AI AT THE PORT: BALANCING EFFICIENCY WITH ACCOUNTABILITY IN GHANA’S CUSTOMS VALUATION SYSTEM

 INTRODUCTION

The Ministry of Finance's introduction of an AI-driven solution—Publican—for customs valuation marks a significant step in Ghana's journey toward digital transformation in public service delivery. By leveraging artificial intelligence to determine Harmonised System (HS) codes and import values, the initiative promises improved efficiency, consistency, and revenue assurance at the ports.

While the potential benefits are evident, the deployment of AI in such a sensitive and consequential domain raises important governance, legal, and ethical considerations, particularly in how this ambition is being operationalised. In the context of longstanding concerns around discretionary practices and corruption in customs processes, the introduction of an AI-driven solution is likely to generate significant optimism as a tool for enhancing transparency and control.

However, the deployment of AI in public service delivery does not simply resolve one problem; it has the potential to create new ones. In particular, AI systems designed to reduce human discretion may, if not properly governed, introduce risks relating to human rights, dignity, fairness, and accountability. The challenge, therefore, is not only to improve efficiency, but to ensure that such improvements do not come at the expense of the very principles public institutions are meant to protect.

The purpose of this article is to examine these implications within the context of high-impact public decision-making. It argues that while AI offers clear efficiency gains, its deployment in public service delivery raises critical governance, legal, and ethical concerns—including risks of over-reliance, accountability gaps, and data governance challenges—and must therefore be accompanied by a deliberate framework that ensures fairness, transparency, accountability, equity, and public trust.

This requires a multi-stakeholder, participatory approach to AI deployment—one that considers not only technical performance, but also legal, ethical, and societal risks. Such an approach must embed principles of security by design and privacy by design, ensuring that systems are developed and deployed in a manner that protects the public from harm while delivering intended benefits.

At its core, this is not a debate about technology. It is a question of governance—how decisions are made, who is responsible, and how the public is protected when systems fail.

 

 

 

THE PROMISE: WHY AI IN CUSTOMS MATTERS

The use of AI in customs administration offers several compelling advantages. It can:

·        Reduce processing time and administrative bottlenecks

·        Promote consistency in valuation decisions

·        Minimise opportunities for discretionary abuse

·        Enhance revenue mobilisation through data-driven insights

It is also important to acknowledge that concerns around discretionary practices and corruption within customs processes have long been part of the broader reform context. The introduction of an AI-driven solution can therefore be understood, in part, as an attempt to reduce human discretion and strengthen transparency in valuation outcomes.

In a context where efficiency, transparency, and revenue assurance are longstanding policy objectives, the introduction of AI is both timely and forward-looking. However, these very advantages—particularly the consistency and perceived authority of AI-generated outputs—raise important questions about how such systems are relied upon and governed in practice.

GOVERNANCE, ACCOUNTABILITY AND THE RISK OF OVER-RELIANCE

A central concern with the Publican system is the emerging over-reliance and overconfidence in AI outputs, a well-documented ethical challenge in AI deployment. While the system is introduced as an efficiency-enhancing tool, its design trajectory— and the risk that AI-generated values may eventually be adopted without meaningful human intervention—raises important governance questions.

AI, by its nature, should function as a digital adviser that proposes, with the human expert retaining the authority to dispose. It must remain a decision-support system, not a decision-maker. Where this distinction is blurred, the risk is not merely technical—it is fundamentally about accountability.

If an AI-generated valuation leads to economic harm to a business, the question arises: who bears responsibility? Is it the system provider, Truedare Ventures, or the deployer, the Ministry of Finance? The absence of clearly defined provider and deployer obligations risks creating an accountability gap that could undermine public trust.

Equally important are questions surrounding the training data and model integrity. What data has been used to train the system? How has bias been identified and mitigated? Without clear answers, there is a real risk that the system may replicate and scale existing distortions within customs valuation practices.

The issue of algorithmic explainability is also critical. Traders affected by AI-driven decisions must be able to understand the basis upon which valuations are made. This raises broader concerns about the right to information, particularly where decisions have direct financial implications.

