Framework to design completely automated reverse Turing tests
Various systems and methods are provided for defining a CAPTCHA generator that is configured to generate CAPTCHA challenges by using at least a first parameter and a first plurality of values associated with the first parameter; defining an adversary program, where the adversary program is configured to automatically attempt to solve the CAPTCHA challenges; performing a first feedback loop that includes generating a first plurality of CAPTCHA challenges, receiving feedback from a group of human users and feedback from the adversary program; and using the feedback received from the human user and the feedback received from the adversary program to modify a weight associated with a first value among the plurality of values in order to generate future CAPTCHA challenges that create less inconvenience for human users but which are more difficult for adversary programs.
1. A method comprising:
defining a CAPTCHA generator, wherein
a CAPTCHA is a Completely Automated Public Turing tests to tell Computers and Humans Apart, and
the CAPTCHA generator is configured to generate CAPTCHA challenges by using at least a first parameter and a first plurality of values associated with the first parameter;
defining an adversary program, wherein
the adversary program is configured to automatically attempt to solve the CAPTCHA challenges;
performing a first feedback loop comprising:
generating a first plurality of CAPTCHA challenges,
receiving feedback from a group of human users with respect to a first subset of the first plurality of CAPTCHA challenges, and
receiving feedback from the adversary program with respect to a second subset of the first plurality of CAPTCHA challenges; and
modifying a weight associated with a first value among the plurality of values, to determine a modified weight associated with the first value, wherein
the modified weight changes how often the associated value will be used in generating a subsequent plurality of CAPTCHA challenges,
the modifying is based, at least in part, on the feedback from the group of human users and the feedback from the adversary program, and
the modifying is configured to accomplish at least one of:
increasing a first success rate associated with human users solving the CAPTCHA challenges, and
decreasing a second success rate associated with the adversary program solving the CAPTCHA challenges.
2. The method of claim 1 , wherein
each CAPTCHA challenge of the plurality of CAPTCHA challenges is generated, at least in part, by randomly selecting a value from among the first plurality of values.
3. The method of claim 1 , further comprising:
subsequent to the modifying, performing a second feedback loop, wherein
the second feedback loop comprises generating a second plurality of CAPTCHA challenges, and
the second plurality of CAPTCHA challenges are generated, at least in part, by using the modified weight to determine how often the associated value is used in generating the second plurality of CAPTCHA challenges.
4. The method of claim 1 , wherein
the modifying comprises applying a Bayesian inference to the feedback from the group of human users and the feedback from the adversary program in order to determine the weight.
5. The method of claim 1 , wherein:
the modifying creates a modified distribution of the first plurality of values associated with the first parameter.
6. The method of claim 1 , wherein
the second success rate is measured by determining a first rate at which the adversary program correctly identified individual characters within the CAPTCHA challenges, and a second rate at which the adversary program correctly solved the CAPTCHA challenges.
7. The method of claim 1 , wherein
the modifying results in less inconvenience for human users than was the case prior to the modifying, wherein
inconvenience is measured by determining how often human users refreshed CAPTCHA challenges and how often human users incorrectly responded to CAPTCHA challenges.
8. The method of claim 1 , wherein:
the first parameter, the first plurality of values associated with the first parameter, and the weight associated with the first value are stored in a table, and
the parameterized CAPTCHA generator uses information in the table to generate CAPTCHA challenges.
9. A computing device comprising:
one or more processors; and
one or more non-transitory computer-readable storage media to store instructions executable by the one or more processors to perform operations comprising:
defining a CAPTCHA generator, wherein
a CAPTCHA is a Completely Automated Public Turing tests to tell Computers and Humans Apart, and
the CAPTCHA generator is configured to generate CAPTCHA challenges by using at least a first parameter and a first plurality of values associated with the first parameter;
defining an adversary program, wherein
the adversary program is configured to automatically attempt to solve the CAPTCHA challenges;
performing a first feedback loop comprising:
generating a first plurality of CAPTCHA challenges,
receiving feedback from a group of human users with respect to a first subset of the first plurality of CAPTCHA challenges, and
receiving feedback from the adversary program with respect to a second subset of the first plurality of CAPTCHA challenges; and
modifying a weight associated with a first value among the plurality of values, to determine a modified weight associated with the first value, wherein
the modified weight changes how often the associated value will be used in generating a subsequent plurality of CAPTCHA challenges,
the modifying is based, at least in part, on the feedback from the group of human users and the feedback from the adversary program, and
the modifying is configured to accomplish at least one of:
increasing a first success rate associated with human users solving the CAPTCHA challenges, and
decreasing a second success rate associated with the adversary program solving the CAPTCHA challenges.
