The Ghost in the Machine: Google Gamifies Machine Learning
David: A series of spells and specters (our mutual travel and presentation schedule) conspired to delay our usual conversation, Mike, but I’m happy to get a chance to speak with you this week.
Mike: It’s a good thing we are squeezing it in. I have a brutal travel schedule coming up so it might not be the last delayed conversation in the series.
David: Fair enough. We will sneak these in when we can!
As for our Halloween topic, a spooky good SEO, Scott Hendison, tweeted a link over the weekend that I found fascinating: https://crowdsource.google.com. Even for those of us who are used to these kinds of initiatives coming from Google, it’s the most brazen public effort we’ve seen to train their machine learning algorithm via user contributions across a whole range of data types.
Mike: It is certainly brazen. There is NO attempt to bury this as an activity within some other program like their Captcha. It’s a gamification of their ML plain and simple, and if I know Google, the reward will be either insignificant or worse: a discount on some “premium product” (i.e., an ad).
David: I really don’t understand the motivation on the part of users who get wrapped up in these fairly transparent efforts. Are they actually convinced by Google’s justification (“helping us create inclusive products that everyone can use and enjoy”)?
“Altruism” where the beneficiary is one of the world’s largest multinational corporations seems closer to a personality disorder than a healthy behavior.
That said, we’ve seen Google execute on this script really well with its Local Guides program. And I have no doubt they’ll be successful with this program as well provided it is sufficiently well marketed.
In fact, I was surprised to see that the Local Guides program wasn’t linked from this page — arguably their most successful initiative yet in terms of crowdsourcing a better product.
Mike: For me at least, the Local Guides program dovetails with the possibility of improvement for a client fighting spam. But this new effort just helps make the machine that is Google spookier yet.
They get such amazing participation in the Local Guide Game, there is no reason to spoil it with more obvious blatant self-interest.
David: I certainly understand and endorse the motivation of industry peers maximizing their Local Guide status such that edits or spam-flags on behalf of clients are approved faster or with more predictability. There’s a very obvious, rational commercial benefit to building that status.
Mike: I would add though that many folks participate in the Local Guides program because they take pride in their community and they think having a more accurate Google helps.
David: Community-building is a noble motive, to be sure, but I don’t think that’s the motivation for the overwhelming majority of Local Guides.
It’s much closer to the directionally narcissistic motivation of the Yelp Elite or Trip Advisor TripCollective, where there’s a certain status conferred based on your “contributions” being viewed by an ever-increasing number of users.
Mike: Well, getting a gazillion views on an image of a diner that I posted to Google Maps does provide me with some weird endorphin rush. But for me at least, a sense of community is a reason (however misguided) that I contribute to Google Maps. It would appear that the motivations for “free” labor on behalf of Google are varied.
David: Look at you! A Level 7! And over 10,000 points!
In all seriousness, this new Crowdsource program seems to be relying on that same “contribution impact” but where the intrinsic value is far less obvious and far more abstract — beyond the value of making Google’s products better.
Mike: Exactly. This, unlike the Local Guides program, seems totally self-serving for Google.
David: Tread carefully, Mike. You know that “self-serving” is a loaded term in Google’s circles these days!
Mike: Very funny.
This program, not being connected to something real in our lives like the store down the street, will need to offer up something beyond marketing and a simple leaderboard status. Don’t you think?
Although, humans are a weird species, and they have been motivated by less.
David: Maybe…the leaderboard may be enough.
The broader Crowdsource effort seems part and parcel of the topic we covered in our last conversation—the increasing importance of photos (and entity relationships in photos) as a new paradigm for local search.
Google has a much easier, more closed feedback loop with respect to training its voice recognition algorithm (not to mention a huge installed base of users), but with images they need help from the easily gratified masses.
Mike: It is interesting that in the training there are four areas that are image-related and four that are language-related. I assume the allure of having a billion people train your algorithm is irresistible to Google.
David: No doubt. This program also highlights Google’s fundamental commitment to the “wisdom of the crowds.” (Insert classic presentation line that the “My” in Google My Business is a complete misnomer here.)
The value of spamming the particular endpoints of the Crowdsource program is not obvious, at least to me. So, maybe Google’s faith in the results of this initiative is more well placed than its faith in similar initiatives that intersect with local search, such as Q&A and Community Edits.
Mike: That being said, I see NO REASON why any human being would participate in this program as it seems to be structured. Maybe humanity will surprise both us AND Google and go on strike for this one.