On the 30^th day of September 2022, I was notified by the National Information Technology Development Agency (NITDA), the IT standards regulator for Nigeria, of my selection as a member of the National AI Volunteer Expert Group (VEG). The VEG serves as an external working group to NITDA and provides strategic advice, support, research, drafting, collating and review on all aspects related to the formulation and implementation of the national AI policy for Nigeria. The duration of this national assignment is 4 months and commences in November 2022.

For keen observers of the tech policy landscape in Nigeria, this particular effort to see that Nigeria finally has a national AI policy was expected. Recall that pillar 7 of the National Digital Economy Policy and Strategy 2020 – 2030 (NDEPS) issued by the Federal Ministry of Communications and Digital Economy (the Ministry), considers AI as an emerging technology that must be harnessed for the development of Nigeria’s digital economy in order to attain 7 of the sustainable development goals (SDGs) as prescribed by the United Nations. These SDGs are; poverty eradication; good health and well-being; quality education; decent work and economic growth; industry, innovation and infrastructure; reducing inequality; and sustainable cities and communities. These SDGs in my view, are the overarching theme that should guide the formulation of Nigeria’s AI policy. Prior to the issuance of the NDEPS, Nigeria had in November 2021 participated at the international level in the development of an international AI policy framework by adopting UNESCO’s Recommendations on the Ethics of Artificial Intelligence.

Now that the government of Nigeria has taken positive steps to develop a national AI policy, the usual questions would be how to frame this policy to ensure that it serves national interests and benefits Nigerians. One cardinal aspect of AI policy development that is common with the AI national policy in other jurisdictions and/or those proposed by AI policy experts is that AI must be responsible and trustworthy and its development grounded in a human rights framework. In my view, these are the same considerations that should be applied in developing Nigeria’s AI policy and they should be framed around the following key elements which I propose below:

i. Safety and security of AI systems

AI systems should be designed in such a manner that their use will not result in an adverse impact on individuals. In developing and deploying AI systems, the government and organisations should ensure that rules and ethical principles are the foundational building blocks required to trigger the safe and reliable operation of AI systems. Designing AI systems to be safe is to proactively ensure that individuals are not subjected to the unintended yet foreseeable risks that could possibly result from AI use. One of the ways in which this can be achieved is through standard-setting by NITDA in exercising one of its statutory powers or through the adoption of globally accepted AI standards that have been tested and validated by the various stakeholder communities. In addition, such standards must encourage the use of sandboxes where innovative AI systems are tested in a secure environment under regulatory supervision as a pre-condition for final release to the public. Lastly, operationalising safe and reliable AI systems should be strengthened with regular audit obligations to be undertaken by independent trusted third parties and risk assessment in every circumstance where using AI may be categorised as high risk based on the level of impact on individual rights and/or safety.

ii. Law enforcement/national security and military application of AI systems

The application of AI systems to law enforcement and military efforts poses serious challenges to both individual autonomy and fundamental human rights. For instance, the subject (or target) in most of these efforts is unaware of being monitored by AI systems and as a result, has not granted consent. Examples of such use by law enforcement and the military include the use of facial recognition technologies (powered by) AI in public places, surveillance technologies that monitor and/or intercept communications indiscriminately and the use of autonomous weapons systems in the battle space. While such use by AI may be necessary to meet the state’s obligation to protect lives and defend the state. However, this must be done in a manner that respects existing fundamental human rights norms and legally recognised rights, and be subject to robust judicial or regulatory oversight.

Being right respecting means that AI systems must be designed or deployed in such a manner that takes into consideration all the rights guaranteed by law such as the rights to life, privacy, freedom of expression, freedom of movement and freedom from discrimination. While oversight requires a pre-deployment assessment of the potential harms to individuals in using such AI systems, and if deployed, a speedy resolution of complaints by individuals alleging an infringement of their rights through the use of such AI systems. In this regard, it is important to mention that the exemption from the application of the recent data protection bill (yet to be presented to the National Assembly) of data processing operations done for the purpose of criminal investigation is absolutely a bad idea, that can be abused by law enforcement agencies and the armed forces.

iii. AI liability

It is important to bear in mind that the adoption of clear standards governing the safe use of AI systems will not totally eliminate all the harm arising from their use. This harm which could be pecuniary or non-pecuniary ordinarily should help establish the liability of the user and/or developer of such AI systems but unfortunately, this is not always the case. For example, to sustain a claim for liability under Nigerian tort standards, there must be a causal connection between the tortious activity and the resulting damage. However, this can be difficult to establish in situations where the AI system facilitated the harm suffered.

