You have had to pass a CAPTCHA test at some point in your life. It has become a part of the Internet experience and we’ve come to grumble less about them. That is in part thanks to the evolution of the test. That evolution has been necessary to keep the test one step ahead of other technology that could allow a non-human to solve the test. It may be hard to believe but these CAPTCHA tests may teach us what being a human is.
CAPTCHA
CAPTCHA, or Completely Automated Public Turing Test to tell Computers and Humans Apart, have been used for almost as long as the general public has had access to the Internet. At first they were a simple block of squiggly text that forced the user to write it into a box to prove that they were human. It has evolved overtime to different tests, from picking out an object in a series of pictures to just simply checking a box that says that the user is not a robot.
These tests have been instrumental in helping to develop AI and machine learning, which has evolved to the point that AI is better at solving the text blocks than humans are. Using Google’s image recognition software AI has also caught up in that regards to the point that the images used tries to hide them or disguise them to confuse the AI.
Can CAPTCHA overcome AI
As CAPTCHA continues to get more complex humans are passing the test less. A new type of CAPTCHA uses a puzzle which requires users to rotate pieces or move pieces based on instructions written into those pieces. It is hoped that AI would not be able to recognize the instructions and thus would not be able to complete the puzzle. The problem is that many humans are not able to complete this test successfully either.
It is not that people are dumb. Creating the perfect CAPTCHA test that humans will be able to solve and AI cannot is a daunting, if not impossible, task. On the human side the test must cross cultural and linguistic barriers that anyone without any training can pass and this puts a severe limit on what can be done. For the most part this limits it to image processing and AI can now pass these tests with ease. The perfect CAPTCHA test is probably impossible to develop.
Understanding what it means to be human
Many security experts believe that any CAPTCHA test can or will be able to be broken by AI at some point. This will lead inevitably to a new strategy and one that has been proposed is continuous authentication, or having AI look for signs of automation. A human will not usually do things in the exact same order on a page. They will not move the cursor the exact same way to the exact same point every time. AI will do that which is causing Google to try to redevelop their own tests away from CAPTCHA with their new concept called “Action” which can be incorporated into multiple pages and allowing the user to customize it.
AI will eventually analyze everything that a user does online to search for suspicious behavior in the hopes that this activity can be tracked and potentially stopped. This will require Internet users to surrender any vestige of privacy as a machine will always be watching. Users who value privacy and use something like a VPN to maintain that will be flagged and those services could be put out of business. That may work well in places like Communist China but it may not go over quite as well here in the US.
AI still needs human programmers
As smart as AI can potentially become humans are still necessary. In order to program AI these people need to be able to understand human psychology. This goes both for an AI program designed to protect us and for the people developing bots to do us harm. This has led to a new wrinkle in the development of CAPTCHA tests: the bot needs to get some things wrong. One of the fundamental parts of being human is that we are not perfect.
AI at this point can complete many CAPTCHA tests better than a human. Bots are even able to carry on conversations with a human passing their non-human-ness off as being foreign intentionally misspelling words and dodging questions by making bad jokes. People wouldn’t expect something like that from AI. The ability to get CAPTCHA tests wrong has led Amazon to receive a patent for the next generation of tests where the only way to pass is to get it wrong.
It is quite possible that within a decade CAPTCHA tests will be a thing of the past and AI will do battle with AI on a virtual battlefield. It can be a bit scary to think that our experience on the Internet could be dependent on AI. What will happen if the bad guys are able to win this new virtual arms race? Let’s hope we never find out.