Captcha and AI: Why Artificial Intelligence Fails at Certain Human Tasks?

The role of Captcha in cybersecurity

THE Captcha (Completely Automated Public Turing test to tell Computers and Humans Apart) have become essential in the world of the web. These small identification tests make it possible to prove that a user is indeed a human, and not a malicious robot. Indeed, they were designed to protect websites against automated attacks that can overload servers, send spam or impersonate a user.

Different types of Captcha

There are several types of Captcha, including:

• Text-based Captcha: involves entering a series of distorted or barely readable characters.
• Image-based Captcha: asks you to select certain images according to a specific criterion.
• Audio-based Captcha: requires transcribing a sound recording containing words or numbers.

However, these tests are not flawless, since some Artificial Intelligence (AI) now manage to deceive Captchas, thus jeopardizing their effectiveness.

Why does AI fail at certain human tasks?

Although AI is becoming more and more efficient, it still remains unable to reproduce certain human skills and abilities. Captchas represent an excellent example to illustrate this situation.

The imperfection of AI in understanding natural language

Although AI algorithms are capable of processing huge volumes of data, their understanding of natural language remains limited. So, when a user is faced with a text-based Captcha, they sometimes have to decipher words whose letters have been distorted or mixed up. This task is usually quite simple for a human, but it poses a real challenge for a machine that cannot easily identify and interpret these characters.

The difficulties of AI in visual recognition

Visual recognition is also a skill where AI struggles to match humans. Indeed, despite the progress made in this area thanks to the development of convolutional neural networks, machines still have difficulty recognizing specific objects or elements in an image. This is true with image-based Captchas, which often ask users to select certain objects (like traffic lights or cars) from a series of photos. Faced with this type of test, AI generally has great difficulty succeeding.

Audio processing, another weak point of AI

Finally, audio processing and transcription are also areas where AI still struggles to compete with humans. Audio-based Captchas highlight this gap in machines, which often struggle to correctly transcribe words or numbers spoken in a sound recording.

To read Claude et le chantage : Anthropic lève le voile sur une révélation surprenante

The exploitation of human skills by certain companies

The ability of humans to pass these identification tests has recently been the subject of controversy surrounding Google. Indeed, a woman wanted to attack Google in a class action, accusing the company of taking advantage of the free labor that Internet users represent. The American justice system rejected this complaint on February 3. Nonetheless, this raises the question of using human skills to train and improve AI algorithms.

The ethical limits of the exploitation of human skills by AI

This affair pushes us to question the ethical limits of the exploitation of human skills by companies. Indeed, if certain tasks can be solved thanks to collective intelligence, it is important to ask what the consequences are for individuals and society in general.

Captchas, as a cybersecurity tool, highlight the current limitations of AI and highlight the importance of certain human skills. As machines continue to evolve and learn, companies must also question their ethical responsibility for using human labor to improve their technologies.

Sources

• https://www.lemonde.fr/pixels/article/2016/02/10/petite-histoire-des-captchas-ces-tests-d-identification-en-pleine-mutation_4862727_4408996.html