Centamental: Private Conversations Aren’t As Private As You Think

1 8 Loading
centamental-big-brother

Centamental turns overheard audio into marketable statistics about how you feel.

Advertisers already pay big bucks for information about your search habits, your emails, and your tweets. In the near future they may be listening in on your conversations with your friends as well. theGOOD, a Portland based interactive think tank, has developed a prototype program for eavesdropping on a discussion and pulling out the important opinions. Called Centamental, this proof of concept platform translates casual remarks you make, say in a store, into valuable data that companies can use to understand how average people feel about their products. I talked with theGOOD co-founder Chris Teso and CTO Shaun Tinney about the potential and the obvious concerns that surround an eavesdropping program bent on understanding how we feel. Watch their commercial for the project in the video below. Many corporations claim to care what you really think…now they have a chance to know for sure.

What exactly will Centamental do? Give the program some audio – either from a hidden microphone, or a recording from a phone conversation, an old movie, etc – and it strips out all the pertinent data about how people feel about things. So, if you said, “I love cheeseburgers from this local burger shack. Big chain burgers aren’t nearly as good.” Then Centamental would spew out something like “+cheese +burgers +cheeseburgers +local food -big chain restaurants”. Where the ‘+’ and ‘-‘ marks (Centamental actually uses hearts and other symbols) indicate positive and negative opinions. The identifiable info about the speaker is tossed aside, only the sentiment remains. Give Centamental enough audio from enough places and you’d get a better idea of how people felt about…well, nearly anything.

When I first heard about Centamental I thought it was too stereotypically Big Brother to be true. Yet speaking with Teso and Tinney made me quickly realize two things: 1) Analyzing public remarks for sentiments about corporate products is already well entrenched in our emergent internet culture. 2) The development of a technology like Centamental was guaranteed to happen sometime.

Companies are already analyzing our online activities to better market products to us. Gmail anonymously sifts through your writing, and nearly all search engines do the same. Now, businesses are looking to twitter feeds to understand public sentiment. TwitterSentiment and TweetFeel sift through thousands (if not millions) of public comments on Twitter and gather the positive and negative opinions about a topic. The mining of the tweetsphere for data is already well established.

Centamental is very similar to these online products, only it uses real world audio. From a technical point of view, however, this isn’t that big of a change. Speech to text technology easily converts audio into text. From there you just sift through large piles of text comments looking for positive, negative, or neutral remarks about things you care about.

In fact, the technology behind Centamental was almost begging to be put together. Teso and Tinney explain that the proof of concept product up on the Centamental site is actually more of a hack at this point than a new creation. They use Ribbit to collect audio, this gets sent to a Google Voice number which converts speech to text, which is then emailed to an address where several APIs are applied that strip out the important data. From collection to sentiment takes around 30 seconds. It sounds pretty simple, but Teso thinks that, given enough audio from the right places, even this basic prototype would be able to answer complex questions like “What do people in the Pacific Northwest think about battery life for the iPhone?”

centamental

Sample sentimental analysis from the Centamental site.

theGOOD has made a habit of innovating on growing trends in the marketplace. They created sellsimp.ly a photo-based market (think Craigslist meets flickr), and theGOOD.uploadr (which makes uploading photos to flickr easier, especially when dealing with meta-data). Centamental, however, is the first that’s pulled them into the national spotlight.

According to Teso, Centamental’s already garnered the attention of half a dozen brands looking to possibly use the new technology. The next step in development would be to improve the analysis process so that Centamental can handle real world language. Sarcasm, slang, and stuttering – the sentiment analysis program of the future will need to handle them all.

Where might we see Centamental in the future? Potentially everywhere, and that’s the obvious privacy concern. Stores could mic their aisles, collecting data about what customers thought about products as they looked at them. Anyone could use directional microphones to record conversations in public places (bus stops, parks, stadiums) to get a better understanding of what ‘people on the street’ felt. In fact, anywhere you could collect audio you could collect sentiment.

I don’t think we’ll see this kind of technology spread so far so soon, however. The low hanging fruit is customer service hotlines. Most already warn you that “your call may be recorded for training purposes”. Many people are calling to complain or request help – they want their sentiments to be noticed.

Teso, who describes himself as a ‘hardcore libertarian’, seemed to think that even the larger ‘listening to conversations in public spaces’ application would be a net benefit for customers. Commenters would be both anonymous and oblivious, which is something that focus groups and satisfaction surveys simply can’t offer. Companies would be able to get information directly from consumers. Centamental’s style of analysis would be unique – and could possibly serve as a means to gather opinions that would otherwise be unheard.

Of course, there are reasons why many of us don’t take our private conversations and post them to company websites. Even with identifiable data stripped out, there’s something very eerie about people collecting information on what you are saying. The potential for abuse, either from governments or businesses, seems very great. But it’s the widespread collection of audio without people’s consent that is really threatening. Sifting through such recordings for anonymous sentimental analysis is one of the milder possible applications.

