Israeli military creating ChatGPT-like tool using vast intercepted communications
The Israeli military has developed a powerful artificial intelligence tool that resembles ChatGPT by utilizing a vast collection of intercepted Palestinian communications. Unit 8200, the military surveillance agency, trained the AI model to comprehend spoken Arabic using extensive telephone conversations and text messages obtained through surveillance in the occupied territories.
Development of the AI model
The project to build the large language model (LLM) was initiated to create a chatbot-like tool capable of answering queries about monitored individuals and offering insights from the surveillance data collected. Unit 8200 accelerated the development of the system following the conflict in Gaza in October 2023.
The former military intelligence technologist overseeing the project mentioned the challenges faced in collecting and processing a massive dataset of Arabic communications. The AI model required a substantial amount of data, with training data eventually incorporating around 100 billion words.
Integration of AI in intelligence analysis
The integration of AI systems like LLMs in intelligence analysis poses risks of bias magnification and errors, as highlighted by experts and human rights advocates. The opaque nature of AI-generated outcomes can further complicate the understanding of decision-making processes.
The Israel Defence Forces (IDF) have deployed various intelligence methods to counter terrorist activities in the Middle East, although they declined to comment on specific queries related to the new LLM.
Concerns and implications
Critics express concerns about the potential misuse of AI-generated insights in making decisions that impact individuals' lives, particularly in conflict zones. The usage of surveillance data to train AI models raises privacy and human rights issues, especially concerning individuals not involved in criminal activities.
Global trends in AI adoption
Several countries, including the US and the UK, are also exploring AI technologies in intelligence operations. However, the approach taken by Israeli intelligence agencies in leveraging AI for surveillance has raised questions about oversight and data handling.
Experts caution that while AI tools like ChatGPT can enhance threat detection capabilities, there is a risk of generating false connections and erroneous conclusions. Understanding the decision-making process behind AI-generated outputs is crucial to mitigate potential risks and consequences.
