Where does the continent fit into the chatbot conversation, and why it needs its own startups that focus on AI.
By Tiana Cline
The ability to be able to have a conversation with someone else is a fundamental part of what it means to be a human being but today, it’s becoming more and more confusing to figure out whether you’re talking to a bot… or not.
Technology has advanced to the point where AI-powered conversational chatbots, like ChatGPT or Meta’s BlenderBot 3, are deceptively realistic. (More so when they go off the rails and say things like ‘I want to be alive’, which is exactly what Bing’s AI bot expressed earlier this year.) While chatbots like these are trained
to avoid expressing personal or biased views, and are known to make factual mistakes and mix up information, they’re becoming smarter, more streamlined and sometimes, a little strange…
“Conversational AI is one of the first applications of NLP in artificial intelligence and this kind of technology has existed since Eliza in the sixties,” explains Professor Mpho Raborife, the deputy
director at the Institute for Intelligent Systems at the University of Johannesburg in South Africa.
Natural Language Processing, or NLP, is a way for computers to understand human language. NLP uses AI tools like machine learning and neural networks to process and interpret large amounts of data.
“With the rise of big data, ChatGPT was inevitable. It’s trained on a very large dataset and the technologies behind it make it a lot more advanced than other conversational AI applications.”
One of the reasons for this, explains Raborife, is that ChatGPT uses reinforcement learning alongside sophisticated algorithms that manipulate its dataset. “It’s so sophisticated that people don’t realize that it really possesses no intelligence whatsoever,”
she adds, “it really doesn’t understand the meaning behind anything, it just know how the words are supposed to work but not what they mean.”
OpenAI’s ChatGPT is a research experiment gone wild that has shown the incredible potential of conversational AI yet Africa is still being left behind. Consider smartphone voice assistants like Siri or Alexa – both examples of conversational AI used regularly; they work with English but struggle to understand local accents. They don’t support African languages.
“We know that the Global North is not going to include our languages. Therefore, we need to find a way to make sure that we create an environment that the startups that come from Africa which do incorporate our languages have the same kind of impact,” Raborife adds. Duolingo, for example, is a popular language learning mobile app that uses an AI system called Birdbrain to personalize its content. Duolingo has 43 different languages (including made-up dialects such as Klingon, High Valyrian and Esperanto) but only isiZulu and Swahili are available when it comes to African language learning courses.
“When we talk about AI, our languages are not supported or even represented by these technologies,” says Raborife. “But conversational AI has the potential to bridge the language divide as it can be used in a variety of ways because it surpasses literacy levels. If people cannot read or write, they can use their voice to access services and this is something that is important to Africa.
This is why we need our own startups that focus on AI.”
Recently launched, Lelapa AI is an African AI research and product house looking to make a real difference within the continent with a special interest in agriculture, education, energy, healthcare, microfinance and language.
“There are so many problems on the continent that AI-type technologies can help with so there’s a responsibility for us to use our skills to help create better lives for everyday Africans,” says Pelonomi Moiloa, Lelapa AI’s co-founder and CEO.
“What we intend to do is use technology in a very smart, resource-efficient way because there are challenges on the continent like rolling blackouts, water shortages and network issues…” Lelapa AI have already debuted an app called VulaVula which offers NLP-as-a-service for under-represented languages.
Moiloa describes VulaVula as an underlying technology that supports other technologies. A banking chatbot, for example, could use VulaVula’s AI language capabilities to help customers interact in their own languages.
“The idea is that if an organization has a place for us to plug our models into and we can collaborate agreements around how that work can be done, then – ideally – if we have the resources and capacity we will help businesses do what they need.”
For Moiloa, creating African data sets to work from may be challenging but it’s also something she sees as a competitive advantage for Lelapa AI. “We’re having to think around clever ways of dealing with Africa’s data scarcity problem. We have to create models that are able to cope with less amounts of data, but models that can get really good at curating and gathering data,” she says.
And the fact that ChatGPT doesn’t necessarily draw its information from African data sets (it was apparently trained using real human conversations) means startups like Lelapa AI have the opportunity to develop responsible methods of dealing with problems that are relevant to the continent.
“I don’t think we should be trying to develop our own ChatGPT but maybe we can develop other textual tools that can help solve the problems that ChatGPT is meant to – but at the moment, it seems to be creating more problems,” she ends.