Artificial Intelligence

State of AI 2019 conference claims UK is the European leader in artificial intelligence for healthcare


More AI start-ups serve the health and wellbeing sector than any other, with one in five focussing on healthcare according to a new report released today. 

The State of AI 2019, launched at the Royal Institute in London this morning, revealed that the UK is the European leader in the sector, with a third of European AI health start-ups founded in the UK, in hubs including London. 

“AI technology has progressed and matured over the last three years in healthcare more than in many other areas as there are profound new opportunities for cost reduction and process automation through AI,” said David Kelnar, Partner and Head of Research at MMC Ventures, who commissioned the report in association with Barclays. 

He told Future London: “Healthcare systems have reached tipping point in terms of need, due to ageing populations, and the cost of treatments. The cost of healthcare in GDP has doubled since the 70s. So stakeholders are willing to embrace innovation to much greater extent than two or three years ago.

“We are now seeing this amazing cohort of bold new entrepreneurs – mixing medical expertise with commercial acumen and a desire to make changes in healthcare at a structural level. People’s idea of healthcare is also expanding, as wellbeing applications are on the rise.”

However, the report also highlighted some concerns.

“There’s a growing recognition around the importance of developing ethical and inclusive AI,” said Kelnar.

“One aspect I’d highlight, around bias, is a particular concern bias increasing social inequality. AI systems learn through training, and the problem is that training data reflects systemic bias from decades around race and gender. 

“It’s important to adopt extensive and rigorous ethical and testing frameworks, with teams of developers to reflect the diversity of the communities they serve, and balanced datasets.”

He also raised the issue of ‘explain-ability’ in these systems. 

“Traditional rules-based software is easy to explain how it works and why a decision was made. The challenge with deep learning in AI is that it is very difficult to look inside and see how they arrived at the decision they did. They tend to give accurate results, but didn’t know how they did it. 

“Being able to explain to people how systems made the recommendations they did, such as if they were denied a loan or diagnosed with condition, is harder.”

The report also highlighted the development of transfer learning as a breakthrough. 

“One of the changes that’s happened in last 24 months has been growth and interest in transfer learning, which is an emerging approach that applies learnt information from previous situations in different but related problems,” said Kelnar.

“Last year was incredible for breakthroughs in language processing using the transfer approach which enabled us to develop systems to interact with real world effectively. Another example is autonomous vehicles – which are impractical and unsafe to go out and generate the training data. But you can simulate environments in a computer and transfer the learning from the virtual environment to the real world. 

“Transfer learning is an important tool to develop systems to interact with real world at scale. Artificial general intelligence systems could do variety of tasks a human could do. We’re a very long way from that today and for time to come. But transfer learning will be an enabler of that progress.”

What is AI?

Artificial intelligence, or AI, is the simulation of human intelligence by machines, and can therefore improve with experience. 

What are the different types of AI?

Machine learning is a sub-set of AI. Instead of the programmer writing a program for a machine to follow, like a list of rules, the program writes the software itself. This allows for more complex problems to be solved as the program can learn, rather than simply follow inputted instructions.

Deep learning recreates the mechanism of the brain in the software through practice and feedback. So the program makes a prediction and receives feedback, modelling the human brain rather than the world. Developers create artificial neurons that mimic the function of neurons in the brain and form a network. This receives an input (such as a picture of a scan), extracts features (such as evidence of a tumour) and offers a determination (such as whether the patient is likely to have cancer.) If the output is incorrect, the connections between the neurons adjust to improve the future predictions. After being fed millions of examples, the connections between the neurons become very sophisticated. 



READ SOURCE

This website uses cookies. By continuing to use this site, you accept our use of cookies.