems producing matches) are open to blame.

Taken together, this seems to suggest that it is at least within common and intuitive practice to see the system as open to blame in making false matches.

I take it, though please correct me if I’m wrong, that blame can in most cases be separated from responsibility. Cheney might be the one to blame in the shooting, in the sense that he was the one who pulled the trigger, and yet might not be responsible, in the sense that he deserves retribution, or any corrective or preventative measures. In other words, I take it that responsibility adds to mere blame at least the idea that steps should be taken to correct the fault (and pick your own theory of justice to fill in the account of ‘correct’ here).

In the case of eHarmony , and all such statistical, information-based models, the results DO get better over time. The more people that sign up for the program and fill out the questionaire, the more data they have to model, and the more accurate the results. Perhaps there are upper limits to the accuracy of the model, but increasing information tends to approach that limit.

Perhaps this is not true responsibility (cf Kant on acting in accordance with a duty). But note that these corrective measures both increase the reliability, and by extenstion the trustworthiness of the system. In other words, whether or not someone or something is responsible for the match, people do tend to depend on these models. Both epistemic and ethical weight is put on the results eHarmony produces. And yet, we seem to lack any good moral agent around to support that weight.