Xtech 2006: Jeff Barr – Building Software With Human Intelligence

Amazon have released a number of APIs, but going to focus on the Mechanical Turk.

[He has ‘ground rules’ and says ‘please feel free to write about my talk’. Er… command and control anyone? Give me a break.]

[Lots of guff about Amazon.]

Artificial artificial intelligence.

The original Mechanical Turk was a chess playing robot. No one could figure out how it worked. Turns out there was a ‘chessmaster contortionist’ inside.

So the MTurk is powers by people. Machines can’t tell the difference between a table and a chair, whereas a person can do it immediately.

HIT – Human Intelligence Task.

Can check people have the appropriate skills, e.g. be able to speak Chinese or tell a chair from a table.

So you make a request, say what skills etc. are required, figure out the fee you’re willing to pay. Workers go to MTurk.com. 45 types of work, can filter by price or skills required etc. Transcriptions are very popular. E.g. CastingWords.com do transcriptions of podcasts. Question and answer kinds of things.

You decide if your workers get paid, but there are ratings on both sides, so you can rate the employers as well as the workers.

Software Developers

– can use APIs with this to include humans in their applications


– can get stuff done that humans need to do


– can make money

– new businesses feasible

Public use – massive scale image clean-up, i.e. which picture best represents the thing it’s of? Got Slash-Dotted to death. People did greasemonkey scripts to help them doing it. Had so many people doing this that they ran out of images. Had more workers than work for a while.

HIT Builder, helps you create your HIT (task thing).

Could use it for market research or surveys. E.g. wanted a survey for developers, so added some qualifications to weed out the non-developers by making people answer questions like ‘which of these four aren’t programming languages’.

Translations services.

Translate written transcripts to audio.

Image Den, photo editing and retouching, e.g. removing red-eye, cutting things up etc.

CastingWords, podcast transcription service.

Need an Amazon account to work, which requires a credit card, and that’s their way of trying to ensure no child labour.

[I think this could possibly have been an interesting talk, but it wasn’t. I like the idea of having APIs for something like MTurk, but this guy was really dull. I guess I could cut and paste from the back channel to spice things up a bit, but that might be mean.]

2 thoughts on “Xtech 2006: Jeff Barr – Building Software With Human Intelligence

  1. Hi Suw,

    Thanks for your comments. Sorry to hear that I came across as “dull.” I try to present all of the facts and to show cool demos built by members of our developer community, but I really try to put the focus on the APIs and on the demos, not to draw attention to the speaker. It wouldn’t be hard to be dramatic or theatrical, but that’s not my strategy. I’d rather my audience walk away with “wow, that’ some cool stuff I saw from some guy” than “Wow, Jeff’s an amazing speaker but I don’t know what he talked about.”

    Make sense?

  2. Hi Jeff,

    Yes, your comment does make sense. I’d like, if I may, to take this opportunity to give you a little feedback on why I thought your presentation was dull. You can take or leave any of it, but my intention is to be constructive.

    There were a few things that stopped me from fully engaging with your presentation. One of the main ones was that you talk really fast. I am a prodigious note-taker (as you may have noticed from this blog!) and I can type as fast as most people speak. I couldn’t keep up with you, though, and even when I wasn’t typing, your words turned into a stream of sound that washed over me. It’s easy to talk too fast, so if you slow down to the point where you feel you’re talking too slow, you’re probably just about perfect.

    The second major barrier was that you gave us too much background. The first 15-20 mins of your talk was either biographical info about you, or history about Amazon. You needn’t have given us either: we can find out about you from your bio in the programme or online; and we all know what Amazon is and unless your history is directly relevant to the topic at hand, leave it out. We can go find that online too if we want.

    The third thing was that whilst we all had a suspicion that the Mechanical Turk is capable of doing some really cool stuff, by the time we got to the end of your talk we still weren’t really sure what. You gave us glimpses, but you needed to slow down and go through one or two examples from beginning to end, illustrating with screenshots or demos where possible.

    I also felt that you missed out on fully exploring the implications of having an API for what are basically human-lead tasks. That in itself is a fascinating concept, being able to tap into real people to do tasks that can’t be automated.

    I really appreciate you coming here and leaving a comment. I hope you will take my response in the spirit with which it is intended, as some friendly advice.

Comments are closed.