Ultimately, the challenge is not the use of AI itself. AI holds significant potential to improve efficiency and consistency in public service delivery. However, where governance, legal, and ethical considerations are not adequately addressed, the system risks being perceived as a “black box”, leading to resistance from stakeholders.

Public acceptance of AI in government will not be determined solely by its benefits, but by the extent to which it is deployed in a manner that is fair, transparent, accountable, and equitable. In this regard, AI is not simply a technological tool—it is a governance challenge.

 

 

DATA GOVERNANCE, PRIVACY AND THE RISK OF FUNCTION CREEP

Beyond questions of accountability and oversight, the deployment of the Publican system raises equally important concerns around data governance and privacy.

The system relies on Bills of Entry (BOE) data, which may include commercially sensitive information and, in certain instances, personally identifiable data. The introduction of such data into an AI system requires careful consideration of how the data is collected, processed, stored, and potentially reused.

At a minimum, the deployment of the system should be preceded by a Data Protection Impact Assessment (DPIA) or, more broadly, a Data Governance Impact Assessment. This is necessary to evaluate risks relating to data misuse, unauthorised access, and unintended secondary use of information. Without such an assessment, the system risks introducing governance vulnerabilities that extend beyond valuation accuracy into the domain of data rights and privacy protection.

A critical question also arises as to the nature of the AI model being deployed. What model architecture underpins the system? Is it a proprietary model, or does it rely on external platforms? More importantly, how is the data being used within that model? There must be clarity on whether customs data submitted through the system will be:

• used strictly for valuation purposes, or
• incorporated into broader training datasets by the provider

If the latter is not explicitly restricted, there is a real risk that sensitive national and commercial data could be repurposed beyond its original intent, raising serious concerns about data sovereignty and control.

This leads to the broader issue of function creep—where data collected for a specific regulatory purpose is gradually used for other unintended or unauthorised purposes. In the absence of clear legal and contractual safeguards, systems such as Publican may evolve beyond their initial scope, with significant implications for privacy, fairness, and institutional trust.

The core issue is not simply whether the system works, but whether it operates within clearly defined and enforceable boundaries. Without strong data governance, even a technically accurate system can produce outcomes that are legally questionable and ethically problematic.

In this regard, transparency must extend beyond outputs to include data usage, model behaviour, and lifecycle management. Public trust in AI systems will depend not only on the accuracy of their decisions, but on the confidence that data is being used responsibly, proportionately, and for its intended purpose only.

LESSONS FROM GLOBAL AI GOVERNANCE: PUBLICAN AS A HIGH-RISK PUBLIC DECISION SYSTEM

The relevance of global AI governance frameworks is not merely in their existence, but in how they help us understand and position systems such as Publican. The key question is not whether Ghana is legally bound by these frameworks, but what they reveal about the nature of the system being deployed and the standards it ought to meet.

From an international best-practice perspective, the Publican system would likely be classified as a high-risk AI system. This is because it directly influences public decision-making with immediate financial and legal consequences for businesses, particularly in the determination of customs values and duties. Systems operating in such contexts are not treated as neutral technical tools, but as decision-shaping mechanisms requiring structured governance and oversight.

Frameworks such as the OECD AI Principles, the EU AI Act, the Hiroshima Process, and the Seoul Declaration converge on a common position: where AI systems affect rights, obligations, or economic outcomes, they must be governed through clear accountability structures, transparency, risk management, and meaningful human oversight. These frameworks emphasise that AI systems must be lawful, fair, transparent, robust, and accountable across their lifecycle.

Ultimately, applying these frameworks to Publican does not constrain innovation. It ensures that innovation is deployed in a manner that is legitimate, trusted, and sustainable within a public governance context. Without these governance and ethical guardrails, the system risks undermining the very public trust and legitimacy upon which its success depends.

In practice, these principles translate into distinct but complementary obligations for the Publican system. The provider, Truedare Ventures, cannot be insulated from responsibility on the basis that Ghana does not yet have a dedicated AI regulatory framework. As a matter of minimum international governance practice, the provider should ensure that the system is built on representative and reliable training data, subjected to bias testing, supported by technical documentation, and capable of providing intelligible explanations for its outputs. It must also be transparent about the system's capabilities, limitations, and appropriate use context. These are not aspirational standards—they are increasingly regarded as baseline expectations for responsible AI deployment.

