Open CV could easily help in that area. Servo controlled stylus for input, OpenCV based watchdog against pre-programmed motions. Maybe mechanical turk or something to generate the reviews.
Perhaps you could fight it by doing statistical analysis on review patterns. If the device is only ever giving great and/or crappy ratings and the same group of devices rates the same group of apps the same way, you could detect this.
Edit: Of course, they could fight detection by having a queue of apps to review and an array of devices. Randomize the matrix and periodically retire a device to prevent it from gaining too much history. Since it's running in a machine it would likely be in pristine condition and capable of being resold as new. Heck, they could buy one, use it for a period of time then return it.
Perhaps you could fight it by doing statistical analysis on review patterns. If the device is only ever giving great and/or crappy ratings and the same group of devices rates the same group of apps the same way, you could detect this.
Edit: Of course, they could fight detection by having a queue of apps to review and an array of devices. Randomize the matrix and periodically retire a device to prevent it from gaining too much history. Since it's running in a machine it would likely be in pristine condition and capable of being resold as new. Heck, they could buy one, use it for a period of time then return it.