AI Safari

By Chioma Nnyamah

The rapid increase in the quality of Artificial Intelligence over the last decade is the stuff of science fiction. Attempts to use either predictive or generative AI as recently as 2015 often resulted in information that was so error-ridden that the user spent more time correcting it than they would have spent writing their own code or doing a brief literature search for the answers to their question. This is definitely not the case today. The number of available tools utilizing machine learning (allowing them to continuously expand their capacity based on incoming/newly generated data) has ballooned and is now staggering. Whether one wants to use Soundraw to create exciting new music or legal.robot to make it easier to understand the complex clauses in the lease they just signed, there is available AI-based assistance to make almost every task more efficient. 

 

Unfortunately, AI cannot truly be the “worldwide” problem-solver we hope it will be until it is utilizing data and generating results that truly reflect the global population. Most AI is built on software algorithms that are taught to quickly pull details from existing databases. Unfortunately, the amount of available information from the African continent is miniscule when compared to Europe, Asia, and the Americas, creating unintentional bias/unbalanced training. Databases containing correct information on basic metrics such as age, occupation, and health parameters of people from Africa are either limited in scope or not entirely accurate. More complex data banks containing more nuanced information such as individual opinions and cultural practices are almost non-existent.  While this problem did not begin with the creation of AI, it must be addressed if we intend to use these tools to create much needed solutions for the social and economic problems faced by many African countries. 

 

This training deficiency might initially seem inconsequential, particularly because AI will still be able to answer most questions with the information it has. The issues will become obvious when governments, businesses, and individuals start trying to implement these solutions and find that many proposed strategies are not possible because the resources being recommended are not available or the approaches have low compliance. It might be as simple as building materials for a proposed construction project being cost-prohibitory because they are not made locally and the AI simply did not know this. It might also be as basic as a meal plan for a diabetic patient being impossible to follow because it excludes the local foodstuff they have access to. 

 

Direct probes of ChatGPT – one of the foremost text-based AI tools – confirms that its training data leans heavily towards more “developed” nations. When asked “Is your data from the African Continent comparable (in raw amount) to the data you have from North America or Europe?” the AI responds, “No, the amount of data and information I have from the African continent is generally not as comparable in raw amount to the data I have from North America or Europe. The availability of data and information in my training corpus is influenced by several factors, including the prevalence of English-language content and the prominence of certain regions in global news, research, and online publications. North America and Europe tend to have more extensive and detailed coverage in the English-language sources that make up my training data. This is due to the historical dominance of these regions in global media and academic research.”

 

While we cannot expect AI to be completely removed from the rapid development of African nations, it is highly important to take a number of key steps to ensure that AI-generated solutions do not create more problems than they solve long term. Some key things we can do include:

 

  • Increased Creation and Maintenance of African Databases: The information simply cannot be considered if it is not compiled and made available. This field alone could create multiple employment opportunities for tech savvy Africans as the value of this data will skyrocket when large companies look to establish locations/provide their services to people on the continent.

 

  • Stronger Pushes for Continent-Wide Internet Access: Large amounts of preference data is gleaned from daily internet usage. Even if it is collected and analyzed, this data cannot be considered substantial when only 40% (570 million out of 1.4 billion people, Statista-2023) are accessing the internet on a continuous basis. In today’s world, internet access is no longer a luxury, it is a basic necessity and should be treated as such.

 

If used properly, AI tools could easily be the leveler that allows governments and people on the African continent to create and achieve a quality of life comparable to their European and North American counterparts. We just have some work to do first.

Subscribe
Notify of
0 Comments
Inline Feedbacks
View all comments
Left Menu Icon
AI Safari