In any case, Centamental may seem ominous at first glance, but it’s also too ingenious not to be applied. Data mining is becoming something of a social science, and its demand is increasing as companies fight to market to customers successfully. Translating real world conversations into statistics that businesses could act on is a seductive possibility. Centamental may still be in the prototype phase, but I guarantee you that we’ll see technology like this in use soon. Or, really, I guess we won’t ‘see’ it at all.

[image credits: Centamental]

[source: Chris Teso, Shaun Tinney, Centamental, theGOOD]

Discussion — 8 Responses

  • ferricoxide October 1, 2010 on 3:55 pm

    One really hopes that technologies such as these will be designed to understand “tone”, sarcasm and irony….

  • Joey1058 October 2, 2010 on 12:08 pm

    This is a bit unnerving. WalMart has a policy of real time monitoring it’s customers now. Logically, it’s a security practice. But what if you get the local kids in on a weekend, and one out of boredom says “God, I hate this place. They never have what I want. I wish I could blow it up”. This is an absolutely normal conversation between peers. But security hears this, and the FBI is outside waiting for this poor kid who only wanted a pair of Nikes. Micromanaging your customer’s desires in this manner is just over the top. Why not just follow us around with a company sales rep directly and eliminate the ad agency if it really has to come down to this.

  • David October 2, 2010 on 4:11 pm

    Oh good, the identifiers are stripped out. This assumes what is and isn’t an identifier is easily identifiable! And how is the rest “tossed out”? Is the speech recorded and then later fed through this system? Or does it stream directly into this system? I tend to think it is being recorded first in which case, it might be that whole conversations are recorded and are still retrievable.

    Is this guy so “hardcore libertarian” that he would refuse to obey a subpoena for information he has gathered in this way? Somehow I think his libertarianism might extend only as far as his concerns for his own bank account and not to me and my privacy.

    • Chris Teso David October 2, 2010 on 9:01 pm

      David,

      Your skepticism is not unfounded. I’d be concerned too. We were concerned when we built the application, and continue to be.

      As the article states, this type of technology is inevitable. We’re trying to get a jump on the game and ensure it’s used logically and responsibly. It is a challenge to strip out identifiers. That said, there are literally no identifiers. You’re right, we record first then turn that audio into topical text. After it’s translated we dump all the audio and save only the topical text for sentiment analysis. The audio is not retrievable. There is no way to associate text to an individual. Even if a persons name is said, there is no way to identifiably prove that person was the one who said it.

      Thanks for the discourse.

      Chris

      • Anonymous Chris Teso October 21, 2010 on 9:51 am

        You are grievously mistaken or you are trying to mislead others when you say that there is no way to associate text to an individual, given your setup. There is a timestamp (obviously) and a location (obviously). From there on, it’s easy to do a bit of data mining.

        Let’s take your “Apple” scenario.

        Take one: Apple installs these mikes in their own stores. How hard is it for them to match actual purchases (personally identifiable due to credit card) to utterances by clients in their stores?

        Take two: You have a network of mikes in public places. An Apple user is running something like foursquare or any other public presence service (as many, many Apple users are wont to do)? Apple would have no trouble at all finding out (based on the locations of your microphones and timestamps on the produced text) who said what when where.

        Take three: Ok, your consumer, (aka the target) does not directly broadcast location. How to you find it? Well, how about Wi-Fi? Google has mapped millions of Wi-Fi hotspots, all data which Apple is already using to provide location info indoors.

        Queries to services such as the iPhone maps have the originating IP matched against a list of known hotspots. Location, timestamp, device type, even, via NDP or ARP. One of the locations is bound to be in or near the user’s home. At which point, you have them by the short and curlies and you can again start to backtrack through the data to see who said what when where.

  • Jeremy October 3, 2010 on 2:10 am

    I’m not sure how to word this, but that part of this stuff that irks me is that it takes on exploitative feel.

    People are making money off what I search for, people are making money off of what I e-mail, and now, people will make money off what I say. What am I getting out of this? I remember the good old days when companies -asked- me for my opinion and then -compensated- me for it.

    By all means, as long as its stripped of personal information, record me. But I deserve a cut.

  • Jdiversen October 3, 2010 on 2:23 am

    I would assume everyone has seen this: http://www.youtube.com/watch?v=Xtuxax8Dtk4

    Spoof feature from Google whispering ads directly into your ear as you have your conversations.

  • Elvenrunelord October 4, 2010 on 7:35 am

    Data mining is becoming something of a social science, and its demand is increasing as companies fight to market to customers successfully…….

    If a company has a product that people really want and really does service a need then they will not need all this fancy marketing. This stuff is only for the “Pet Rock” type crap that is pointless other than as a profit snipe for capitalists.