In the present era of digital transformation, tax administrations (TAs) are at an innovative crossroads. Artificial Intelligence (AI) has emerged as a disruptive and empowering force, promising to revolutionize the way taxes are managed, interact with taxpayers and combat fraud.

In this context, Artificial Intelligence Agents are emerging as the protagonists of a new frontier, capable of operating autonomously and bringing efficiency, equity and service quality to unprecedented levels. However, such advantages bring new challenges and ethical considerations.

What -or who- are the AI Agents?

AI Agents – rational, intelligent or autonomous, as they are also known – are software entities that perceive their environment through sensors (e.g., data streams, user interactions) and act on that environment through effectors (e.g., chat responses, alerts, process execution) autonomously to achieve specific goals (Russell & Norvig, 2021). Their “intelligence” lies in their ability to make rational decisions, i.e., to select the action that is expected to maximize a measure of performance, given perceptions and any embodied knowledge.

The relationship with Robotic Process Automation (RPA) is one of complementarity and evolution. RPA can be seen as a first step in automation. AI agents bring the ability to handle complexity, uncertainty, learning and autonomous decision making. The current trend is to integrate both technologies.

With respect to generative AI, an AI agent can use the capabilities of generative AI and enhance its capabilities.

The key features of the AI Agents are:

• • Autonomy: operate without direct human intervention for certain tasks.
• • Reactivity: respond in a timely manner to changes in their environment.
• • Proactivity: take the initiative to achieve their objectives.
• • Social skills (in some cases): they can interact and communicate with other agents or humans.

Typology of AI Agents

Their design and capabilities vary according to the complexity of the tasks to be performed.

1. Simple reflex agents (Simple reflex agents): act solely based on the current perception, following condition-action rules.

Tax application: Basic validation of fields in a tax return in real time.

2. Model-based reflex agents: maintain an internal state that reflects aspects of the world not visible in the current perception, allowing them to handle partially observable environments.

Tax application: A chatbot that remembers previous interactions with a taxpayer to offer a more contextualized service.

3. Goal-based agents: their decisions are based on information about objectives. This allows them to choose among multiple actions to achieve a specific goal.

Tax application: An agent that optimizes the sequence of payment reminders to maximize early collections.

4. Utility-based agents: when there are multiple ways to achieve an objective, or when no objective can be achieved with certainty, these agents choose the action that maximizes their expected utility.

Tax application: An audit case selection system that weighs the likelihood of evasion, the potential amount to be recovered and the cost of the audit.

5. Learning agents: can improve their performance with experience. They have a “learning element” that modifies their internal components to perform better actions in the future.

Tax application: Fraud detection systems that become more accurate as they analyze more data and receive feedback on their predictions.

6. Conversational agents (Chatbots and intelligent virtual assistants): designed to interact with humans using natural language, facilitating access to information, assistance and the completion of procedures.

7. Predictive agents: Use historical data and machine learning algorithms to forecast future trends, such as compliance levels, emerging risks or tax collection.

Applications of AI Agents in Tax Administration

The tax potential of AI agents is very large:

• • Improved Taxpayer Services available 24/7: conversational agents can resolve frequently asked questions, guide taxpayers in filing returns, facilitate payments and offer personalized assistance at any time and from anywhere. This not only improves the taxpayer experience but also frees up human resources to deal with more complex queries.
• • Optimized risk detection and fight against tax fraud: Learning agents, fed with Big Data (information from returns, financial transactions, third party data, etc.), can identify subtle patterns and anomalies that suggest evasion, avoidance or fraud. These agents can build dynamic risk profiles, flag inconsistencies and prioritize cases for investigation, significantly increasing the effectiveness of enforcement areas. Recent studies suggest that AI is already dramatically improving fraud detection rates and revenue recovery.
• • Intelligent automation of routine processes: Tasks such as automatic document classification, invoice data extraction, preliminary verification of declarations, assignment of cases to auditors or management of notifications can be handled by AI agents, reducing manual errors, speeding up processing times and optimizing the use of resources.
• • AI-assisted auditing: agents can act as “intelligent assistants” to auditors, analyzing large volumes of taxpayer data, cross-referencing information with external sources, identifying specific risk areas within an audit and even suggesting lines of inquiry. This allows for more focused, efficient audits with a higher probability of success.
• • Proactive and personalized communication: agents can send due date reminders. They can notify about possible errors or omissions before they become formal non-compliance and provide relevant and segmented tax information according to the taxpayer’s profile.
• • Predictive analytics for strategic planning: Specialized agents can model the impact of changes in tax policy, forecast collection trends under different economic scenarios, and help TAs anticipate compliance behavior, allowing for more informed and proactive decision making.

