While OkCupid plays with your heart, researchers are in love with the data

For most companies, like OkCupid, these kinds of experiments are about business — it’s about optimizing products and profits in a space where startups are quickly trying to figure out what works and what doesn’t. This kind of experimentation is also not new — mandatory standardized exams like the SAT have always included an experimental section used to try out fresh questions.

What is new, however, is the scale of data collection and the insights it could potentially bring. The data stores of companies like Google, JawBone or Uber are also potentially treasure troves of fascinating new information for academics and researchers.

As more and more daily activities are recorded by our increasingly ever-present technology, we’re potentially creating the most accurate, long-term datasets that have ever existed. In a given day, data centers from phone or browser-based services can record what kind of food you eat (if you order online or enter it into an app yourself), who your close friends are or who you’re attracted to. Increasingly, so-called ‘Internet of Things’ devices will do this data collection; sensors on our bodies like FitBit that record when and where we move and how we actually behave.

As a result, both academics and non-academics are looking at private companies’ users to study the world. And in the case of non-academics, it might come without standard institutional oversight.

Linguistics is one example of an area where modern data collection allows for unprecedented research opportunities. For instance, what if you could record everything people said to each other and analyzed word choice, grammar and tone?

Nuance Technologies, the company that made T9, the predictive text system used on mobile phones, and now the Android Swype platform, has an opt-in program that allows users to share their text messages with the company in order to study usage and receive more up-to-date word suggestions. The platform has even detected small variations in how different groups of people communicate, such as people texting using little-used Native American languages.

If you’re a linguist today, this is where the interesting data is. As opposed to studying the English language through looking at books and newspapers — which might include jargon or be overly formal in style — you could analyze how people are actually talking.

“Absolutely these consumer companies that experiment with a lot of text data are attractive to researchers of all kinds,” said Lynn Cherny, a data analysis consultant and former Stanford researcher. “I know several people who have taken a job at Facebook because it means they are able to work on interesting problems at a huge scale.”

But with private sector industries from health to craft beers collecting new and more personal types of consumer data, there is a growing question about how much oversight such “experiments” receive or need.

In academic settings, experiments and studies usually must pass an institutional review board (IRB). IRBs are strict procedures meant to ensure the privacy of participating subjects. One of the principles is known as informed consent: a subject must be informed of everything the researcher is doing and any possible outcomes. If that doesn’t happen, their consent isn’t valid.

Holding web-based services to the same standard, however, is problematic, at least on the business end. If you’re trying to build a company, Cherny said, “how do you get someone to still sign up and participate in your business and not scare them away with lots and lots of potential horror stories about what could  happen?”

Facebook says its users opt-in to its experiments by agreeing to its Terms of Service. The reality is that no one reads Terms of Service or Privacy Policies. One study found that if someone were to read all privacy policies the average person sees in a year, it would take 244 hours, or roughly ten days, and that’s without breaks to eat, drink or sleep.

And is making your data available for experimentation as a condition of using a useful service really the same thing as opting in?

Even if consumers agree to a company’s privacy policies, when should a company be more forthcoming about experimenting with their users? If the tech industry sees all experimentation as essential, should there be a line between testing different shades of blue on your homepage or curating news feeds — things some users might expect — and engineering social hookups, which at the very least seem more personal and consequential? And if researchers want the unprecedented insights into human behavior, how can that work in a way that doesn’t leave a bad taste in peoples’ mouths?

For Cherny it comes down to respecting autonomy to a degree and understanding users’ expectations, even if they had technically “agreed” to a set of rules and permissions when they signed up for the service. “It’s about the morals of design. Are you going to work with users or use what they’re doing against them and lose their trust?”

Lam Vo contributed to this report.