Tech

Google makes us train AI models with reCAPTCHA


Google offers dozens of services for different types of companies and users. Among its services is reCAPTCHA, which is used to protect websites from bots.

All users were subjected to reCAPTCHA while using the Internet. That’s where these tests appear under the heading “Confirm you’re not a robot.” This is because websites that use this service are trying to prevent bots from entering.

reCAPTCHA consists of some images that are shown to users. They are asked to identify certain elements within those pictures. Examples include identifying traffic lights within those images, cars, a pedestrian crossing path, and more.

We may not have known that doing these tests directly helps Google’s AI models. And that’s what say Google it officially. Identifying elements within images helps prepare data for artificial intelligence models and improve the company’s machine learning techniques.

Read also: How Google caused the establishment of one of its most important competitors

Google’s reCAPTCHA service

AI technologies need to be constantly learning, just like humans. This is done through a famous process known as machine learning.

In order for a machine to learn, it needs to see illustrative patterns, just like humans do. For a child to recognize, for example, the shape of an airplane, he must see pictures of it. And when he sees her more than once, he can easily recognize her in the future.

As for the machine, it should be provided with various images of the thing that you need to identify quickly and effectively in the future. This is the user’s role in reCAPTCHAs. We illustrate images of cars to AI, as well as images of pedestrian lanes, or images of motorbikes, and all images that appear in the tests.

Read also: Sony wants to make robots that can win the Nobel Prize

Each time the user identifies the images in the test, they are called “Labeling”. Thus, he passes the test and heads to the site, while at the same time learning the artificial intelligence of Google.

But the most important question here is: How does Google know that the user has chosen the correct answer?

The importance of this question comes from the fact that user efforts are directed at defining unknown images of AI and machine learning systems. Thus there is no way to be sure that the user’s choice is correct.

Here, Google uses a smart method, as the user asks him – for example – to select five images out of ten images, and all of these images must include a motorcycle inside.

In this case, Google places four known photos of it from motorcycles, along with six unknown photos. If the user has successfully identified the first four images, then the fifth is actually a motorcycle image. Thus it is certified and sent to artificial intelligence.

Read also: The US military relies on artificial intelligence techniques

All of this data is used to develop Google’s AI and machine learning models. Along with its self-driving car project, Waymo. On the other hand, site owners benefit from the protection of their sites on an ongoing basis.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button