Pakistan government to use Big Data and AI to find tax evaders

Finance Minister Asad Umar recently spoke in the National Assembly about amendments to the Finance Bill and hinted at using data and AI to identify tax evaders. “Algorithms are now very advanced. Global tax collection agencies don’t rely solely on traditional methods for collection, they use modern technology,” he said. “We want to use anonymised data to run algorithms which will help us identify tax evaders.” But what algorithms was Umar referring to and how do they work? In order to understand this we need to first understand what Big Data is and how AI works. What is Big Data? Let’s just say that Big Data is data that is too large to process on one computer. Suppose you wanted to work with an Excel file of four gigabytes. Odds are that the average PC will not be able to open this file in the first place. Even if it opens, suppose you want to analyse it for information -- it is very likely that your PC will either become very slow or will simply stop responding. There are technologies through which big data can be analysed. Banks and multinational companies all over the world are already using them. How does AI work? If we were to oversimplify, AI or artificial intelligence, works something like this: Ismail is a known tax evader and his bank knows it. The bank would have other information about Ismail like where he lives, his level of education, declared income, marital status. Now another person, let’s call him Ahmed, is not a tax evader and the bank has similar information about him. A bank has hundreds of thousands of customers and it has data on all of them. Through AI, banks can make a machine ‘learn’ that people with similar characteristics to Ismail are more likely to evade taxes and people who are like Ahmed are less likely to. Now, the bank is in a position to flag people who evade tax and these flagged individuals can then be investigated by the authorities. The FBR will have even more data about each individual as it is a government agency that has access to government databases. When Umar spoke in the National Assembly, he did not directly use the words ‘Big Data’ or ‘AI’. But he mentioned having an ‘abundance of data’ on non-filers and tax evaders and the way he described algorithms is similar to how someone would explain AI to a layperson. According to a paper presented at the International Conference on Computational Science and Computation Intelligence in 2016, governments can benefit from using big data technologies when trying to find tax evaders by: 1) Increasing the number of audits -- new technology is faster than traditional approaches so more people can be audited 2) Improving tax evasion detection -- better algorithms and techniques mean there’s a higher chance of detecting tax evasion 3) Improving complex fraud investigations -- patterns can be discovered from data coming from various sources 4) Improving tax revenue collection – using the latest technology can help them with this 5) Analysing all company/individual data available How does this affect Pakistan? If done right, this will help widen the tax net in Pakistan. Tax auditors would become faster and more efficient. However, if not done right, it could spell trouble for people who are not tax evaders or involved in fraudulent activities. For example, Ahmed is not a tax evader but could be flagged as one by the bank. Right now, rumours of an all-powerful Big Brother out to monitor every aspect of our lives, similar to the Social Credit System in China, seem sensationalist at best.