In the Bouches-du-Rhône, the AI made it possible to collect 5 million euros by tracking the undeclared swimming pools
AI in Tax Enforcement: Detecting Undeclared Swimming Pools in France
Introduction to AI-Driven Tax Collection
Since 2021, the Bouches-du-Rhône tax administration in southeastern France has been leveraging artificial intelligence to detect undeclared swimming pools. This innovative application of AI has not only streamlined the tax collection process but has also significantly boosted revenue. In 2022 alone, this system facilitated the collection of 5 million euros in tax revenue by identifying 11,200 swimming pools, out of which 7,200 were verified and resulted in tax updates. This program serves as a model for how AI can be integrated into government operations, potentially impacting tax collection practices in the United States as well.
The Role of AI in Tax Administration
The AI tool used in Bouches-du-Rhône utilizes satellite imagery and machine learning algorithms to identify features like swimming pools that may have been omitted from property declarations. According to “France 3 Provence-Alpes-Côte d’Azur”, the system works by scanning aerial and satellite photos for telltale signs of an undeclared pool.
This technology has proven effective but has also faced hurdles. In 2021, the system had a 30% error rate, largely due to its inability to distinguish between different pool types and other blue features in satellite images such as tarps or other water bodies. However, these inaccuracies have been greatly reduced through various fixes and improvements over time, making the tool well worth its implementation cost.
“AI allows us to have an improved database, it saves us time and precision, without dismissing the human from this activity. […] It is the human who validates or not what the AI said to him,”
Limitations and Human Intervention
Despite the advancements, AI systems in tax collection still require significant human oversight. While AI identifies the pool, tax officers analyze the data provided to determine whether the detected pool is legitimate and a tax assessment required.
Tax official Claire Sarrail, who is also instrumental in CGT-Finance nationally, underscores this challenge: “AI is only used to identify what is in blue on the plans, the rest requires human analysis.” She narrates an instance where she spent two full days to differentiate a swimming pool from a work cover recommended as a pool by the algorithm. This hard-coded classification is typical of almost all such AI systems and is a testament to the human element still indispensable in modern technology.
ction>
Practical Applications and U.S. Implications
The successful use of AI in Bouches-du-Rhône raises intriguing prospects for similar applications in the United States. Imagine IRS using satellite imagery to detect undeclared home extensions, thereby reducing non-compliance with property tax laws. Such measures have actual political and populist backing and allow local governments to provide better public services from funds raised by such methods. However, privacy concerns and the potential for technical errors must be carefully addressed.
For instance, several U.S. jurisdictions use drones and aerial photography to identify properties that may not be properly taxed, such as New York City’s efforts in the 2010s to address underreporting of rental properties in the city, where the city’s Finance Department used data analytics and Google Maps to identify unlicensed dwellings. While federal agencies are just getting started with AI and machine learning, this demonstrates the feasibility of integrating such technologies at state and federal levels.