James Butterworth is a PhD student in the Department of Computer Science at the University of Liverpool. He is researching how neuro evolution – a digital version of Darwin’s theory of evolution – can be used to harness artificial intelligence (AI) across a range of applications, including drones.
James appeared on the University of Liverpool Podcast to discuss how neuro evolution is helping to drive advances beyond the reach of human capability, as well as the potential pitfalls and security issues that lay ahead in a future where robots operate autonomously.
5 insights to help explain the potential of neuro evolution
- Bringing Darwin into the digital world
Neuro evolution is a term that describes the digital equivalent of Darwin’s theory of evolution. It is being used across a range of fields, but is easiest explained through a gaming example.
When researchers programmed a computer to play the game Super Mario World by itself, the computer had no information other than a single goal (known as a ‘fitness function’): move the character to the right as far as possible.
Over several attempts, the computer had mixed success. But by using a genetic algorithm, it selected and combined its two most successful performances to produce an improved ‘player’. When you do this process several hundred times, the results can be startling.
- The potential is vast
James’s interest in neuro evolution was sparked during his undergraduate degree in artificial intelligence. One of the things that fascinates him most is the speed of improvement.
For an early project, he worked with a team of colleagues to create simulated ‘fighting agents’ that had a field of view and could fire bullets. Over time, they began firing their bullets with intelligence and dodging incoming bullets.
“By 200, 300, 400 generations, you’ve got an agent that can perform far better than if you were in control,” he says. “That was like ‘well, if it can learn to do this automatically in the space of three or four hours, then what else can we teach these things to do?’”
- It is creating real-world solutions for challenging problems…
As a PhD student, James’s effort is focused on using neuro evolution to help power the activities of real-life drones. In the event of a natural disaster, such as an earthquake, he believes drones could be ‘trained’ through genetic algorithms to “traverse the entire plane as fast as possible”.
Computer simulations, he says, would provide the early work, preventing real drones from crashing into buildings and each other during their early generations. But once the simulations are at an appropriate level, the AI can be transferred to operating drones.
For anyone trapped in a disaster zone, he says a swarm of drones “might be something that’s quite orderly, that won’t upset you or scare you as much, or it might be something quite terrifying and things moving around really fast, trying to find people really efficiently”. Regardless, it will be something that finds you quicker than a human.
- … And pushing beyond human limitations
In all kinds of fields, including medicine – where AI systems are starting to diagnose illnesses more accurately than doctors – James is adamant that neuro evolution could have a revolutionary effect.
“Rather than just saying ‘that’s good enough, I think that’ll work, that’s fine’, let’s just push it and really see how far it can go,” he says. “There are lots of examples where AI has come up with a lot better solution than humans can.”
- As advances continue, security will become critical
Despite his enthusiasm about AI and neuro evolution, James admits he does have concerns about its potential use in the future.
“If all of this software becomes open source and all of these ideas become more common knowledge, then there is the potential for people with less ethical intentions to play around with these things and basically train them on whatever they want to train them on.”
About this podcast
The University of Liverpool Podcast aims to bring listeners closer to some of the academic experts, authors and innovative thinkers from the University who, through their in-depth analyses, research and discoveries, are affecting positive change in the world today. Each episode features one or more of our academic experts discussing research in their specialist field. Subscribe to the University of Liverpool Podcast via iTunes, Tunein and Google Play Music (US and Canada only).
Interested in learning more about the University of Liverpool’s online Computer Science programmes? Find out more about the MSc in Advanced Computer Science or MSc in Software Engineering.