Is it possible that, in the future, all taxpayers´ doubts regarding taxes are resolved by virtual assistants? Will we see a world in which taxpayers do not have to file tax returns, which will be made by intelligent machines? Will a verification procedure be sorted out  without the intervention of officials? These are some of the possibilities potentially offered by artificial intelligence (hereafter AI),  which we must pay attention to from now on. As the European Commission points out, due to the growth of computing capacity, the availability of data and the advances in algorithms, we are facing one of the most strategic technologies of the 21st century (European Commission (2018), Communication of 7 December 2018, Coordinated Plan on artificial intelligence, Brussel)

The possibilities are huge, but also the risks. For this reason, we need to reflect on AI from a variety of perspectives (technological, economic, sociological, legal and ethical) and we welcome the contributions that allow us to enhance their advantages and control the existing problems and risks. From a tax perspective, the new possibilities must be analyzed, and specifically, the role AI can have in the transformation of the countries’ tax administrations and tax systems. In addition, we must consider how the position of taxpayers may be affected in order to strengthen their rights and guarantees where appropriate,  as well as the possible new framework of relations that may arise between the two sides of the tax relationship.

For tax administrations, AI can play a key role, as the use of these techniques finds a particularly suitable area of extraordinary potential.  This is because, as the experts point out, although AI is developed through algorithms and data, the success of the results that it can achieve is mainly found in the data that are captured, and especially in their volume. In this sense, there is no doubt about the privileged position in which tax administrations find themselves, because of the immense amount of data they have, and because of the quality of this information.

Such a position of privilege requires the use of AI by the Tax Administrations, not only for the direct fight against tax fraud, but above all with the aim of providing a better service to taxpayers, increasing their compliance of tax obligations. In the era of transparency that we are going through, tax administrations have more and more data, which in many cases are obtained via the taxpayers´ reporting obligations that they manage. These obligations have increased substantially and pose a very significant indirect burden on individuals and companies, which governments must “compensate” in some way.

At what point are we in the use of AI by tax administrations? According to the report of the OECD Tax Administrations 2019: Comparative Information on OECD and other Advanced and Emerging Economies, more than 40 tax Administrations are making use of AI, or plan to do so.

In the area of prevention, many countries are already making progress in various ways. First, to assist taxpayers by informing them of their tax obligations or by resolving doubts through virtual assistants or chatbots that, unlike traditional information programs, can assist dynamically. According to the information contained in Tax Administrations 2019, by 2017, 10 countries already have chatbots, 7 are implementing them and 23 are planning to do so. In Spain, the Tax Agency  collaborated with IBM Watson to resolve doubts related to the immediate supply of VAT information, creating an AI-based virtual assistant whose functioning is proving very positive. According to information from the Tax Agency , the number of emails received has been reduced by 80%, and the number of queries to the virtual assistant has increased tenfold in the first week. Secondly, tax compliance is also encouraged when, if an irregular situation is detected, taxpayers are deterred from carrying on with it. In this context, the Spanish Tax Agency  has sent letters to small businesses informing them that, according to their information obtained through AI, the revenues declared by them are below the average in the sector.

On the other hand, in the fight against fraud, big data and AI are used in many countries (Spain, the United States or Canada) in order to assess tax risks, which allows for segmenting taxpayers according to the probability of non-compliance, with controls being initiated in the most likely cases of fraud.

But what other areas allow using big data and AI?

The prospect for the future is to be able to reach all areas and tax procedures, all the functions and competences of tax administrations. Let us think of the whole course of automation of administrative functions. As far as tax procedures are concerned, in the management area there is also an important field for including AI and improving the efficiency of the procedure, following the example of the reference value for real estate that the General Directorate of Cadastre has developed with neural networks in Spain. In the collection process, some Administrations can already predict bad debts in order to prioritize the enforced collection (Finland, Ireland, Singapore or Sweden). But, going further, it is possible to imagine that AI will be used for other purposes in the collection field; for example, it could be used for the adoption of precautionary measures or for the derivation of tax liability procedures. A tax review procedure could also be resolved through these techniques. It is even possible to imagine the complete resolution of a procedure in an automated manner, without the intervention of officials, although we are still far from that reality.

In short, many benefits can be achieved through the use of AI, and in our area, they should ensure better tax compliance, in the sense of easier and fairer compliance for taxpayers, in a world with fewer errors on the part of administrations. It is true that this is the view of AI from a clearly optimistic perspective. Faced with this concept, however, there are also risks on what could be a misuse of AI, which requires an appropriate assessment from an ethical perspective and the adoption of a set of principles that should govern administrative action.

• Firstly, the principle of prudence, by which we avoid the complexity of the algorithms or of the scope of the projects in which they are used. Progress is made based on secure results or pilot programs which are put in place to test the results concretely and progressively, and to introduce caution before their widespread application. In this context, we should also keep in mind that the use of these techniques implies a significant economic cost for countries, and it is therefore important to assess the means used in terms of the results that can be obtained. Furthermore, the principle of prudence must also be present when assessing the validity of the conclusions arising from these programs. Such conclusions should not replace, at least at the start-up stages, the work of tax officials but should complement them.

• Secondly, the principle of non-discrimination. As we know, algorithms are based on hypotheses developed by scientists, which implies the risk that human errors or their biases may be transferred to the algorithm itself, conditioning the validity of the new hypotheses and their results. Let us remind you of the case of an algorithm that became racist within 24 hours due to exposure to racist material.

• In the third place, the principle of proportionality, under which we must evaluate the degree of interference that occurs in the rights and guarantees of the taxpayers with the decisions that are derived from programs that use AI. For example, we may think of the impact that sending a letter to a taxpayer can have by informing  that the administration has certain information about his tax situation. This seeks to encourage  voluntary compliance and is different from starting an intensive control . Moreover, this principle should lead us to the utmost caution when fundamental rights can be affected. The use of presumptive statistical techniques, as well as the use of AI tools, may serve as an indication for certain management actions, but not as a single test.

• Fourthly, the principle of transparency implies the adoption of measures that will enable taxpayers to know why a decision has been taken, without limiting their right of defense. One solution would be to require the external auditing of algorithms that ensure their proper functioning or even ex post certification as a condition for the validity of decisions. The interesting French model includes, in the Law for the Digital Republic, the right of the administered to have knowledge of the use of algorithms for individual decision making, as well as to have information about the functioning of the algorithm (how it has contributed and to what extent to the decision making, the data being processed and the processing parameters).

• Finally, data governance is relevant to ensure data security, for which tax administrations are responsible, while respecting privacy and confidentiality. In addition, administrations should also take responsibility for the quality of the data and the integration of all information should be encouraged.

In short, the decisions that can be derived from well-designed AI algorithms and fed with taxpayer data should allow a better application of the tax system, and the existence of higher quality tax acts, which promote unity of administrative criteria, more legal certainty for taxpayers, a reduction in resolution time and less conflict. In addition, they should lead to a reduction in the possibilities of tax fraud, which undoubtedly benefits society in general. However, technology must be at the service of taxpayers and the administrative action in tax application procedures must be inspired by the principles of prudence, proportionality, non-discrimination, transparency and information governance.

26,366 total views, 39 views today