Positioning South Africa for the AI boom
Terence Singh, key account manager at KPMG Matchi, a global matchmaking platform that connects financial institutions and other large companies with emerging technology solutions, says AI is not yet mainstream in South Africa.
“We are in its infancy in adoption, maturity and education. That doesn’t mean we’re backwards, though,” he says. “What we’re seeing currently is a mix of small companies using an AI engine to create a specific application, in which case big companies buy that service from them, and companies like IBM that have their own AI (in this case, IBM Watson), and they work with companies to build specific solutions using the AI technology they already have.”
Harmony Mothibe, co-founder and operations director at BotsZA, says adoption has been slow and unpredictable compared to the US, China and Europe. “But the adoption of AI in general has been steadily increasing in the last two years, particularly in the field of Natural Language Processing (NLP). Financial institutions are among the early adopters of this technology to help provide exceptional experience for their customers and improve customer support. The next two years will see exponential growth in the field of ML as companies are trying to find their feet in this fast-paced field.”
AI solutions are being used in South Africa in everything from curbing rhino poaching to running customer service chats, and Singh believes we’re only limited by our imagination in terms of the role AI can play.
He says, however, that when he lectured on the Fourth Industrial Revolution, students started out excited by the lecture and ended up depressed, hearing about technologies already available that can replace what humans do. “I tell them it’s their fault,” Singh says. “They’re demanding frictionless services and wanting everything to happen faster. This is the consequence.”
For example, South Africans love to moan about dealing with call centres, but are we willing to accept job losses resulting from replacing inefficient call centre agents with an AI system that can cut call waiting time, optimise efficiency and reduce costs? “There is always a price to pay,” Singh cautions. “No job is not at risk.”
While he believes AI will yield new jobs, as automation has over the last century, there’s no place for unskilled jobs in the future. This is potentially catastrophic for South Africa.
One potential solution, Singh says, although unpopular, has proven effective for China – ban the technology, while at the same time throwing money at developing it locally. Singh notes that the western world is hypocritical in this regard, shunning the idea of banning something, but at the same time levying extreme import tariffs, which curtail demand in the same way. He cites South Africa’s prohibitive import duties on imported vehicles as the reason it became a world-class automotive manufacturer.
“We could put a 10-year ban on certain kinds of technology while pushing money into raising our education and outcomes in that area,” he says. “At the moment, we’re throwing money at the problem of education without the outcomes. The solution is not difficult; it just requires political will.”
Ian Sharland, national IT governance and cybersecurity manager at Bidvest Data, disagrees, arguing that given access to technology, it would be almost impossible to police. He suggests the real solution lies in fixing South Africa’s education system, starting with mathematics and language skills – the foundations for understanding and developing algorithms.
AI applications in South Africa
Sharland believes that the public perception of what AI can do sometimes outstrips the reality. Having worked on AI/machine learning projects previously, he says the major limitation of AI is that an algorithm can’t think laterally. “Humans are still the best problem-solvers,” he says. “We can take discrete sources of data and draw connections between them. We solve problems naturally, without thinking about how we think. It’s difficult to break that process into a series of rules for an AI to follow.”
Many great discoveries have come from a person superimposing wisdom learned in one context onto a problem in another.
“When it comes to AI, our imaginations are far ahead of where we thought the technology would be by now,” he says. “But on the other hand, we don’t fully grasp what the technology is currently capable of, so we’re not yet exploiting it to its fullest extent.”
AI also requires enormous amounts of data, and in a format that’s usable. Sharland explains that AI is essentially a probability engine and relies on mathematics, so any problem needs to be articulated numerically.
Given this, he says he can’t understand why financial auditing is not yet handled by AI. The data is readily available and already in numerical format.
In his own line of work, Sharland says AI is used in cybersecurity in antivirus, through what’s known as heuristic virus detection. “Virus detection used to be predominantly signature-based,” he explains. “The virus created would be stored in a file with a unique numerical signature. Files could be blocked if they matched that signature. But virus creators are smart and learned to build algorithms that subtly change the file – not the functionality, but the unique signature. Using AI to assess similarity, antivirus detection can gauge the probability of a file being the virus and block it.”
