‘People used to think AI was the terminator robots’

World leaders and tech giants met in the UK this week for a landmark AI safety summit, as the technology becomes an increasing, often invisible, part of our daily lives. On Wednesday, the US and China were among 28 countries, excluding South Africa, to sign a declaration resolving “to work together in an inclusive manner to ensure human-centric, trustworthy and responsible AI”. 

Without this, “there is potential for serious, even catastrophic harm, either deliberate or unintentional” they agreed in the Bletchley Declaration, which also recognised AI’s potential “to transform and enhance human wellbeing, peace and prosperity”.

That frontier, developing AI “for the good of all”, is where University of Cape Town (UCT) professor Deshen Moodley, chair of the South African Research Chairs Initiative (SARChI) in Artificial Intelligence Systems and co-director of the South African National Centre for Artificial Intelligence Research, can be found.

Machines are already contributing to preventive health and becoming “more in touch with emotions”, he says, embracing AI’s advances into the mind-body continuum.

As an example, Moodley tells a story a lawyer shared with him: “My Apple watch told me I had a problem with my heart. I went to see a cardiologist who told me I did have a heart problem and that I was lucky I had come to him (in time).” 

“Low-cost wearables like a watch can detect someone’s mood and emotional state, picking up stress and wellbeing through a heart-rate signal. For ‘human robots’ who are not in touch with their emotions, like me, this could be a wake-up call,” he says,

Moodley’s priority is designing next-generation AI systems to benefit society. His track record — building a successful IT start-up and working for multinational corporates before pivoting to academia — gives him an edge over ivory tower researchers when it comes to turning ideas, and idealism, into reality.  

AI in daily lives  

Moodley focuses on augmented AI — machines supporting people. This is distinct from autonomous AI, which is typically seen as an existential threat.

“People used to think AI was the terminator robots and that’s what we would call autonomous AI. In that case, you create this AI that can think on its own and can act on its own, and the human is out of the loop. That is very scary because then one starts thinking that you’re going to create this other race of machines, and where do we fit in?  

“Augmented AI means the machine is now working in tandem, in collaboration, with a human,” he says. 

“On a daily basis you are using quite sophisticated AI, but you’re not aware of it because there is a nice interface around it. For example, Google Maps is sophisticated AI, but because the interface is so simple and intuitive, one thinks there isn’t a lot of complexity,” says Moodley. “There is such sophisticated machinery going on behind Google Maps. 

“[With the app] there is a real-time ‘pseudo reality’ of the city’s traffic network, which is being updated every second. Every car that is moving has a representation in the model, which is being updated as the signals are coming through. 

I’m looking at next-generation AI, not just theoretically, but how to design systems for social good. I partner with collaborators and bring the latest thinking in AI to the problems they have.

—  AI expert Professor Deshen Moodley

“It does this with the data which comes in from our GPS. If you have Google Maps on your phone, you are not just a user, you upload your location in real time,” he says. “AI is not just a monitoring system, it is a tool to support decisions.”

AI to reimagine a better society  

Moodley says: “I’m looking at next-generation AI, not just theoretically, but how to design systems for social good. I partner with collaborators and bring the latest thinking in AI to the problems they have.” 

AI systems can be designed to radically rethink how society works, for example changing food markets to make sure nobody goes hungry. Proposed changes to markets can then be tested using AI “agent-based” modelling in large-scale social simulations.

In an agent-based model, for example, 10 autonomous actors will be inserted into a virtual environment and told they can do x, y or z.  When the model runs, they interact with each other and the environment following the rules of engagement, and patterns of behaviour emerge.   

This system attempts to mimic how humans would actually act, even accommodating erratic and irrational behaviour. “We can play it out over 100 times and then see what happens to societal structures ... you discover new information and patterns, and understand the dynamics in a very different way.”   

 AI’s agent-based models are a more dynamic system than “trying to fit a person into a mathematical model”. A mathematical model uses averages, what the average person would do, says Moodley, but averages can be misleading.

“There is no average person. People are sometimes driven by emotion, their behaviour is uncertain and they never behave in a linear way,” Moodley says, noting that agent-based modelling takes into account uncertainty and is well established in computer science research.  

The holy grail of AI: expanding scientific discovery  

AI has the potential to enhance society and expand scientific knowledge by automating scientific discovery, he says.  

Scientists with expert knowledge manually pour over data to detect patterns and anomalies to develop theories and models. “The data can confirm theories or they find things that are unexpected and need to be investigated,” Moodley adds.

“One of my big thrusts is using AI systems to automate science. While everyone else is worried about their jobs, I think scientists should also be worried about AI,” he quips, noting however that scientists will be working in tandem with machines on such discoveries.  

 “Scientific knowledge discovery is at the forefront of AI. That is the next holy grail for AI systems.  

 “We can use ‘machine learning’, one branch of AI, to pick up unexpected patterns in streaming data — that is live sensor data being continuously generated — to see if something pops up that looks odd and needs to be probed.” 

