AI is now a strategic priority for government across West Africa. Guided by policy documents such as the African Continental AI Strategy, many now recognise the technology’s potential to improve service delivery and promote inclusive access for citizens, backed by investments in digital public infrastructure.
Meanwhile, many countries are adhering to globally recognised regulations or establishing their own and are even in the business of building their own models.
In Nigeria, for example, the government has partnered with the private sector to develop an open-source language model for local languages.
AI has the potential to bring the region’s public sectors and institutions into the future, relieving them of obstacles caused by limited infrastructure, manual processes and outdated systems.
That said, AI is not a one-size-fits-all solution, and its implementation does come with its own challenges, especially as projects move from the proof of concept to the production stage and organisations need to start scaling and extracting value from their deployments.
But the future of AI in the public sector isn’t defined by the models you use. It’s how you use them, and for any organisation, that’s the right place to start.
Modernising IT and putting data to good use
It’s important to keep in mind that when we refer to AI or ML, we are using a single term to describe a whole host of technologies, everything from generative AI (GenAI) and intelligent automation to natural language processing and optical character recognition.
These technologies form the basis for government agencies to leverage any data, structured or unstructured, that they have access to.
They also represent the chance for agencies to modernise their IT, culminating in use cases across all kinds of departments and services.
Sitting at the heart of all these technologies is software that enables AI applications to communicate with large language models (LLMs) and generate responses on their data.
This is a process known as inference and it is the operational phase of AI. However, running complex LLMs can be expensive, and running inference can end up accounting for the majority of an organisation’s budget.
That’s where software solutions like Red Hat AI Inference Server come in. Powered by virtual large language model (vLLM), it provides fast and cost-effective inference by maximising infrastructure (i.e., GPU) utilisation, is certified for all Red Hat products, and can be deployed across other supported Linux and Kubernetes platforms.
Modernising IT involves keeping operating costs as low as possible, and how government agencies run AI models becomes part of that effort.
Agencies can also use AI to modernise legacy systems, identify bugs in existing code and move to newer programming languages. From there, it’s a free-for-all.
In government procurement, models can collect information about potential suppliers and select preferred ones based on strictly outlined criteria. In law enforcement, pattern recognition can help agencies identify suspects and track specific threats (within the confines of relevant privacy regulations).
In healthcare, AI can analyse patient data to streamline approvals and research processes, as well as collate data to identify broader health trends or issues in the general population.
Across any use case, the advantages of AI are constant. Agencies can deliver more effective services and better allocate resources, while citizens enjoy an enhanced and more seamless experience when interacting with those services.
At the same time, improved data distribution and collation means agencies can streamline dependent processes and better predict citizens’ changing needs, thus increasing their agility and overall service availability.
The cost question
Deploying and maintaining AI models can be an expensive exercise, with cost variables encompassing GPU clusters and other hardware, human expertise and ongoing fine-tuning.
Even in the face of avoiding potential exorbitant licensing fees by training and deploying their own models, costs for organisations can escalate as they have to fork out for computation, storage and talent.
According to one survey published by Benchmarkit, 80% of enterprises miss their AI infrastructure forecasts by more than 25%, while 24% miss theirs by 50% or more.
Model costs can also be unpredictable without centralised governance. Agencies and development teams benefit from having a streamlined experience for managing AI models, achieving improved productivity and consistency, as well as guaranteed lifecycle management.
In terms of reducing expenses associated with model deployments, government agencies in West Africa need to choose the right model for the right job.
They need to evaluate the complexity of the task the model needs to perform and then fine-tune it. By doing so, they elevate the model’s efficiency and optimise its cost.
Agencies should also utilise monitoring tools that are typically available on cloud platforms to track resource usage. They should also prioritise open-source tools and platforms that enable greater flexibility and tailor-made solutions.
AI by the people, for the people
As the global market for AI models matures, organisations are focusing less on training and more on operations. In other words, the focus is now on being able to take a model and apply what it’s learned to the real world.
That is the difference between AI training and AI inference, and by transitioning between the two, government agencies can start to unlock value from their projects.
Inference does have its unique challenges, particularly as model complexity increases and hardware evolves at an accelerated pace. Agencies should note that models are only as cost-effective and high-performing as the inference server.
What makes a solution like Red Hat AI Inference Server impactful is its ability optimise model inference across hybrid environments, as well as its open source nature.
vLLM is the standard in enterprise AI inference, and the versatility it offers through open source means agencies can tailor it according to their model needs.
When it comes to AI in the public sector, it needs to be developed by the people, for the people. With the help of platforms that centralise model development, management and governance, institutions in West Africa can put their best digital foot forward, all while keeping costs in check and delivering reliable and consistent services.
