The scariest thing about that video is how easily they refer to Soylent as 'it', as in 'it cuts down your text', 'it fixes grammar errors', even though when Soylent is in use it's clearly a 'they'.
Or perhaps that distinction is as nonsensical as referring to the bundle of neurons in your head as 'they'...
On one level it's quite cool to see human intelligence pipelined as a service, but on another I worry that this may represent a new form of slavery with the turkers being paid at such a low level that other social problems may arise as a consequence.
While I'm sympathetic to this worry about crowdsourcing (which Jonathan Zittrain has been writing and speaking about for years), I'm still optimistic that the architecture of ecosystems like Mechanical Turk can protect people from exploitation. And hopefully the market will set fair compensation per task (although I don't think it has yet).
This would be really useful for programming, particularly refactoring. Imagine being able to quickly apply a relatively mechanical refactoring that still required human judgement (pretty common), and have other programmers do it for you.
I imagine it would work something like this:
- Select the files/directories/blocks of code that you want the refactoring applied to.
- Describe (in english) what the refactoring work should be, and do an example on one piece of the code.
- The work gets split (by file?) and distributed via a mechanical turk-like interface (extra points if it opens tasks in your favorite editor with syntax highlighting). As workers submit their work, the code is automatically built and tests are run (could be a callback to your local development environment, so you don't have to worry about shipping all your code up to the cloud and figuring out how to build it there).
Could be pretty awesome, particularly if you could enlist an army of freshman CS majors, bored programmers, and eager overseas engineers. I wonder if intellectual property concerns would be too high for people to actually use it though.
EDIT: If I still worked at Google, this would be a fun project to try internally as 20% project. There are lots of mechanical refactoring tasks that need to be done across large portions of the codebase, good build/test infrastructure to easily verify success/failure for a given task, and lots of bored/eager engineers that feed off of reward-based systems:)
Mechanical Turk has been the brute-force behind a lot of unique concepts, it seems. Really cool use of crowd-sourcing, and you've gotta love the name. "Soylent is people" indeed.
wow! At 4pm today I was giving a talk in my lab about how cool this work is.
What really impresses though is not that it works. It is the Find-Fix-Verify design pattern which tries to set some good practices in crowd-sourcing. For example in GUI/web engineering MVC is considered good practice.
Refference:
Bernstein, M.S., Little, G., Miller, R.C. et al. Soylent: A Word Processor with a Crowd Inside. UIST '10, ACM Press (2010)