Another branch of AI, ‘knowledge-based AI’, takes theories and looks for all possible explanations to support or dismiss them. 

 These “knowledge-based techniques” use logic, expert and background knowledge, and existing theories to analyse data. “If they find multiple occurrences (or patterns), a theory can get stronger and new knowledge can be created,” says Moodley. 

AI tools like ChatGPT 

AI tools also complement a range of disciplines, he says, suggesting ChatGPT is to tech and science students what the calculator was (and is) to students in humanities.  

“People who were not good with numbers got the calculator and Excel, so machines helped them. Now people who struggle to write have got ChatGPT. Traditionally people in computers were not strong linguistically and artistically, and now they can use AI tools.  

People used to think AI was the terminator robots and that’s what we would call autonomous AI. In that case, you create this AI that can think on its own and can act on its own, and the human is out of the loop

—  AI expert Professor Deshen Moodley

 “Some people are using AI tools to learn things they thought they would never understand. Some people are frightened of them, some are using them ruthlessly,” says Moodley.  

His view on AI tools is clear: you can use them for research, but you must check the replies. “Treat ChatGPT as access to knowledge and make sure it makes sense.”

Ahead of AI curve  

Research and development are Moodley’s lodestar and, in 2001, he left the corporate world, taking a large salary cut with the move. “My value system changed and I had a strong development pull.”

From Pietermaritzburg, he grew up under apartheid and attended a public Indian school. “My school friends are still my closest friends,” says the professor.  

• In October 2006, Netflix launched a contest to improve its recommendation engine for viewers with a $1m prize.

• The winner improved Netflix’s own algorithm by 10% and the competition boosted machine learning.

 •This created a trend, where companies “outsource” machine learning problems for prize money or honours, say experts in the field. 

—  AI DRIVES NETFLIX RECOMMENDATIONS:

When he started high school in 1986, he heard about computers and saved up his pocket money to buy one. “Everyone thought I was crazy. It was a second-hand machine and I taught myself programming, making the machine do what I wanted.”

 Computers enthralled him and he studied computer science at the University of Durban-Westville (UDW). In his third year he took a course in AI and became engaged in the field. 

 He did his honours project in AI and his master's in neural networks in 1995, “before most people had heard about it”. Corporate jobs followed his graduation and he was the first “non-white software engineer” on the development team of the first company he joined in Durban. 

 From 1996, he was immersed in the industry, working with big banks and multinationals in South Africa and the UK. As a London consultant he remembers visiting the Chivas Regal headquarters in Scotland to troubleshoot problems with its distilleries, its charms wasted on a non-whisky drinker.  

 “Then I moved back and joined a South Africa start-up. We worked on a farm in Honeydew (on the outskirts of Joburg),” he says, remembering the frantic days and nights of work with Systems Fusion.   

In 2000, the company moved to the US, occupying the entire 18th floor of the World Trade Centre in Denver, Colorado.  “I was director of the company and we were doing very well when I decided, at 27, it was time to move back.” 

When people around the world cannot discern fact from fiction because of a flood of AI-enabled mis- and disinformation, I ask: Is that not an existential (threat) for democracy?

—  US Vice-President Kamala Harris at the AI summit in London, warning that vulnerable populations are already being harmed. 

Collaborative researcher 

Moodley started a PhD in health informatics at the University of KwaZulu-Natal (UKZN) and insisted on teaching first-year students for the initial few years. “My reward was taking kids from the rural areas of KwaZulu-Natal and, after 14 weeks, they could programme in Java. 

“Anyone can learn maths and coding with the right teaching ... and empathy. There is more than one way of understanding,” says Moodley, who prioritised teaching over research in his early days in UKZN's computer science department, which he later headed.   

During his doctoral studies in 2005, he went on a  two-month scholarship to the Massachusetts Institute of Technology (MIT), where he was exposed to cutting-edge AI. There he met Tommie Meyer from the CSIR and, recognising AI’s potential and the need “to create a critical mass” for research in this field, they cofounded the Centre for Artificial Intelligence Research Unit (CAIR) in 2011. 

 Coming from the corporate world, Moodley could “see things differently and felt free to change the status quo”, so the pair initiated the centre’s expansion across universities and disciplines.  

The centre has expanded to eight universities and 11 research groups, and falls under the CSIR and department of science and technology. CAIR co-director, he says: “CAIR is now the flagship programme for building capacity for AI research in South Africa.” 

Moodley is now in Paris on a 10-month research fellowship to Sorbonne University with 14 international fellows, pursuing the design of augmented AI systems. This will allow an exchange of ideas, like “hacker houses” across the ocean in San Francisco, where AI trailblazers hole up together to accelerate innovation.   

“Nobody imagined AI research would grow this fast,” says Moodley, who recently attended a major AI conference in Macau as part of an African delegation. At the vanguard of the continent’s AI revolution, he is resolute that Africa will not be left behind.