AI Crowd Estimation: How Many Attended the Degel HaTorah Rally?

Mass political and religious gatherings have always sparked debates regarding crowd sizes, with organizers and opponents often presenting vastly different numbers. Recently, the Degel HaTorah movement held a massive demonstration on the Rav Landa Bridge in Bnei Brak, drawing a sweeping sea of participants under the guidance of prominent rabbinic leaders. The sheer visual impact of the crowd, as documented from above, prompted a modern technological experiment to determine the exact scale of the assembly.

To achieve an objective perspective, aerial photographs taken by professional photographer Shuki Lehrer were fed into six of the world's most prominent artificial intelligence models: Meta AI, Gemini, ChatGPT, Grok, DeepSeek, and Claude. The goal was to leverage advanced computer vision and density estimation algorithms to count or extrapolate the number of individuals present in the given geographic space, bypassing human political bias.

The resulting figures from the six AI models showcased a staggering variance. While some models provided conservative estimates starting at roughly twenty thousand participants, others scaled their calculations drastically upward, with the highest projections reaching an astounding one hundred and fifty thousand attendees. This immense gap highlights the methodological challenges artificial intelligence systems face when interpreting complex, densely packed real-world imagery.

Analyzing crowds from aerial perspectives involves numerous variables, including shadow interference, overlapping bodies, variable clothing colors, and difficulties in defining exact perimeter boundaries. Traditional crowd estimation methods, such as the Herbert Jacobs method which calculates density per square meter, often require precise human calibration. When AI models attempt these calculations autonomously without standardized calibration metrics, their outputs can fluctuate wildly depending on how they interpret spatial depth and density distribution.

Ultimately, this digital experiment demonstrates that while artificial intelligence is rapidly evolving and finding applications across various industries, counting massive crowds from static photographs remains an imprecise science. The wide discrepancy between the six leading systems serves as a reminder that human expertise, contextual understanding, and standardized measurement techniques are still essential when evaluating public events and large-scale demonstrations in complex urban environments.