It's not the 'clicking the box', it's the way the mouse moves across the screen that Google use to identify a bot or not.
FWIW I disagree with Rev. I've been up north for the last couple of days seeing a well known company/customer (e-tailer in computing and tech), and one of the subjects that came up was bots, as we have a product in our portfolio that detects and can manage them. BAD captchas are easy to overcome with automation. For any decent captcha or bot-manager it's the automation that gives a bot away.
A story; (and this is a real life example we did for an airline), the airline in question had a problem with bots scrapping their site for prices, seat availability and so-on. They implemented a (bad) captcha to try and combat the bot, but the bot just mutated and was taught how to evade the captcha manually. However, the bot was using a mobile phone user-agent when it was scrapping, so when we inlined our product the bot manager was expecting to see phone 'style' heuristics - that is, we capture sensor data from the 'phone' via the browser (we inject js beacons), so we expect it to be moving as the user types, and not sitting flat on the table and so on. As the bot wasn't providing good sensor data - moreover it was the same every time - the bot is detected and managed. And when I say managed I don't mean blocked - the best thing to do with bots like that is to feed them false data (that airline ticket is £9,000 and so on)
Anyway, long story short, some captchas are bad and some are good. The Google one is ok as captcha's go.