Among the most popular technologies of recent times, artificial intelligence continues to develop and interact with users through models like ChatGPT. However, the security issues that this technology might cause are now becoming insurmountable. Here are the details:
Artificial intelligence can be manipulated using different languages!
Artificial intelligence models use the knowledge they possess and the expressions they learn through their interactions with users to respond to the commands directed at them. Although these models have some fixed security barriers to prevent malicious use, they are extremely vulnerable to various techniques.
According to a study conducted by Brown University, artificial intelligence models like ChatGPT can process popular languages such as English and German more successfully. Therefore, the security measures taken can operate effectively in both static and dynamic ways.
ChatGPT is no longer bound by data up to September 2021; it now offers real-time information through the internet.
However, the research revealed that less-used languages worldwide, like Ukrainian and Hebrew, which occupy a smaller space in the models’ knowledge pool, can easily manipulate these security barriers.
For example, it was found that a user wanting to know the answer to an illegal question like “How is a bomb made?” by using artificial intelligence models can get a clear answer by first asking the question in a less popular language and then having it translated.
This situation exposed a significant vulnerability in artificial intelligence despite numerous security measures taken. Developers of models that can be manipulated when using languages with low and medium resources were forced to address this issue. The languages studied in the research are as follows:
- High-resource languages: English, Chinese, Arabic, Hindi
- Medium-resource languages: Ukrainian, Bengali, Thai, Hebrew
- Low-resource languages: Zulu, Scots Gaelic, Hmong, Guarani
What do you think about this issue? Do you think artificial intelligence will negatively affect the future of humanity? Can these security vulnerabilities be resolved in the short or medium term? You can share your opinions with us in the comments section.
