Do you know you are training Google self-driving cars so they don’t kill people? Yes, by solving captcha

When Google's driverless cars roll out on busy streets after their ongoing trials end, and when they don't kill a pedestrian, or hit a cyclist, or run over a fire hose, you can thank yourself. That is because you have taught -- and you are teaching -- these Google cars now called Waymo cars, to see the roads and everything on them. Yes, you. As much as these cars are a testament to the ingenuity of Google's hardware and software engineers, they are also going to be a product of the wisdom of the crowd that millions online users, and they include you, impart to Google's systems daily.

Confused? Don't be. Every time you solve Google-- supplied Captcha -- and Google supplies this Captcha on thousands of popular websites -- you teach Google's driverless cars. Of course, you don't probably know that. But have you ever wondered why for the last two odd years, increasingly the Captcha that you are solving on file-sharing websites involves identifying cars in tiny pixelated images, or recognising fire hose, or store fronts, or bicycles, or buses? This is not random. It's by design because Google wants you to identify these images so that its artificial intelligent systems can learn from your knowledge.

Let's take a step back here before talking about what Captcha has become and how Google uses it. Let's first talk of what Captcha used to be and how Google used it earlier.

A brief history of captcha

Captcha was not always used the way we remember it now. It was introduced in early 2000s by Carnegie Mellon University researchers, lead by computer scientist researcher Luis Von Ahn or Big Lou, and the idea was to filter out spam bots that were pretending to be real people.

The idea was that robots would not be able to read a bunch of letters and numbers that were printed in a distorted format, but humans would. Bots are good at logic. Humans are good at abstract, and printed words and numbers, or in other words images, can be considered abstract. So researchers devised a program that would display a bunch of garbled codes for humans to type in and gain access to a website.

Captcha code was a hit and soon it was being used everywhere on the world wide web for keeping bots at bay. You would probably still find them on IRCTC's online portal.

But then in 2006 Ahn had this idea of using captcha for deciphering old smeared text in archival texts. "So we asked, 'Can we do something useful with this time?'," he told the New York Times in an interview. And thus recaptcha was born.

So, from 2006 to 2009, which is when Ahn's start-up was acquired by Google, when millions of internet users across the globe were solving recaptcha codes, they were in a way working for the New York Times and helping it decipher the old smeared text -- all for free!

Google used it for Books Project

The shear ingenuity of this technology impressed Google, which on acquiring it in 2009 used it for its book scanning project. This is the reason why, if you recall, you were solving the text and number puzzles while solving captcha. Enter the word that you see, the website would ask you. When you entered the word, it was matched with the image, with the printed letters, and Google's AI systems learned which printed letter or number corresponded to which digital letter and number.

At its basic, machine learning and AI is all about identifying patterns. This requires data, huge amount of data that no single technology company can create on its own. To get this data, Google decided to use the millions of clicks that online users make daily. By the end of 2011, the captcha had helped Google digitise millions of books.

With book scanning project nearing completion, Google turned sights on giving its AI and machine learning systems ability to identify images. You must have remembered news about Google's AI learning to identify cats on its own. Now, this AI used in Google Lens is so good that it can identify a Doberman from a Golden Retriever.

Google's image identifying AI was also helped in parts by millions of web users. In 2012, Google changed captcha as it was popular back in the day. Google started including images in its recaptcha code. Initially, it started with distorted text in the images, which online users were asked to identify to prove that they were humans. Then in 2014, Google introduced a modified version of recaptcha -- no captcha ReCaptcha-- to replace the old recaptcha codes. These images were essentially random images and internet users were asked to identify similar looking images, say images of a cat or images of a tree, to prove that they were real people.

In the following years, with Google's driverless cars becoming an important project for the company, the 'no captcha ReCaptcha' gradually started including image-based puzzles that would help AI and machine learning systems behind these cars to learn more about the roads on which would be driving.

The important bit about the AI systems is that they need vast amount of data to practice and learn. The more the data, the better and more accurate they get. Google has a dedicated team working on improving the accuracy of the AI systems running at the core of Waymo cars. However, even, those teams cannot create the amount of data, or co-related data, that millions of online users can generate. This co-related data can then be used for training its AI systems for things ranging from identifying street signs to identifying an animal and from detecting the season of year to detecting the object in front. This in turn can be used for improving the Waymo car's decision making capabilities to access and respond in various real-life situations, like how it has to react in case there's a cow in its way or if a tree blocks its way.

May be a thank you note, Google?

Google is not the only company utilises user-generated data to train its algorithms and machine learning systems. Companies like Facebook and Aamzon do that too. When you upload an image on Facebook, that image is used by Facebook's systems for training purposes. This is the reason why recently when 10 Years Old Challenge became viral many said that it was giving Facebook the kind of data that its machine learning systems could use to understand ageing. Facebook, though, has denied that it was behind the trend, calling it spontaneous and one popularised by users, but it hasn't explicitly said that it won't use the images people have uploaded on Instagram to use its systems.

But even with this context, the way Google has used captcha has been a brilliant solution to a big problem that any big tech company faces: how to get relevant and good quality data.

The AI and machine learning systems are beating humans at games like Chess and Go, but humans still win at -- and will continue to win for considerable time -- at tasks where logic either fails or is very difficult to apply. Identifying a car in an extremely blurry image is one such task.

This is not to say that Google's Waymo cars are a project built by the millions of online users, or that its book projects is powered by the wisdom of crowds. But it is also undeniable that with the help of millions of online users who solved captchas -- and who still solve them daily -- Google's engineers have been able to train their AI and machine learning systems at a rapid pace and with minimal investment. So, yes every time you have solved a captcha you have helped Google's driverless cars get a tiny bit smarter.

You have done that for free, and most likely without realising. You have believed that by identifying bicycles in blurry images you are just trying to unlock the download for that song you want to hear. But no, you aren't just accessing the song. You are part of a big crowd-sourced system that is building the future. And while you are not paid for it, and won't be paid for it, may be Google should consider sending you a tiny thank you note.