10. The computing device of claim 9 , wherein
each CAPTCHA challenge of the plurality of CAPTCHA challenges is generated, at least in part, by randomly selecting a value from among the first plurality of values.
11. The computing device of claim 9 , wherein the operations further comprise:
subsequent to the modifying, performing a second feedback loop, wherein
the second feedback loop comprises generating a second plurality of CAPTCHA challenges, and
the second plurality of CAPTCHA challenges are generated, at least in part, by using the modified weight to determine how often the associated value is used in generating the second plurality of CAPTCHA challenges.
12. The computing device of claim 9 , wherein
the modifying comprises applying a Bayesian inference to the feedback from the group of human users and the feedback from the adversary program in order to determine the weight.
13. The computing device of claim 9 , wherein
the modifying creates a modified distribution of the first plurality of values associated with the first parameter.
14. The computing device of claim 9 , wherein
the second success rate is measured by determining a first rate at which the adversary program correctly identified individual characters within the CAPTCHA challenges, and a second rate at which the adversary program correctly solved the CAPTCHA challenges; and
the modifying results in less inconvenience for human users than was the case prior to the modifying, wherein
inconvenience is measured by determining how often human users refreshed CAPTCHA challenges and how often human users incorrectly responded to CAPTCHA challenges.
15. One or more non-transitory computer-readable storage media to store instructions executable by one or more processors to perform operations comprising:
defining a CAPTCHA generator, wherein
a CAPTCHA is a Completely Automated Public Turing tests to tell Computers and Humans Apart, and
the CAPTCHA generator is configured to generate CAPTCHA challenges by using at least a first parameter and a first plurality of values associated with the first parameter;
defining an adversary program, wherein
the adversary program is configured to automatically attempt to solve the CAPTCHA challenges;
performing a first feedback loop comprising:
generating a first plurality of CAPTCHA challenges,
receiving feedback from a group of human users with respect to a first subset of the first plurality of CAPTCHA challenges, and
receiving feedback from the adversary program with respect to a second subset of the first plurality of CAPTCHA challenges; and
modifying a weight associated with a first value among the plurality of values, to
determine a modified weight associated with the first value, wherein
the modified weight changes how often the associated value will be used in generating a subsequent plurality of CAPTCHA challenges,
the modifying is based, at least in part, on the feedback from the group of human users and the feedback from the adversary program, and
the modifying is configured to accomplish at least one of:
increasing a first success rate associated with human users solving the CAPTCHA challenges, and
decreasing a second success rate associated with the adversary program solving the CAPTCHA challenges.
16. The one or more non-transitory computer-readable storage media of claim 15 , wherein
each CAPTCHA challenge of the plurality of CAPTCHA challenges is generated, at least in part, by randomly selecting a value from among the first plurality of values.
17. The one or more non-transitory computer-readable storage media of claim 15 , wherein
the first parameter, the first plurality of values associated with the first parameter, and the weight associated with the first value are stored in a table, and
the parameterized CAPTCHA generator uses information in the table to generate CAPTCHA challenges.
18. The one or more non-transitory computer-readable storage media of claim 15 , wherein the operations further comprise:
subsequent to the modifying, performing a second feedback loop, wherein
the second feedback loop comprises generating a second plurality of CAPTCHA challenges, and
the second plurality of CAPTCHA challenges are generated, at least in part, by using the modified weight to determine how often the associated value is used in generating the second plurality of CAPTCHA challenges.
19. The one or more non-transitory computer-readable storage media of claim 18 , wherein
the modifying comprises applying a Bayesian inference to the feedback from the group of human users and the feedback from the adversary program in order to determine the weight.
20. The one or more non-transitory computer-readable storage media of claim 19 , wherein
the modifying creates a modified distribution of the first plurality of values associated with the first parameter.