For instance, where the harm suffered is damage to data, an intangible, it may be difficult to argue in court that such an intangible asset qualifies as a physical property that is capable of being damaged. Another limitation of the traditional concept of tort in substantiating liability is the difficulty in proving that the act challenged for causing the harm is a direct consequence of the conduct of the user or developer of the AI system, especially if the AI system involved is self-learning or its initial action has been modified as a result of external data acquired during the normal course of operation.

For this reason, the AI national policy should broadly consider the importance of framing rules for imposing liability especially as AI technologies gain traction and mature in Nigeria. These considerations may take cognisance of the various actors (developers, deployers and users) present in the entire value change of AI and the multiple scenarios of AI use/development that could give rise to liability for which an individual suffering the resulting harm could confidently pursue damages.

iv. Trustworthiness, explainability and fairness of AI systems

AI is trustworthy if it complies with the law and adheres to ethical principles. Explainability means that to trust AI systems, an individual that has been adversely affected by AI decision-making must be provided with an explanation of how and why the AI system arrived at such a decision. Fairness means that AI must not churn output that could discriminate, lead to bias or amplify social inequalities. For AI to remain trustworthy, explainable and fair, AI experts have suggested prescriptive governance mechanisms requiring humans to always intervene or have control over AI systems. Human control is important to mitigate the opportunity for AI in their decision-making, to deviate from what the developers originally intended and/or what was expected by the users. In this regard, the AI national policy may want to consider prescriptive mechanisms that can be implemented to achieve this such as human-in-the-loop, human-on-the-loop and human-in-command. Irrespective of the preferred mechanism, it is very important for the human behind the machine to be properly identified and tasked with ensuring that the operations of critical AI systems are aligned with their required functionalities.

v. Competition, innovation and growth of local content

It bears emphasis to note that very few tech companies have the resources to invest in continuous AI research and development, and because of this AI systems are concentrated in a handful of tech companies from the global north. To ensure that the AI market in Nigeria remains competitive for homegrown AI systems, the AI national policy must obligate specific fair market practices to level the playing field. Understandably, competition law exists in Nigeria but these rules would most likely apply after the harm to the market and/or competitor has occurred. The proposal made here is for the application of specific ex-ante rules to apply within the AI market irrespective of whether market power has been found to exist or not. A suggestion I would readily propose in this regard would be to require AI developers, deployers and users to grant access to their datasets which can be used in the development of AI systems. Public sector organisations that develop AI may also be encouraged to release their AI system to the general public as open source in order to create opportunities for community collaboration that has the potential to accelerate innovation beneficial to Nigerians and to the nation. Alternatively, data governance models like data trusts may be considered if it has a chance of tilting the power asymmetry from AI-powered platforms to users and/or data subjects.

Conclusion

Without prejudice to the foregoing, the AI national policy must also be subject to a robust public consultation that is undertaken when the AI policy is still at a formative stage; adequate information must be provided to enable participants to properly respond; adequate time must be provided in which to respond and lastly NITDA must consider all responses to the consultation. This is to give the final version of the AI policy legitimacy if stakeholders were given an opportunity to contribute and participate in its development. It is also important to stress that all the elements proposed are meant to apply irrespective of the sectoral or cluster application of the AI systems.

AI is globally recognised as having serious potential that benefits the digital economy and society as a whole. This recognition is consistent with Nigeria’s commitment to harness AI to meet the 7 SGDs. However, the flip side of AI use is to realise its socially disruptive nature and the various challenges its application poses to both individual and society. Whilst Nigeria’s national AI policy will no doubt establish a framework on how AI may be developed, deployed and used in order to unlock its benefits, NITDA must as a matter of importance also ensure that the resulting benefits are not overshadowed by the risks associated with AI use. The answer to this will ultimately rest on the approach adopted by the federal government and by extension NITDA in the AI national policy.

This post represents my personal opinion and should not be construed as the opinion of the National AI Volunteer Expert Group or NITDA.