Singh notes that call centres, chatbots, automated sales calls and help desks are the most obvious applications for AI technology locally. But the possibilities are endless. Advice-driven industries, such as financial planning, are also being targeted, and Singh says there are a number of simple, everyday applications ripe for exploration – like using AI and shopper behaviour data to create a service that predictively draws up a shopping list for you, gets your approval, purchases the goods and arranges for their delivery at your home. These technologies and data streams exist – they just haven’t yet been exploited in South Africa.
Mothibe says that companies with the most data – or, rather, meaningful data – and the right skill sets have more opportunities than those without. “Skills shortages and lack of education around AI and its application are the limiting factors to adoption in the South African environment,” he says.
BotsZA has seen the impact of its chatbots in HR, particularly in recruitment, where many repetitive processes can be automated. “We have developed a chatbot called Hazie, which is a job placement bot that uses NLP and ML to connect job seekers to their dream jobs and also help recruiters source the best talent,” Mothibe says. “Hazie handles frequently asked questions, candidate screening, welcome aboard, interview scheduling, etc.”
Who’s using AI in SA?
CLEVVA: CLEVVA (with emphasis on the last two letters – VA, which stand for Virtual Assistants) helps to deliver consistent, compliant customer service experiences across staffed and digital self-serve channels. The company offers one platform with a host of VA apps, designed to let non-coding teams design, build and maintain these.
Merlynn: Merlynn aims to replicate and scale human expertise using AI. Its proprietary AI technology, TOM (Tacit Object Modeler) is housed in a process that allows non-technical experts to work through a series of steps, culminating in the creation of a virtual expert – a tangible, robotised version of the expert’s expertise.
DataProphet: An advanced ML company focused on developing and deploying bespoke solutions for Industry 4.0, particularly in manufacturing. The company says its AI, OMNI, can guide your production team to eliminate defects, scrap and minimise downtime.
BotsZA: Specialising in applications powered by AI, ML, automation and Internet of Things (IoT), BotsZA offers chatbot solutions for a range of industries and applications, from travel (hotel reservations and flight bookings) to e-commerce, banking, insurance and customer service.
NumberBoost: A data science studio that builds custom AI solutions using computer vision (e.g. licence plate identification or people counting), NLP (e.g. sentiment analysis or automated FAQ-answering chatbots) and data science (e.g. recommendation engines or fraud detection).
IBM Watson IoT Platform: Companies can use the Watson IoT Platform to create a range of solutions in NLP, ML, image and video analytics and risk and security management. Applications span the full gamut. For example, MTN, together with Wageningen University (WU) in the Netherlands and Prodapt, have been using IBM Watson as part of the MTN Connected Wildlife Solution to help predict threats and combat the poaching of endangered rhinos at Welgevonden Game Reserve. Rogerwilco, digital marketing agency, plans to use IBM Watson to create data-led marketing solutions for its clients.
Aerobotics: Marketing itself as a global precision agriculture company, Aerobotics tree crop protection for farms. Using aerial imagery and ML algorithms, it helps farmers to identify pests and disease early on.
AI will take some jobs, but make others more effective
Rather than replacing human decision-making, Sharland believes AI will do away with the drudgery involved, freeing people up to focus on the actual problems that matter, and to move towards more creative environments.
This brings us back to Wöcke’s idea of AI either heralding opportunity or loss – depending on the response. His philosophy centres around differentiation and IP as essential elements of what makes us human.
His approach to AI (and that of Merlynn) posits that rather than using AI to mine historical data in search of answers, we should leverage human expertise by teaching human wisdom and experience to AI. “Some AI experts believe the truth is in the data, but when I’m faced with a pressing problem – like a medical issue – the first thing I think of is, ‘Is there someone I trust I can call?’ Imagine if we could replicate the expertise of those people.”