The deployer, the Ministry of Finance, bears a corresponding obligation to ensure that the system is used within a framework that preserves procedural fairness and institutional accountability. This includes maintaining meaningful human oversight, ensuring that AI outputs do not become automatically determinative without review, providing accessible mechanisms for challenge and redress, and clearly establishing where responsibility lies when decisions influenced by the system result in harm.

The critical point is that AI systems do not carry responsibility—institutions do. Where this distinction is not clearly maintained, governance gaps emerge, and accountability becomes difficult to enforce.

In this context, the absence of a domestic AI statute should not be interpreted as a regulatory void or haven. Rather, it places a greater responsibility on both the provider and the deployer to adhere to generally accepted international standards of AI governance. The issue, therefore, is not one of legal compulsion, but of institutional responsibility and public trust.

 

 

 

 

THE WAY FORWARD: BUILDING TRUST THROUGH GOVERNANCE

To ensure that the Publican system achieves its intended objectives while maintaining public trust, governance must precede automation. This requires a deliberate set of measures.

1. Maintain Meaningful Human Oversight

AI must remain a decision-support system, with customs officers empowered to exercise independent and meaningful judgment, not constrained by system outputs. This requires adequate training in the logic underpinning the system.

At its core, AI should function as a system that proposes, with the human expert retaining the authority to review, override, and decide. The human must remain in control of the system—not the other way round.

2. Define Clear Accountability Structures

The respective responsibilities of the provider (Truedare Ventures) and the deployer (Ministry of Finance) must be clearly defined to ensure that liability is not diffused in the event of harm.

Harm cannot be attributed to the AI system itself. It must be traceable to a responsible human institution or actor—either the provider or the deployer—within a clearly established accountability framework.

3. Ensure Transparency and Explainability

Stakeholders must be able to understand how AI-generated values are derived. This is essential for both trust and effective dispute resolution.

AI deployment in public service delivery offers significant benefits, but without transparency and explainability, it risks resistance. Where systems are perceived as opaque, public trust erodes—not because of the technology itself, but because of how it is introduced and applied.

4. Strengthen Data Governance and Bias Mitigation

The system must be subject to regular audits to ensure that training data is representative and that biases are identified and corrected.

The Publican system should initially operate as a parallel validation system, rather than a fully determinative one. AI-generated values must be tested and validated by customs officers and valuation experts. Transition to full deployment should be based on demonstrated consistency, not assumed accuracy.

Deploying the system without this validation risks introducing systemic errors at scale, with direct economic consequences for businesses—particularly in the absence of a transparent and effective redress mechanism.

5. Protect Procedural Fairness and the Right to Challenge

Affected parties must retain the ability to question and appeal AI-influenced decisions, supported by accessible and meaningful explanations.

This is particularly important in situations where businesses may be required to comply with higher valuations determined by human officers despite differing AI outputs. Without clear mechanisms for challenge and review, the system risks functioning as a “black box”, undermining fairness and legitimacy.

CONCLUSION

The introduction of AI into Ghana's customs valuation system represents a significant opportunity to modernise public service delivery. However, efficiency alone cannot be the measure of success. The real test of this initiative will lie in the strength of the governance framework that underpins it.

AI is not the problem—its governance is. If deployed without adequate governance, AI risks undermining the very objectives it seeks to achieve. When deployed within a framework that prioritises fairness, transparency, accountability, and equity, AI can enhance trust and improve outcomes. Without such a framework, even the most advanced systems risk resistance—not because of what they offer, but because of how they are implemented.

The absence of a domestic AI statute should not be interpreted as a regulatory void or haven. Rather, it places a greater responsibility on both the provider and the deployer to adhere to generally accepted international standards of AI governance. The issue, therefore, is not one of legal compulsion, but of institutional responsibility and public trust.

Ultimately, the effectiveness of AI in public administration will depend on the governance structures and change management processes that guide its use. AI is not just a tool of efficiency—it is a governance system that must be deliberately designed to protect rights and build trust.