Oluwafiropo Tobi Ogundare, Red Hat regional sales lead for West Africa and Mauritius.

Nigeria is experiencing significant capital spending on artificial intelligence (AI). 

The country has the highest amount of funding provided to AI companies and the second-highest number of AI specialists in Sub-Saharan Africa, indicating strong investor confidence, a robust and growing talent base, and an overall potential for scaling innovation in the field.

Central to most AI enterprise initiatives is the training and deployment of models, specifically large language models (LLMs) that have brought generative AI to the forefront of public and business interest. LLMs offer a lot of business potential with many different use cases. 

That said, organisations need to know when and how to use them and question whether a particular project benefits from them. 

In many cases, Nigerian businesses should opt for a multi-model approach or instead use smaller ones, all while making AI development a cornerstone of their digital transformation.

A model proposition

What’s important to remember about AI models is that they are an agnostic technology, meaning their usefulness is not limited to a specific industry or business function. AI models transform the traditional organisation on three primary fronts:

• The customer: With the help of conversational AI, businesses can assist and interact with customers in real time. Using machine learning (ML) algorithms, programs can understand and respond to customer queries and direct them to service agent. On top of that, programs can personalise the customer experience and target advertisements based on their search and purchasing behaviour. In doing so, businesses can uncover data trends which can be used to enhance product up-selling and cross-selling initiatives.
• The business: Generative AI tools that can produce content, including text and images, opens up all kinds of avenues for businesses. These tools can even be used to generate software code based on existing enterprise data. Additionally, in processing and evaluating large amounts of data, models can identify trends and patterns which can then be used to inform organisational planning and decision-making.
• The technology: By integrating AI into IT operations, businesses can automate and optimise workflows to unlock new efficiencies. Businesses can automatically generate code, increase the performance levels of their applications, shore up their cybersecurity processes, and get proactive through predictive maintenance.

AI models are not a new technology – LLMs have been around since the 1960s with early ones capable of completing simple tasks like spell checking – but they have evolved over the years and are now not only more advanced than ever, but are more accessible to businesses worldwide thanks to an increased availability of computational resources and infrastructure (i.e. cloud computing). 

That said, LLMs and LLM training can have high resource and energy requirements, and the productivity and efficiency gains businesses unlock may not meet their initial expenditure.

Questions: Go big or go small?

Small language models (SLMs) are a potential and ideal alternative to LLMs in Nigeria owing to the higher computational and energy demands of LLMs. Indeed, SLMs are a practical solution to sustainable AI development in countries where digital infrastructure and access to data sets in limited. 

They are cheaper and easier to train and deploy, offer improved data protection and management, lower latency due to decreased processing times, and can offer comparable or even improved performance than LLMs.

The emergence of SLMs has to the creation of an entire ecosystem of open source models that are available to enterprises and can be deployed. Examples of these models include Meta’s multilingual, text-only Llama models, and Granite, IBM’s series of LLM foundation models. 

The open source nature of these models enables businesses to tailor them to their specific organisational needs. It increases the speed at which businesses can innovate and gives them direct sight of the data and methods used to train the models, and it results in reduced costs from a development standpoint. 

SLMs also enable Nigerian enterprises to take a multi-model approach to their AI development plans, utilising multiple specialised models rather than just a single LLM.

MLOps: Injecting intelligence into enterprise IT

Whether businesses are using either LLMs or SLMs, embracing AI requires them to adopt workflow practices and principles into their IT operations that streamline model development. Machine learning operations (MLOps) helps organisations integrate models into the software development process. 

They achieve this by training teams on how to treat and prepare data, building ML pipelines, and deploying software that follows best practices.

MLOps also entails continuous monitoring and maintenance of data and processes, and they require cross-team collaboration. 

In the same way that DevOps brings development and operations teams together, MLOps brings operations and data teams together to collaborate and complete projects. Using a single, integrated MLOps platform in a hybrid cloud environment, businesses can manage model lifecycles, streamline development, and run them at any scale.

All this goes to show that any organisation in Nigeria can deploy specialised AI solutions. With the right approach, platforms, and vendor support, they can take a step closer to becoming intelligent enterprises.