Technology Salon


Sponsored by

a discussion at the intersection of technology and development

How Ghanaians Can Engage with and Create Generative AI Solutions

ghana genai developers

The release of OpenAI’s ChatGPT and similar chatbots like Google’s Gemini and Anthropic’s Claude has re-ignited the discussion around how we can manage and direct artificial intelligence (AI) solutions. A common challenge with these and other AI and machine learning (ML) systems is their Silicon Valley heritage that often excludes the lived experiences of Africans – including Ghanaians.

At the recent Technology Salon Accra on How Can Ghanaians Manage and Direct Generative AI Solutions?, we asked deep questions like:

  • How can we create more space for Ghanaian participation in these models?
  • Who should participate from African communities? When, how and why?
  • Would meaningful participation improve GenAI accountability?
  • What would greater accountability look like? What might it achieve?
  • Where is the role of government or civil society in managing GenAI?

Our lively discussion engaged multiple Ghanaian and international experts, where the following themes emerged.

Current GenAI Solutions Exclude Africans

The large language models we use daily are based on the corpus of English language thought published on the Internet. This is a huge content set – terabytes of data at last count – yet, like the Internet overall, it is under representative of Africa in general, and many of its countries, specifically.

Yet, there was an effort to deliver structured Ghanaian data to OpenAI that could’ve been included in its ChatGPT models. A team in Ghana was collecting and submitted this data through a data crowdsourcing effort. Sadly, OpenAI shut down that process early in the ChatGPT lifecycle.

This leads to Generative AI solutions that can do wonders, within the general confines of Western approaches published in English. Missing is data that showcases African culture, customs, and data that could increase accuracy and improve results, especially when looking at African problems. Worse, there are many instances of visual Generative AI producing tropes and stereotypes long ago debunked.

How to Localize Large Language Models?

We do have access to untrained Large Language Models that we can then train on a specific data set, and thereby get responses in line with the training data. These models are affordable too – in the direct technology costs – yet can be prohibitively expensive when you include the human resources cost to hire and manage the talent needed to train and deploy the solutions.

The human resources costs are high for any organization seeking to utilize, or even regulate AI. A recent report highlighted how the UK and European governments cannot compete for talent with the private sector, and they pay orders of magnitude more than low- and middle-income governments, especially African ones.

If a Ghanaian organization were to get the human talent to develop and deploy an LLM, what then? For example, Ghana’s largest newspaper, Daily Graphic, is building an LLM to look at its corpus of news data, but this isn’t a free tool supported by advertisers like the newspaper itself. The Daily Graphic LLM will be behind a paywall to capture the model’s value to users.

When Open AI is worth $80 billion – more than BMW, Dell, or InfoSys – we need to talk about value creation. We also should we asking:

  • Should Ghanaians be adding their data to LLMs? Or is their data being shared regardless of their opinion on the matter?
  • If they do, what happens to the ownership of that data? Is there any way to track the data’s use or misuse? What to do if it’s misused?
  • How is that data governed, and which data protection laws apply? Is there a way for LLM’s to “forget” data it ingested?
  • Who gets a share of the value created? How do we value the data or remit payment for it now and in the future?

What Does GhanaAI Look Like?

Let’s say that multiple Ghanaian organizations follow Daily Graphics lead and create Generative AI models based on their data, or data they can access. Or maybe a national movement to create a GhanaAI? Is being a consumer of a more localized LLM be the pinnacle of innovation for Ghana?

Might there be aspirations to be a GenAI exporter? A country with the skills and expertise to provide GenAI goods and services to other countries or regions? A GhanaAI could be a powerful way to lead West Africa and the world in this new technology.

Will the forthcoming National AI Power Policy propel Ghana to the forefront of technology-focused countries? If so, then there is a group often forgotten that needs to be central to GhanaAI – civil society. Ghanaians themselves need to understand the options and opportunities of this new technology and keep it from perpetuating or increasing existing biases and inequalities.

Civil society already has tools like the AI Readiness Index to assess their country’s status, and there are efforts like the African context sensitive AI solutions to ensure GenAI is responsive to local concerns. Finally, we also need to keep a human in every decision loop.

Still, the real question for any GhanaAI solution, the test to see if it’s truly reflecting local conditions, is to ask it who makes the best jollof rice.

One Response

  1. Elizabeth Njoroge says:

    I enjoyed the discussions and thoughts on how to progress generative AI in Ghana but more so providing insights for Africa.

    With alot of awareness on various use cases of generative AI and probable use of your own data is a game changer.

    I am hopeful that we can achieve more in Africa by looking at the opportunities generative AI has reather than barriers and blockers.

    Thanks Wayan