Developed 20 years ago, so that hackers do not steal content, insert malicious posts, engage in dishonest transactions or slowing websites to complete inoperative, the acronym of this ubiquitous defense line clearly defines his mission: a completely automated Turing test to tell computers and people.
For almost two decades, Captchas are widely used as a means of protection against bots. With time, along with the dissemination of their use, the techniques of bypassing or fraud, Captchas constantly improved. However, the Captchs have also evolved in terms of complexity and diversity, becoming more and more difficult to solve for both bots (machines) and people. Considering this long -term and still significant technological competition, it is extremely important to examine how much time legal users require to solve modern captchas and how they are perceived by these users.
Today, Captchas still remain one of the main problems of user.
However, scientists from the University of California in Irvine came to the conclusion that bots seem to solve them better than people.
In their work, scientists examine Captchas in natural conditions, assessing the results of users in solving them and their perception of currently used captchas. They collect this data through manual controls of popular websites and user research, in which 1,400 participants were solved by 14,000 captchas in total. The results show significant differences between the most popular types of Captchas: surprising, time needed to solve, and the perception of users does not always correlate. A comparative test was carried out to examine the impact of the experimental context – in particular the difference between the direct Captchas solution and the solution to them as part of a more natural task, such as creating an account. Despite a few potentially misleading factors, the results indicate that the experimental context may affect this task and should be considered in future Captcha studies. Scientists also examine the refusal by users to perform tasks caused by Captchas, analyzing participants who start the task and do not finish it.
They discovered that bots not only solve various forms of Captchas, such as image recognition, slide puzzles and a distorted text, better, but also faster.
As evolutions, Captchas in terms of complexity and diversity is becoming more and more difficult to solve for both bots (machines) and people. However, with the development of computer vision and machine learning, bots of bots in recognizing the distorted text have increased significantly, achieving an accuracy of over 99%. Bots can successfully overcome Captchas with a distorted text in almost 100% of cases. Human accuracy in Captchas solving ranges from 50% to 84%. In addition, the solution to these tasks takes people up to 15 seconds, while bots can do it in less than a second.
Based on this study, scientists came to an obvious conclusion: there is no simple way, based on small paintings or other features to ultimately distinguish people from bots. Instead, they recommend the use of artificial intelligence to develop “intelligent algorithms” that can more effectively distinguish between the bot from human activities.
You can read a full research article about the following to combine.
