Tell HN: Mechanical Turk is twenty years old today

Join the discussion

• Hacker News
• I was working for a marine biologist applying computer vision to the task of counting migrating river herring. The data set was lousy, with background movement, poor foreground/background separation, inconsistent lighting, and slow exposures that created blurry streaks. Moreover, the researchers had posted the frames to some citizen science platform and asked volunteers to tag each fish with a single point—sufficient for counting by hand, but basically useless for object detection.

  In desperation I turned to Mechanical Turk to have the labels redone as bounding boxes, with some amount of agreement between labelers. Even then, results were so-so, with many workers making essentially random labels. I had to take another pass, flipping rapidly through thousands of frames with low confidence scores, which gave me nausea not unlike seasickness.

  by rgovostes
• Using the Propser.com data set (a peer-to-peer lending market), I used MTurk to analyze the images of people applying for a loan. This was used in a finance research project with 3 University of Washington professors of Finance.

  The idea was that the Prosper data set contained all of the information that a lending officer would have, but they also had user-submitted pictures. We wanted to see if there was value in the information conveyed in the pictures. For example, if they had a puppy or a child in the picture, did this increase the probability that the loan would get funded? That sort of thing. It was a very fun project!

  Paper: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1343275

  by stevejb
• During the beta, the only consistent HITs were to identify an album based on a picture and 4 or 5 choices (if Im remembering correctly). These paid pretty well since the workforce volume was very low and the service was brand new. Well, I noticed that the image link contained an ASIN. So, I wrote a greasemonkey script that would look it up on Amazon and highlight the most likely correct answer. I then turned around and shared it with the forum I frequented. It became extremely popular and spread to other forums before we moved it to a private forum. The damage was already done though.

  People kept asking me to automate it, but I felt it was against the spirit of mTurk. So, another member would take my updates and add an auto-clicker. That lasted for a couple of weeks at most before the HIT volume dried up and very few would be released. I guess Amazon caught on to what was happening. But before that, several forum members made enough to get some high dollar items: laptops, speakers, etc. Eventually, I relented and created a wishlist. Thats how I ended up with the box sets for the first run of Futurama seasons.

  by goykasi
• We asked user to evaluate 300x300 pixel maps. Users were shown two image and had to decide which better matched the title we chose. Answers were something like "left", "right", "both same", "I don't know". Due to misconfiguration the images didn't load for users (they only loaded in our internal network). Still we got plenty of "left" and "right" answers. Random and unusable. Our own fault of course.
  by mtmail
• I’ve run millions of jobs on MTurk.

  For a major mall operator in the USA, we had an issue with tenants keeping their store hours in sync between the mall site and their own site. So we deployed MTurk workers in redundant multiples for each retail listing… 22k stores at the time, checked weekly from October through mid-January.

  Another use case.. figuring out whether a restaurant had OpenTable as an option. This also changes from time to time, so we’d check weekly via MTurk. 52 weeks a year across over 100 malls. Far fewer in quantity, think 2-300. But it’s still more work than you’d want to staff.

  A fun more nuanced use case: In retail mall listings, there’s typically a link to the retailer’s website. For GAP, no problem… it’s stable. But for random retailers (think kiosk operators), sometimes they’d lose their domain, which would then get forwarded to an adult site. The risk here is extremely high. So daily we would hit all retailer website links to determine if they contained adult or objectionable content. If flagged, we’d first send to MTurk for confirmation, then to client management for final determination. In the age of AI this would be very different, but the number of false positives was comical. Take a typical lingerie retailers and send it to a skin detection algorithm… you’d maybe be surprised how many common retailers have NSFW homepages.

  Now some pro tips I’ll leave you with.

  - Any job worth doing on mturk is worth paying a decent amount of money for.

  - never runs. Job 1 tile run it 3-5 times and then build a consensus algo on the results to get confidence

  - assume they will automate things you would not have assumed automated - And be ready to get some junk results at scale

  - think deeply on the flow and reduce the steps as much as possible.

  - similar to how I manage Ai now. Consider how you can prove they did the work if you needed a real human and not an automation.

  by social_quotient
• My wife had dozens, well probably over 100, handwritten recipes from a dead relative. They were pretty difficult to read. I scanned them and used mturk to have them transcribed.

  Most of the work was done by one person - i think she was a woman in the Midwest, it's been like 15-years so the details are hazy. A few recipes were transcribed by people overseas but they didn't stick at it. I had to reject only one transcription.

  I used mturk in some work projects too but those were boring and maybe also a little unethical (basically paying people 0.50 to give us all of their Facebook graph data, for example.)

  by comrade1234
• If there's any justice, a good number of comments will focus on the ethical nightmare MTurk turned out to be. Apologies to the people who worked on it, but it's fair and appropriate for observers to point out when someone has spent their time and energy creating something that is a net negative for the state of society. That's probably the case here.
  by frmersdog
• I used MTurk heavily in its hey-day for data annotation - it was an invaluable tool for collecting training data for large-scale research projects, I honestly have to credit it with enabling most of my early career triumphs. We labeled and classified hundreds of thousands of tweets, Facebook posts, news articles, YouTube videos - you name it. Sure, there were bad actors who gave us fake data, but with the right qualifications and timing checks, and if you assigned multiple Turkers (3-5) to each task, you could get very reliable results with high inter-rater reliability that matched that of experts. Wisdom of the crowd, or the law of averages, I suppose. Paying a living wage also helped - the community always got extremely excited when our HITs dropped and was very engaged, I loved getting thank yous and insightful clarifying questions in our inbox. For most of this kind of work, I now use AI and get comparable results, but back in the day, MTurk was pure magic if you knew how to use it to its full potential. Truthfully I really miss it - hitting a button to launch 50k HITs and seeing the results slowly pour in overnight (and frantically spot-checking it to make sure you weren't setting $20k on fire) was about as much of a rush as you can get in the social science research world.
  by pvankessel