AI systems bring together information so they can learn and find solutions to problems. These tools can ingest vast amounts of data, which is why AI has become such an asset in the field of science. One of the most significant breakthroughs of AI in research, which earned a Nobel Prize, was AlphaFold, developed by Google Deep- Mind. This deep learning algorithm can predict the structure of proteins and can be used for drug design, and has saved many years of work in research labs.
For Professor Benjamin Rosman, from the School of Computer Science and Applied Mathematics at the University of the Witwatersrand (Wits), this promise led to Wits’ recently launched Machine Intelligence and Neural Discovery (MIND) Institute.
Rosman says he’d been thinking about launching an AI Institute at Wits for about five years, but it wasn’t until October 2023 that the concept finally started gaining momentum.
Rosman and Wits vice-chancellor Professor Zeblon Vilakazi were at the Science and Technology in Society Forum in Kyoto, Japan, and decided it was the right time to establish an AI research hub at the university. Rosman remembers AI being at the centre of so many conversations at the conference – from healthcare to education – that they both realised how well positioned Wits was to establish itself as a research leader in the field. In November 2024, the MIND Institute was launched. As the founding director, Rosman says the institute is focused on the fundamentals of AI and “actually building AI itself”.
Fresh insights
He adds that many AI models are modelled on neural processes in different parts of the brain, and he doesn’t want to replicate this at the institute.
“One of the MIND fellows is looking at how we model parts of the brain that aren’t already being modelled in a lot of these computational processes. Specifically, they’ve been looking at the brain stem. If we can build models based on different parts of the brain, can that give us fresh insights into more efficient ways of doing computation?
“In conversations about the MIND Institute, people often assume that we’re just building a massive facility with a million computers in it, but we’re trying to challenge this perception. Many believe that if you throw more computers, more data and more money at AI, you’ll be able to solve more complex problems. But we’re saying that’s not a scalable solution. We’re trying to bring bright minds together to come up with new, clever ways to do more, but with less.”
It seems everyone he meets these days is “doing AI”, Rosman says, but this is not necessarily the case. “They’re taking other people’s stuff that’s been developed elsewhere, whether it’s ChatGPT or another tool, and are applying it to their problems so that they can find solutions.” He says there are some researchers developing AI in Africa, for Africans, but many are using downstream AI tools, which means there’s a gap in what this technology can deliver. “If we don’t do something about it, the gap is only going to widen between the people developing it and the people consuming it.”
Communication barriers
The MIND Institute plans to focus on developing AI and the application of AI to solve problems, while also asking questions around ethics, policy and societal impact. “We’re really hoping to do all three, but everything is grounded in having a strong group of people thinking about AI as a science itself,” says Rosman.
While MIND doesn’t currently have a physical home on the campus, the goal is to bring together researchers from fields like psychology, neuroscience, neuroanatomy, philosophy, linguistics, evolutionary sciences, engineering, computer science, ethics and statistics, among others, to get a broader picture of common problems and to think about how to solve them from a fresh perspective. Scientists and researchers working across other disciplines in many cases often don’t have access to the latest AI tools. Rosman says they’ll often be using tools that are out-of-date, which means that they may be doing things inefficiently. By creating a group of people with a common interest, the MIND Institute serves as a platform where fellows can turn to each other for help.
These academics are already doing amazing things in their disciplines, says Rosman. “If they’re working in a fancy neuroscience lab and they’re busy studying and cutting open brains, we’re not planning on equipping them with improved computing or building them fancy new labs. We’re not trying to replicate what exists already.”
What they are trying to do is bring together academics from different disciplines, break down the communication barriers between them and inspire them to think outside the box by learning from each other and learning to speak the same language.
This is also important when considering ethics. “If we’re worried about the impact of these technologies on our lives, and on the world in general, we must make sure that all kinds of voices are in the room when we’re developing it.”
The hope is for this group, which currently includes 34 fellows from across the university, to be provided with a small amount of grant funding, from MIND, so they can collectively work on innovative projects. And if only one or two of the projects leads to something interesting, then MIND has achieved what it set out to do. “I like to describe it as taking all the smartest people we can find, putting them in a box, shaking it up and seeing what magic comes out,” Rosman says. “We’re not looking for innovations that are just a small, incremental improvement on something that exists already. We want to encourage people to try radical things and then we’ll provide support to anything that looks vaguely promising.”
As for funding, the MIND Institute is backed by the university, and has also received financial support from IBM, Google, and others. While the Google partnership is still in the very early stages, the university’s relationship with IBM is well established. “Wits and IBM have been research partners for years,” Rosman says, adding there are IBM staff visiting researchers in Wits departments. The idea is to draw on their expertise and collaborate with them on a range of different academic activities.
Catastrophic forgetting
Rosman says many AI researchers keep building bigger models with more compute power and more data. We’re even seeing nuclear energy being considered for AI datacentres because of the massive amount of electricity needed to power AI workloads, he adds, describing this as totally unsustainable.
Turning to the natural world for solutions, Rosman says that a group of fellows is looking at sleep and memory consolidation and asking if AI models need “rest”, too. Part of the project is driven by a phenomenon called catastrophic forgetting, which occurs when a neural network in an AI system forgets previously learned information while learning new tasks. “So, if we look at what sleep does for living creatures, can rest assist with memory consolidation for AI models? This is a great example of how we’re trying to use cross-disciplinary ideas to rethink and hopefully improve AI.”
“If we don’t do something about it, the gap is only going to widen between the people developing and the people consuming it.”
Benjamin Rosman, Wits
Dr Victoria Williams is a postdoctoral fellow in the School of Computer Science and Applied Mathematics and a MIND Institute fellow. With a background in anthropology, neuroanatomy and psychology, Williams is passionate about intelligence and the brain, and she’s currently using her expertise to better understand AI systems. One project she’s working on is around machine psychology and machine empathy. “We’re studying the extent to which large language models (LLMs) can show empathy. In particular, we’re looking at something called cognitive empathy, which is our ability to understand someone else’s mental state.”
Williams and her fellow researchers have been running detailed psychological assessments on LLMs to understand their strengths and weaknesses based on how they perform when completing standard empathy tests. And, thus far, their research shows that the language models outperform humans, including psychology students, when it comes to cognitive empathy.
“The aim of the MIND Institute is to get people thinking about problems from different angles and perspectives,” says Rosman. “There are so many different world-class researchers and research groups at this university and many of them don’t know the other exists. By pulling all these people together, we can really lay the foundations to take the cool things they’re already doing to the next level.”
* Article first published on brainstorm.itweb.co.za
Share