Cybersecurity experts have issued a stern warning to users regarding the input of personal information into Generative Artificial Intelligence (AI) assistants like ChatGPT, emphasizing that this data gets logged and can be repurposed to further train the application, leading to a loss of control over it and potential privacy ramifications.
These AI generative technologies have become commonplace for users, aiding various tasks in personal and professional spheres. From precise information searches to text composition in diverse formats, story translations across languages, product comparisons, and recommendations, these technologies, like ChatGPT developed by OpenAI with over 100 million active users weekly in less than a year, and Google’s Bard, are significantly integrated into daily routines.
Secure&IT cybersecurity company’s experts have flagged the potential consequences of sharing personal information with such assistants, cautioning that these platforms could store and use this data to refine their AI models, effectively relinquishing control over personal information.
Natalia Patiño, Secure&IT’s legal TIC consultant, stressed that while ChatGPT supplements human work, it doesn’t replace it entirely, emphasizing the necessity for users to validate its usage regularly. Additionally, she clarified that this technology isn’t foolproof, operating solely based on pre-trained data rather than reasoning independently.
TRANSFER LEARNING AND DATA USAGE
Secure&IT elaborated that ChatGPT employs a transfer learning technique, utilizing extensive datasets to continually enhance its learning model. When users interact with the chatbot, their queries shape the model’s responses, which evolve based on user input and contextual data. However, experts highlighted that these systems utilize all user-entered data, storing it for potential reuse, necessitating a careful consideration of the information shared with the chatbot, especially personal data.
Patiño emphasized that sharing personal or confidential information implies losing control over it, defining personal data as any information related to an identifiable individual. She advised users against inputting directly identifiable or indirectly traceable data, considering the opacity of current AI-based chats.
CHATGPT RESPONSE QUALITY AND BIASES
Secure&IT emphasized that the quality of the assistant’s response hinges on user-inputted prompts and shared context, advocating for clear and keyword-rich queries to ensure accurate responses. Nonetheless, they cautioned that incorrect or incomplete answers, even hallucinatory responses, might occur, leading to seemingly logical yet false or invented information.
The experts also underscored the existence of biases in AI tools like ChatGPT, mentioning feedback bias, wherein systems perpetuate existing prejudices and stereotypes by learning from biased user feedback. Patiño cited examples such as age or gender discrimination in workplace environments, where biased AI models might hinder the hiring of older individuals or suggest only male profiles for managerial roles.
EU REGULATION AND FUTURE IMPACT
In light of these concerns, the European Union is actively working to regulate AI usage. Secure&IT highlighted the pending approval of the AI Regulation, aimed at addressing technical, ethical, and implementation challenges posed by AI across multiple sectors.
As the conversation around AI ethics and privacy intensifies, experts emphasize the need for increased user awareness and regulatory measures to ensure responsible AI development and usage, mitigating potential privacy risks and biases embedded within these powerful systems.