Tuesday, March 16, 2010

Crowdsourcing the Singularity

“Come on. Learn, goddammit.”
- David (Mathew Broderick) to Joshua, War Games, 1983

I’m reading the Singularity is Near by Ray Kurzweil. For those of you who’ve never heard of the Singularity, “it’s a future period during which the pace of technological change will be so rapid, its impact so deep, that human life will be irreversibly transformed.”

According to Kurzweil, we’re currently living in the 4th Epoch and are rapidly approaching the 5th Epoch, which is the Epoch where all of the Singularity goodness begins.

"Looking ahead several decades, the Singularity will begin with the fifth epoch. It will result from the merger of the vast knowledge embedded in our own brains with the vastly greater capacity, speed, and knowledge-sharing ability of our technology. The fifth epoch will enable our human-machine civilization to transcend the human brain's limitations of a mere hundred trillion extremely slow connections."

- Ray Kurzweil, The Singularity is Near
The singularity will begin when we build intelligent machines like Joshua from War Games. Combining human understanding with silicon’s raw processing power will result in an explosion of innovation and progress. This argument makes sense. However, I think Kurzweil may have missed an intermediate step between the 4th Epoch (today) and the 5th Epoch (the beginning of the Singularity). Let’s call it Epoch 4.5.

Epoch 4.5 is when our brains’ “mere hundred trillion extremely slow connections” will help computers transcend early stage artificial intelligence. We’ll do this by greatly expanding crowd-aided artificial intelligence, resulting in computers that provide a Joshua-like user experience. The crowd will support, train and augment these computers’ intelligence in real-time on a vast scale.

On a small scale, Mechanical Turk is already doing this, which is why Amazon’s tagline is - “artificial, artificial intelligence.” Here are some other real world examples:

  • reCaptcha uses CAPTCHAs to digitize books. Every time you enter a CAPTHCA to register for a website or buy something, you’re helping a computer understand text.
  • Google has created an ESP game where two people look at a picture and type in words that describe the picture. When the words of two players match, the players get points. Google then uses those words to index the images for Google Image Search. -
As crowdsourcing becomes more prevalent, larger pools of on-demand workers will be available. The computers of Epoch 4.5 will leverage this ever present crowd to address their intellectual weaknesses.

As a small thought experiment consider an advanced Roomba. Let’s call this Roomba SuperRoomba. SuperRoomba’s AI isn’t much better than Roomba’s; however, SuperRoomba is connected to the crowd. When SuperRoomba’s AI is stumped, it queries the crowd for help. Here are a few examples:
  • One of my former Long Island neighbors orders a SuperRoomba. SuperRoomba carries the latest speech recognition software, but he still can be confused by a nasally, Long Island accent. When presented with a confusing word, he sends the offending audio clip to a worker in the crowd. The worker corrects SuperRoomba’s understanding, training him to recognize “Dawg” as “Dog”
  • My old neighbor outfitted his SuperRoomba with a security camera. While video analytics can detect motion and a small subset of activities, it can’t tell the difference between a real threat and my neighbor’s dog. While patrolling, if SuperRoomba encounters sudden movement, he sends a snapshot to a worker, who evaluates the threat level.
Where will this crowd come from? To make Epoch 4.5 a reality, millions of workers will need to be available to serve these requests in real time. The best source for this labor is the same labor pool that currently staff factories. Factories using tens of thousands of workers construct complex products in mass every day. In a similar manner, online information factories could process computer requests. There are certainly enough underemployed people in the world to staff these virtual factories. According to the UN, 1.4 billion people make less than $1.25 a day. Most of these people work in terrible conditions and would jump at the chance to work in a safe office environment.

SamaSource, a non-profit that brings crowdsourcing work to African refugees, has already shown a demand for this work, as well as the ability to reach these workers. Imagine the impact if an additional 1.4 billion brains came online to guide, correct and train the next generation of artificial intelligence. We could provide safe and decent-paying jobs to the world’s poorest and simultaneously bring about a revolution in computer intelligence.

Friday, February 5, 2010

Is That a Bomb in Your Pants or are You Just Happy To See Me?

There was a lot of uproar over the "Underpants Bomber" in the days after Christmas. The talking heads on the Sunday morning shows pounded the TSA hard and wanted to know how this guy got through. It isn't possible to point a finger at one reason. Tons of small mistakes led to the security breach. My sense is that most people feel airport security is ineffective.

Recent research backs that feeling up. A test at O'Hare found that TSA agents only found 60-75% of planted guns, knives and bombs (all fake of course). It's nearly 10 years after 9/11 and we're still not very good at airport security. Why? Well, according to research by Harvard Prof. Jeremy Wolfe, it turns out that human brains really suck at visually searching for rare events. When an item shows up only 2% of the time, which is the frequency of contraband in luggage, people miss the item 30% of the time. However, if the item appears 50% of the time, people miss the item 7% of the time.

One solution to the problem would be to insert staged contraband images into the image feeds so that 50% of the images contain a gun, knife or bomb. If TSA had done this last year, they would have intercepted 4,504,455 more prohibited items (most of it probably 60z bottles of hair gel).

At ReTel, we've experimented with similar techniques to improve the visual search capabilities of our auditors. Understanding the way the brain sees is a big part of successfully leveraging the crowd to analyze video.

Thursday, January 21, 2010

Crowdsourcing and San Fran Awesomeness

Last night I went the Crowdsourcing Work Meetup event in my new hometown - San Francisco. The energy, passion, and excitement in this city is incredible. Chicago never could have staged this event. I feel really lucky to be here. ReTel is at the geographical and ideological epicenter of a brand new technology that is going to change the world. My co-founder, Adam Rodnitzky said it best - "This is turning into a movement." He's right. The wave is building and it is going to change lives all over the world.

Lucus Biewald, the CEO of Crowdflower, moderated and organized the event. He's a great guy, and if you're interested in crowdsourcing, you should read his blog - http://blog.doloreslabs.com/ . The other speakers were great as well. Here is a list:

  • Aaron Koblin, a local artist, who created The Sheep Market, Bicycle Built for Two Thousand and Ten Thousand Cents projects.
  • Leila Janah, a social entrepreneur who runs Samasource, our fantastic non-profit partner.
  • Sharon Chiarella, VP at Amazon in charge of Mechanical Turk.
  • Panos Ipeirotis NYU Stern school professor who writes one of my favorite blogs on crowdsourcing and Mechanical Turk.

  • If you want too see an amazing visualization of the power of crowdsourcing, go to Aaron Koblin's website - http://www.tenthousandcents.com/. It's amazing.

    If you want to get into the underlying coding and theory behind crowdsourcing, go to Panos Ipeirotis blog -http://behind-the-enemy-lines.blogspot.com.

    If you want to learn about how crowdsourcing is going to save millions from poverty, go to http://www.samasource.com/.

    Ok, I've got to go. I've got a board meeting this afternoon. :)