Sami Kaski, Professor of Artificial Intelligence at The University of Manchester, says that human-focussed artificial intelligence can play a key role in addressing urgent issues such as climate change.

In recent weeks there has been more than one instance of scientists warning that artificial intelligence developments could lead to a‘doomsday’ scenario akin to somethingfrom science fiction.

This kind of narrative runs the risk of discrediting the huge potential for AI to act as a force for good. We’re already starting to see some of the possible benefits – for instance, at The University of Manchester we have worked on improving the capabilities of robots used in the hazardous work to decommission nuclear power reactors. 

We must take a considered approach that focuses on how AI can be autonomous, but with the primary focus of serving the needs of humanity. We will need much more capable AI, but capable in a way that keeps humanity at the heart of the algorithms.

Human-compatible AI

The sheer variety of human behavior and tasks humans would need help in is one of the major challenges when it comes to AI helping humans. That is why the systems need to learn on the fly. Ongoing work onhuman-robot interaction is a starting point– but work is still needed to design AI that is more broadly useful to humans while leaving them in full control.

AI needs to infer their user’s goals and then recommend actions in a way they understand. This means that they would need models of how human users work in the real world to efficiently collaborate with them, in line with the human cognitive concept called the ‘Theory of Mind’. 

According to ‘Theory of Mind’, humans make assumptions about what others want, think and believe – these are used to infer these states of mind, even though they’re not directly observable.To enable successful human-AI interaction, and human-AI teamwork, we must put these behaviours at the heart of our algorithms.This is a complex idea, but something that scientists are making significant progress towards.

Knowledge sharing

I’ve been working with an international team of colleagues based in the Finnish Centre for Artificial Intelligence (FCAI) and the UK, including the Alan Turing Institute in Manchester, to create “Virtual Labs”, an AI-powered research platform that relies on AI to help scientists and engineers in design and decision-making tasks.

It will help researchers seamlessly design and run computational experiments, operating with physical and virtual assets working in close collaboration with AI-powered research assistants. The trusted assistants will be able to access existing data and ultimately a vast library of knowledge around relevant topics and issues. The aim is that this will accelerate the pace of delivery while, at the same time, improving the quality of research.

While search tools already exist for relevant published knowledge,earlier academic ‘know-how’ is the bottleneck for automation – what we call tacit knowledge – it needs to be seamlessly integrated in the research process. Ultimately it could be easily accessible so research teams worldwide can build on the progress others have made towards the same solution and ensure they haven’t missed someone else’s eureka moment on what could be a critical area of research

This will help fast-track new research and innovation tosupport breakthroughs in diverse areas of study, from the development of new advanced materials to the design of new drugs.

Tackling the serious issues

The overarching aim of our “Virtual Labs” concept is to use AI tohelp human scientists and engineers work towards a “common good”.We believe it has the potential to tackle some of the biggest challenges that humanity facesat a faster pace than before,such as climate change, sustainability, health and societal well-being.

New scientific knowledge is needed more urgently than ever.With the development of “Virtual Labs” what we are looking to do is to make sure we have the most efficient AI-based tools as we just don’t have time to waste in meeting some of the big challenges ahead.

In the future, new innovations will be made in “Virtual Labs” where researchers seamlessly operate with physical and virtual assets to accelerate the pace of delivery while, at the same time, improving the quality of research. 

In those vital areas where we have no time to waste, AI has the potential to be quite the opposite of an existential threat. In fact, it can be a key asset in speeding up progresstowards a safer, more prosperous future for us and our planet.

1. The preprint of the article ‘Virtual Laboratories: Transforming research with AI’ can be viewed here: Virtual Laboratories: Transforming research with AI (techrxiv.org)
2. To find out more about Ai and digital research at The University of Manchester visit: Digital Futures | The University of Manchester