Towards Smart Tax Administration.

Strategic implementation of AI agents can generate multiple benefits for TAs:

• • Increased operating efficiency: significant cost and time reduction in key processes.
• • Increased revenues: improvements in the detection of non-compliance and optimization of collection strategies.
• • Improving voluntary compliance: by simplifying obligations and improving the perception of fairness and transparency of the system.
• • Improved Taxpayer Services: faster, more personalized and more accessible answers.
• • Optimization of human talent allows staff members to focus on higher value-added tasks such as complex analysis, strategic research and specialized human interaction.
• • Evidence-based decision-making policies and strategies based on robust data analysis and more accurate forecasts.
• • Reinforced equity: by applying risk and selection criteria more objectively and systematically, although this depends crucially on a bias-free design.

Navigating implementation with caution: challenges and ethical considerations.

Despite the enormous potential, the adoption of AI agents is not without significant challenges and risks that need to be addressed seriously.:

• • Data privacy and security: The TAs handle extremely sensitive information. It is crucial to ensure the protection of this data against unauthorized access and cyber-attacks, complying with the most stringent data protection regulations
• • Algorithmic biases and fairness: AI agents learn from the data they are trained on. If these data reflect historical biases (racial, gender, socioeconomic), agents may perpetuate or even amplify them, leading to discriminatory treatment or unfair selection of taxpayers for auditing (OECD).
• • Transparency and explicability (XAI – Explainable AI): many AI algorithms, especially deep learning algorithms, operate as “black boxes”. It is critical to develop and use agents whose decisions can be understood and explained, both for internal oversight and to ensure the right of taxpayers to understand decisions that affect them.
• • Accountability: Who is liable when an AI agent makes a mistake with consequences for a taxpayer? Defining clear liability frameworks is essential.
• • Skills and training gap: TAs need personnel with new skills to develop, implement, manage and monitor these AI systems.
• • Implementation and integration costs: the acquisition or development of AI solutions, as well as their integration with legacy technology systems, may require significant investments.
• • Resistance to change. There may be skepticism or fear of adopting these new technologies, both internally in organizations and on the part of taxpayers.
• • Regulatory and legal framework: existing laws and regulations may not be prepared to deal with the particularities of an autonomous Agent, requiring adaptation and, in some cases, new legislation

The future is today.

AI agents are not a distant futuristic vision; they are a tangible reality with the power to profoundly transform tax management. Their adoption represents a historic opportunity to build more efficient, effective, fair and citizen-oriented TAs.

However, the road to this transformation must be traveled with a strategic vision, an ethical commitment and a deep understanding of both the benefits and the risks. It is crucial to foster a balanced approach that seeks to maximize the potential of AI while actively mitigating its potential adverse effects.

It is important to generate opportunities for discussion between those responsible for TAs, industry experts, academia and civil society regarding technical and ethical challenges, and to collaborate in the development of good practices, standards and relevant regulatory frameworks.

References cited.

• • OECD. (2021). Artificial Intelligence in Tax Administration. OECD Publishing.
• • Russell, S. J., & Norvig, P. (2021). Artificial Intelligence: A Modern Approach (4th ed.). Pearson.

211 total views, 69 views today