In recent years, there has been a lot of enthusiasm to change the society about the ability of Artificial Intelligence (AI), from improving medical diagnosis to reducing learning gaps. Tax authorities around the world have also joined this campaign and have focused their sites on a main problem: overthrowing tax fraud. India is also excited by the AI’s capacity that they tighten the deception of tax.
However, this enthusiasm needs to be carefully angry with planning and testing. AI tax administration is unlikely to strengthen the administration until they are equipped to use it and are equipped to subjugate it to harsh tests. A recent study of India highlighted this, incorporating an AI-based equipment to prevent tax fraud.
The increasing number of countries to improve tax administration is integrating AI devices. A report by 2023 OECD has shown that around 80% of the surveyed tax administration had implemented or implemented AI devices for virtual assistance, future stating analysis and transcription with other uses. In India, tax officials are searching for AI to improve dispute resolution and auditing.
This trend aligns with a broad push to avail technology to improve tax collection. For example, a random assessment in Ghana found that providing electronic tablets loaded with a geophysical database to property tax collectors increased by 103% in the collection. Another assessment in Senegal found that a computer-aid property assessment system was much more accurate and fair than the manual assessment of the property tax collectors.
The benefits of technology are not unavoidable, though. In India’s study, researchers used an offshoot machine learning of AI to detect goods and service tax (GST) fraud. While the equipment was successful in detecting fraud, it did not increase enforcement or revenue. The project provides a caution story on implementing AI and other technologies in tax collection.
GST presented several central and state taxes to make the collection more efficient and simplified. However, the system has been reduced by mass theft. The estimated revenue loss in FY 2024 was equivalent to 10% of the total GST collection as per the Directorate General of GST Intelligence.
Some firms have discovered creative ways to avoid taxes. They can evaluate goods and services to pay low tax, claim incorrect discounts, or claim more input tax credit (ITC) than permission.
The claim of ITC fraud through fake firms has been particularly problematic. Fake firms are institutions that are present on paper to generate pure imaginary challan and ITC. Taxpayers can use these fake credits to offset some, if not all, their tax liability. In August, for example, a major GST scam in Delhi was exposed, where 500 fake firms produced invoice 718 crore to claim GST refundable 54 crores.
To catch fake firms, Berkeley, a research team led by a uninterrupted Mahajan, at the University of California, created a novel machine learning tool. The machine learning technique is a branch of AI that can process large amounts of data, detect patterns and make predictions.
Using tax data and inspection records, researchers trained the equipment to generate a rank list of firms that had a high probability of being fake. Researchers used common signs of fraud, such as claiming high level ITCs too low or any real tax payment or sharing registered address with several firms. Although traditional methods consider the same criteria, the equipment included a broad set of factors to detect fraud.
Depending on the field inspections, the ML Tool was found to be highly effective – 53% of the firms flagged by the ML Tool were confirmed, while compared to 38% firms marked by traditional methods. It is impressive how difficult it is to identify fake firms and their customers. To make tax evader officers hard to catch them, tax can be bounced among many fake firms.
Despite the accuracy of the tool, it did not increase enhanced enforcement of fake firms to cancel the increased or rapid, or increase enlarged enlarged against customers of fake firms. In other words, the tax collection did not improve, even though the equipment was relatively accurate.
The aim of the Government of India is to become a global leader in AI through its Indiai Mission. States like Telangana have also set up dedicated AI Research Centers. This study offers policy makers important lessons to effectively benefit this technique.
The greatest takeway is: Even though the AI ​​can reduce the burden of tedious works and can empty time for better other functions for humans, governments need to strictly evaluate whether AI equipment desired results Will give birth. It is uninterrupted to assume that AI will naturally lead major reforms.
In this case, the AI ​​Tool offered more efficient and accurate ways to detect fake firms, which could enable tax authorities to focus on taking action. But this did not provide evidence to the authorities to help them investigate and make a case against them – the authorities still have to invest a lot of time to investigate each case. Spending great time and money on the construction of AI tools, only falling into humiliation for it, would be a tremendous waste.
More fundamentally, policy makers should first ask whether AI is the best solution to a problem. Researchers say that the ML models developed by them may detect firms that are likely to be inspected or canceled anyway by traditional methods. Therefore, simple technologies may have high and fast returns, for example, improve coordination between tax authorities in various courts.
AI’s appeal makes sense. It is a common purpose technique that improves widely accessible, highly adaptable and constantly. But deploying AI for the sake of AI is unstoppable.
This article is written by Guillermo Heera Nimagadda, Policy and Training Manager, J-Pal South Asia.
