How do we create Artificial Intelligence (AI) that can think for itself while being 100% trustworthy? This next big step in AI technology is the area that Eddington-based researcher Pietro Barbiero is exploring at the Department of Computer Science and Technology at the University of Cambridge.
Born in Turin in Italy, Pietro studied a BSc in Computer Science and Mathematics for his Masters, because it offered the opportunity to study the foundations of machine learning.
He said, “I wanted to do something good and concrete for society, but initially I didn’t know exactly what path to take, so I decided to study mathematics because it can be applied to so many real-world scenarios.”
While doing his Masters Pietro worked in a start-up for the first time, called Biolint (a hybrid word born from biological and intelligent). The research involved applying machine learning to bacterial infections to try to produce a system that identifies the correct antibiotic for specific cases of sepsis, which creates more targeted treatments and lessens the risk of antibiotic resistance.
Pietro works in the field of AI and specialises in Neural-Symbolic AI, which aims to provide the link between basic data perception and high-level reasoning, to usher in a new generation of AI tools that are advanced, efficient and trustworthy.
“The issue of trustworthiness is critical. Many AI models are powerful and accurate but not interpretable, and thus not yet worthy of human trust. Therefore they raise legal and ethical concerns.”
He aims to create a deeper understanding of how these systems function, because the way a powerful algorithm works in making predictions is difficult to fully quantify.
“Most of the time when we are discussing topics we use a form of logic to describe what we see, and describe the causal relationship between different facts and events. This is the structure that allows us to think about things, not simply to perceive, but to reason about things. That is what is currently lacking in AI models.”
Neural networks are currently like sensors. They perceive the environment and then respond to what they see. But it’s just raw feedback – there is no thinking or reasoning about what is being perceived. Researchers are now looking to augment these networks with a deeper level of understanding.
Concept Learning and Down’s Syndrome
Pietro is also working on a project to better understand the causes of different morbidities in people with Down’s Syndrome, such as accelerated ageing and obesity.
He is developing methods to study types of data and how that data is perceived, and then converting them into human concepts, which involves Concept Learning, a new type of science that was only born in 2017.
He said: “I am trying to map the representation these models build into human ontologies. Once we get to this level we can start to create reasoning systems and form hypotheses on how these concepts are interrelated, to eventually enable us to develop new strategies to help people.”
Mapping the body with AI
Pietro is also currently attempting to build a representation of the human body in an AI system, using data to describe what is happening in different organs, cells, and how they communicate via blood and within the human genome.
The system aims to offer the capability to explore what happens for example if you alter the behaviour an organ, and how this impacts the rest of the body.
It is an ambitious project with one of the main hurdles being data acquisition. He said: “Sometimes the bottleneck for progress with this sort of research is budget. Each sample of data to train our models costs $100. To train a model properly you need millions of these data samples.
“Search engines like Google work so well because there is such a huge amount of data available. It’s cheap to generate data about photographs or text. But it’s difficult and expensive to generate genomic data.”
The future of Concept Learning
“The systems I’m working on are forming the preliminary base for something new. We are building the foundations of AI reasoning and there is much work to do to make it scale.”
He is participating in an application by a consortium for a European grant to fund further work when he finishes his PhD (he is in his final year). The strategy is to secure the grant and try to find some more funding to carry on with his current research projects.
He is also collaborating with a start-up based in Cambridge that is working on ‘explainable AI’. It’s called Tenyks (the inverse of Skynet from the Terminator films!) so far funded by two PhD researchers from his department. Researchers at Tenyks are concentrating on improving robustness and reliability within AI networks.
Working in Cambridge
“I love what I do and really enjoy working in this department. It is so friendly and stimulating and the best part is talking to people and walking through the corridors and meeting fellow researchers. It is amazing.
“There is plenty of collaboration and so many visitors from other universities. We get different perspectives even within the broad area of AI from people who are working on different things.”
Pietro appreciates the freedom he is given to pursue the work that most interests him, and the way the University promotes the wellbeing of staff, prioritising a healthy work-life balance.
Living in Eddington
He arrived in Cambridge with his wife during Covid restrictions in 2020 and initially moved into a 15-square-metre apartment. As soon as they were offered the opportunity they moved to Eddington.
“The first time we walked into the apartment it felt like luxury accommodation. Everything was included in the specification… it was also super-convenient because it was so close to work.”
Coming from an industrialised city like Turin, living in Eddington felt like being in the countryside. He loved being able to cycle into the city centre in only 10 minutes. He also appreciated having a supermarket on his doorstep, which he found incredibly useful.
When his wife recently got a job on the southern side of Cambridge they decided to move, a decision that has come with some regrets.
Pietro explained, “We have already noticed things that we miss from our Eddington apartment. For instance the energy efficiency. While we were still in Eddington we turned off the thermostat when we went away on holiday over Christmas, and when we returned it was almost as warm as it was before we left!”
With the exponential growth in AI research, it is thanks to PhD researchers like Pietro that the University of Cambridge maintains its position at the forefront of innovation in this discipline, helping to build a better world in areas such as healthcare.