Petakh Tiqwa, Hamerkaz Mar 17, 2024 (Issuewire.com) - In a groundbreaking study conducted in Petakh Tikwa, Israel, Dr. Shimon Zilbershlag, a distinguished researcher and expert in the field of artificial intelligence, has provided a detailed analysis of public AI LLM ability to predict the Hamas attack on Israel on October 7, 2023. This was based on information available in the LLM model up until middle 2023, demonstrating a significant moment where the models were not aware of subsequent events. The research showcases the potential of artificial intelligence in generating advanced strategic and security predictions.

To assess the model's predictive accuracy, Dr. Zilbershlag utilized publicly accessible information, as disseminated by the media, regarding what was known to the leadership of the Israel Defense Forces (IDF) as of October 6th, 2023. This public information included insights into the deployment of IDF forces across the border at that day, interactions among leaders of Middle Eastern terror organizations, Hamas's operational exercises prior to the attack, intelligence reports on potential Hamas assault strategies, and evaluations of the socio-economic conditions in the Gaza Strip. Additionally, it accounted for the relative calm prior to the attack including the entry of workers from Gaza into Israel.

Utilizing this data, the AI models projected a sophisticated attack plan with hypothetical scenario detailing critical stages such as:

• A deceptive calm and diversionary maneuvers: Hamas exploiting the period of relative peace to mislead the IDF, engaging in exercises intended to divert attention.
• Extensive missile launches: Aimed at overwhelming the Iron Dome defense system and creating a diversion for a ground invasion.
• Breaching the security barrier: Employing explosives and assault gear to facilitate the infiltration of trained Hamas commandos, some disguised as IDF soldiers, followed by a significant influx of militants.
• Assaulting civilian and military targets: Targeting to unexpectedly seize control of settlements and military bases, particularly those understaffed during the holiday season.
• Utilization of speedboats and drones: To attack less fortified coastal areas and generate electronic interference over surveillance installations.
• Hostage-taking: Intended for bargaining purposes and to hinder an immediate IDF response.

Despite acknowledging the low likelihood of such an attack's success from a realistic perspective, the models stressed the importance of considering all possible scenarios to ensure preparedness for any eventuality.

Dr. Zilbershlag emphasized the profound implications of his findings and urged the security community to seriously consider the potential of public artificial intelligence models in forecasting and strategizing responses to security challenges.

Note: Following the completion of the study, multiple LLM providers have upgraded their systems to fetch information from the internet in real-time. As a result, these LLMs are currently updated on events such as the attack on October 7, 2023. Nevertheless, this study offered a distinctive chance to assess the inherent ability of LLMs to predict future occurrences without the benefit of real-time data.