The Daily Technocrat

I already wrote about this new facial-recognition technology on the web called Riya. I started getting into a bit of a discussion on the Riya blog about possible uses for Riya beyond just photo albums, such as combining it with other web services like Mechanical Turk for distributed “Intelligence Engineering”.

Amazon’s Mechanical Turk is an interesting concept. These days we’re trying to use computer for a lot of things, but some things computers just can’t do. What Mechanical Turk does is pays people a small fee to do small tasks that require human logic. These tasks are then fed back into our computer programs, and the programs are more complete in the end.

One thing about Mechanical Turk that is annoying is that it doesn’t pay enough for your time. Now I don’t think people with jobs would consider the Mechanical Turk as a source of revenue, unless they’re really bored.

However, if we can take the mechanical turk (7 seconds per job, 3 seconds between jobs, 3 cents per job, 4 hours per day = $43.20 per day) and make it easier/less stressful so we can boost output and have a user work longer (since less brain power is needed), we might be able to make it a viable income…? (3 seconds per job, 2 seconds between jobs, 2 cents per job, 4 hours per day = $57.60) (or 6 hours @ $86.40)

Of course, there would need to be much more use of the MT API, as all of the jobs would soon be used up.

…but anyways, I’m sure that $432 per week for a 30-hour work week would be welcome somewhere on earth…

In addition to making MT for feasible as a paying job, I’m sure we can find a better use for MT. Instead of posting pictures so someone can have an accurate idea of what the front of Jon’s Bait Shop in Van Nuys, CA looks like, why not post jobs on MT that will actually help us?

If we can take the programming of an Artificial Intelligence program, say, for a fire-fighting robot, as post 100,000 pictures of fires-in-progress, we might be able to use MT to gather the intelligence of thousands of fire fighters and build a pretty good AI.

For another example, why not use MT to program the AI for the vehicles in the
Grand Challenge? We all know how to drive (most of us). If a computer could learn from 1 million our visually-based decisions and learn how to navigate the desert, at $.03 per MT job, $30,000 would be a small price to pay! Add in the functionality of Riya, and you would have the added ability not only to stay on the road automatically, but also be able to read signs and symbols (i.e. if Riya sees ‘Railroad Crossing’, pan camera to left and right, take another shot left and right 10 seconds later, and compare to see if train is coming)

This is just the beginning, any other ideas?