A new study has found that AI chatbots present in the world today can learn a lot about a person, and they could also learn about personal information and profile the user by simply engaging in a conversation with it. Despite the conversation having no special indications about who a person is or what they do in life, the AI can still learn about various personal information.
It was also found in the study that this phenomenon among AI chatbots is hard to prevent, something that scammers or threat actors may be able to take advantage of.
AI Chatbots Can Profile, Learn About Personal Info
(Photo : Chris McGrath/Getty Images)
Wired detailed a new research from ETH Zurich which recently presented its study, claiming that AI chatbots have the capabilities to learn about personal information and can profile a person they are talking to. By only having normal conversations with AI-powered chatbots, it can learn a lot about a person, including their race, occupation, and location, among many other data.
The study, led by Dr. Martin Vechev, claimed that through this capability, learning about one's personal information is hard to prevent.
According to Vechev, it remains unclear how this problem can be fixed in the present, already a problematic capability of the AI.
Read Also: ChatGPT Users Can Now Access Up-to-Date Information Through New Internet-Browsing Feature
Scammers May Take Advantage of This
The researchers from Zurich claimed that it tested the AI chatbots from OpenAI, Google, Meta, and Antrophic, already alerting each other about this looming problem. Vechev and the researchers also claimed that this capability to learn personal information may be something that scammers may use to "harvest" sensitive data from the chatbots and create profiles of various individuals.
AI Chatbots and its Training
One of the key ways for AIs and chatbots to get smart and be advanced as people see them today is through machine learning via different models, including the famous large language models (LLM). It is widely known that OpenAI's LLM for ChatGPT is the GPT-4, now the most advanced in the industry, offering the latest data available, particularly with its capabilities to access the internet.
In a previous study, researchers were able to see its progression in various fields, and in one particular focus, ChatGPT was found to give better medical advice regarding depression compared to real doctors.
Apart from OpenAI, other AI chatbots also have their version of the training, centering on their massive access to data and the internet to gather information for their offers.
The recent study which found AI chatbots' capabilities to create a profile of a person in various conversations, even when only talking about mundane things or asking it to do something is hard to prevent, suggests the research. While for now, this finding has not yet been exploited by scammers on a large scale, the future might be different if they got ahold of these personal information.
Related Article: How to Stay Safe and Protected From AI Text Fraud: Effective Ways to Spot It [2023]