Wöcke believes by identifying the best human experts in any area and teaching AI to mimic them – from plumbers to insurance assessors to neurosurgeons – bottlenecks can be alleviated in access to services. Imagine cloning the best financial planners in the country and then being able to replicate what they do in a half-hour consult thousands of times over within the space of a second.
While this can boost revenue for companies and boost access to important services like healthcare, education and legal advice, it does leave behind anyone with skills deemed irrelevant. Wöcke stresses this is why the focus in education should be on teaching innovation, problem-solving and differentiation, rather than a cookie-cutter approach.
“I’m an expert at being me. The question should therefore be, ‘What do I know that people find useful?’ Then looking at ways to scale that. We need to be asking what we can do better than anyone else, than as well as anyone else. Generally, that boils down to taking better decisions in a specific field.”
To prepare for tomorrow, he suggests positioning oneself to benefit from AI. “It’s upon us. And we know that not making a decision is a decision to go backwards, so we need to start identifying our IP assets and how we can scale those. To survive the future, you need to get very philosophical.”
Getting practical
Wöcke notes that having an AI strategy in place is now expected. Unfortunately, he says, everyone tends to apply the same technologies in the same ways in business. “The corporates in South Africa are all using RPA – robotic process automation. And they think because it’s got the word ‘robotic’ in there and the promise of automation, it’s AI. It’s not. But it is the door that AI will typically come in through into a company. Building on top of RPA, companies need to be asking, ‘How do we do this more intelligently?’”
Sharland’s advice for companies wanting to prioritise AI as a competitive advantage is to make it someone’s job. “Give it to someone; measure them on it; give them the budget and the freedom to think,” he says. “Innovation requires navel-gazing, so there needs to be room for that. Give them the space and protection and put the correct incentives in place to manage their output, not their inputs.”
He notes that companies also need to carefully consider what they mean by innovation. Everyone says they want to innovate, but in the South African context, that is used to mean cost-cutting. “Innovation requires investment,” Sharland says. “How many companies on the JSE have R&D as a line item in their financials? And if you’re not measuring innovation, how can you manage it?”
He suggests that innovation starts as a strategic decision by company leadership, but requires a long-term view by investors and management. Innovation is not immediate. Companies need to measure their performance in different ways and not just on quarterly results.
The big question: where to from here?
According to Wöcke, the question now is, “When AI disrupts itself – what is that likely to look like?”
He believes that the next phase of AI will focus on empathy ethics, being more people-centric and democratisation. “When you look at historical data, there’s no ethics there. Do you put it in as a cold rule? Or, do you point at someone who has fantastic client relationships and an impeccable record and say, ‘Let’s be like them’? The second big point is to break AI down to be usable by the mainstream employee so that they can actually create value for the organisation using AI. That’s what’s going to allow businesses to take their ‘secret sauce’ and create scale. In my world, and I’m very biased, AI should be about creating scale,” Wöcke says.
Defining AI
According to Nick Polson and James Scott, authors of AIQ – How artificial intelligence works and how we can harness its power for a better world, what we should think of is an algorithm. “An algorithm is a set of step-by-step instructions so explicit that even something as literal-minded as a computer can follow them,” they write. “On its own, an algorithm is no smarter than a power drill; it just does one thing very well… But if you chain lots of algorithms together in a clever way, you can produce AI: a domain-specific illusion of intelligent behaviour.”
They go on to explain that AI systems tend to follow a “pipeline-of-algorithms” template, considering data from a domain, performing a chain of calculations and outputting a prediction or decision. They also highlight the fact that AI algorithms have two distinguishing features:
· They deal with probabilities, not certainties (e.g. an AI agent at a bank might note that the probability of fraud on a transaction is 92%).
· Unlike traditional algorithms (like the ones that run word processors), the instructions are not fixed ahead of time by a programmer. The AI learns these directly from its “training data” (e.g. the AI will analyse historical data for fraudulent and non-fraudulent financial transaction sets and find the patterns that distinguish one from the other, essentially training itself what to do based on data and the rules of probability).
