Artificial Intelligence

Study finds bots are better than humans at 'Are you a robot?' tests

Researchers found bots are faster and more accurate than humans at solving these online puzzles that are intended to dupe them.

"I'm not a robot" test on a computer screen
Scripps News
SMS

Web developers have long used CAPTCHA, also called "bot tests," "are you a robot?" tests, as a security tool to prevent automated bots from accessing their websites. But a comprehensive new study suggests these tests may now be doing more harm than good.

Researchers at the University of California Irvine found that bots are better and faster than humans at cracking these online puzzles that are designed to block them from accessing sites and masquerading as legitimate online users.

Federal regulators launch investigation into ChatGPT creator OpenAI
Federal regulators launch investigation into ChatGPT creator OpenAI

Federal regulators launch investigation into ChatGPT creator OpenAI

The Federal Trade Commission is investigating whether the company engaged in "unfair or deceptive" data security practices that harmed consumers.

LEARN MORE

The study — which has yet to be peer-reviewed — looked at 200 of the most popular websites and found 120 of them still used CAPTCHA to trip up software.  

Researchers then recruited 1,000 online participants to take 10 of these tests on the websites and gauge their difficulty. Their findings showed that various bots described in scientific journals were both faster and more accurate at solving the tests than their human counterparts.

For instance, when humans solved distorted text CAPTCHA tests, it took them about nine to 15 seconds on average with an achieved accuracy rate of 50% to 84%. Bots taking the same tests were able to complete it in under a second with near perfection.

OpenAI unveils GPT-4, its most advanced AI language model
OpenAI unveils GPT-4, its most advanced AI language model

OpenAI unveils GPT-4, its most advanced AI language model

GPT-4 exhibits "human-level performance" on various professional and academic benchmarks, the company said.

LEARN MORE

"The bots' accuracy ranges from 85-100 percent, with the majority above 96 percent," scientists wrote in the study.

Gene Tsudik, one of the researchers on the project, said the findings suggest CAPTCHAs do very little to steer bots from accessing websites and only provide an additional burden to authentic online users. 

"We do know for sure that they [the tests] are very much unloved. We didn’t have to do a study to come to that conclusion," he said. "But people don’t know whether that effort, that colossal global effort that is invested into solving CAPTCHAs every day, every year, every month, whether that effort is actually worthwhile."