The risks associated with using artificial intelligence (AI) systems can be extensive and have severe business impact if left unaddressed.

By Nevellan Moodley, head of financial services technology at BDO South Africa

AI is being used everywhere, across every business sector. In fact, you may even be interacting with it on a daily basis; completely unaware. Not only are the vast majority of us using it, AI is changing the way we do business – and these changes are only set to become more and more prevalent.

The use of AI in business can offer many benefits such as automating workflows, enhancing productivity, and so much more. In many cases the benefits are so significant that they outweigh the risks. However, the risks associated with using AI systems can be extensive and have severe business impact if users are not fully aware of them and organisations do not have rigorous systems in place when considering the implementation of AI systems and solutions in a business environment.

Unfortunately, the risks associated with using AI technologies are often interdependent and risk in one category can trigger a risk in another, creating a detrimental domino effect that can have severe repercussions.

Based on our research, we suggest considering the following risks and mitigation strategies when integrating AI solutions into your business operations.

Data quality, bias and fairness

AI solutions are only as good as the data used to inform them. Poor quality data puts the learning capabilities of the AI system at risk and can negatively affect that business’s strategic decisions, leading to poor and inaccurate predictions, errors, and a failure to achieve the required objectives. This becomes especially pertinent when considering biases during AI system training.

By applying good data hygiene – cleaning and pre-processing the data used to train AI systems, businesses can reduce algorithmic biases and achieve improved accuracy in the results. Appropriate data types and models necessary for training the AI system should always be used and model training also needs to be diverse, broad, and inclusive enough to extend over various scenarios and emphasise on any minority groups present in the datasets.

Data privacy, security and transparency

Data breaches have the potential to create massive reputational damage and expose businesses to potential regulatory fines. This is why it is crucial that sensitive data is adequately stored to mitigate the risks of cyber-attacks and breaches.

Being prepared for cyber-attacks by conducting assessments, listing attacking scenarios and how to respond to them, and documenting recovery approaches is an effective way to address privacy risks.

Because AI algorithms and systems are inherently complex, it is important for businesses to understand how the systems operate and maintain the required level of control. Throughout the process, remaining transparent around AI practices and processes is paramount to enhance trust in both stakeholders and employees.

Ethics and inhuman behaviour (human-AI collaboration)

Concerns around the ethical use of AI are commonplace. In some cases it may seem discriminatory for users who don’t have a clear understanding, and negative societal perceptions that AI will replace the need for humans still persist.

These concerns are understandable and it is crucial for organisations to establish ethical frameworks to effectively understand and address these concerns. To help employees understand that AI can supplement their processes and not replace them, it is helpful to develop training programmes that educate and upskill workers to perform their tasks alongside the AI system.

The human aspect must always be incorporated into the implementation of AI systems to create a collaborative space for AI and human work.

Knowledge, cost and third-parties

Implementing, operating and maintaining any AI system requires a certain level of technical expertise. Without this, organisations could face operational risks. This is why organisations often turn to outsourced third parties for assistance. It is important to remember that extensive dependence on outsourced providers can bring its own challenges such as the risk of inadequate integration and solution failure. It could also result in higher costs.

Establishing clear goals for the required AI solution as well as establishing core components of simplicity and thorough documentation in the deployment of an AI solution can go a long way to regulating the complexity and reliability of the model.

If third party providers are used, clear contractual agreements must be developed in which the scope of deliverables, responsibilities, services, and service level agreements are clearly defined. Vendors must be provided with all the necessary information to successfully implement the business’ required AI solution, and in return must provide the business with the fluency to correctly operate the solution to avoid any problems arising at a later stage.

Environmental and social implications

There are still a number of concerns around the environmental and societal impact of AI solutions. For example, on average, over 113 000kg of carbon dioxide is emitted when training just one AI system. Organisations must prioritize environmental sustainability by safeguarding against the use and training of AI models that rely on the consumption of energy or natural resources beyond what is environmentally sustainable.

AI systems must be designed with consideration and consciousness and businesses must adopt a “AI for good” approach that includes capabilities like measuring, monitoring, and removing the existence
of greenhouse gas emissions which can alleviate the effects on human health and the environment.

Industry leaders have a collective responsibility to actively identify and address the risks associated with this highly transformative technology. There is a place in the future where responsible AI technologies can benefit people, the planet, and still profit substantially. We simply need to navigate the related risks effectively to build an inclusive, responsible, and prosperous AI-driven space.