14 January 2019

Ethics and AI

Just before Christmas I attended the TechUK Ethics & AI conference in London. It was an excellent event with great speakers and an knowledgeable and interesting bunch of delegates. "AI" was being interpreted more as machine learning than conversational AI/virtual humans, but here are a few of the key take-aways:

1) Kate Coughlan of the BBC presented some research they'd done on public attitudes towards AI - showing which areas excited people, and which they were wary of. Overall them seemed more wary than excited. Interestingly the notion of digital immortality came out as one of the few positives!

2) Ethics (AI and other tech) should be about making sure that people are “safe” even if they don’t care or have the time

3) The Royal Society and  Ipsos Mori also doing interesting public attitudes research. The Royal Society found that "only 9% of those surveyed had heard the term ‘machine learning’" - hence the AI catch-all! The Royal Society has also issues an AI Narratives report.

5) The RS presentation included an image from Boyle's notes (see image at top) which showed "extending life" as one of his top priorities - https://www.telegraph.co.uk/news/uknews/7798012/Robert-Boyles-prophetic-scientific-predictions-from-the-17th-century-go-on-display-at-the-Royal-Society.html

6) Luciano Floridi (https://twitter.com/Floridi) was the keynote. He has a useful 5 factor AI Ethics model which is based on a bio-ethics one.:

  • Beneficence (do only good): Promoting Well-Being, Preserving Dignity, and Sustaining the Planet
  • Non-maleficence (do no harm): Privacy, Security and “Capability Caution”
  • Autonomy (of the human, not the AI): The Power to Decide (Whether to Decide)
  • Justice: Promoting Prosperity and Preserving Solidarity (and eliminating discrimination)
  • Explicability: Enabling the Other Principles Through Intelligibility and Accountability

All in all a great day, and certainly going along next year.

3 January 2019

Virtual Route Learning for Birmingham Community Healthcare NHS Foundation Trust

A recent study by the Birmingham Community Healthcare (BCHC) NHS Foundation Trust found that after brain injury over 70% of patients reported a reduction in their ability to navigate and this had a major impact on quality of life.  If you take a moment to think about all of the cognitive skills you draw on just for a simple journey, this can become quite a complex task for someone with difficulties in all of those skills. In an outdoor environment we often rely on tall distinctive landmarks to help us build a ‘bird’s eye’ or overhead map of the environment or we use local landmarks such as post boxes or shop fronts that help us learn a route using more of a ‘worms eye’ view. 

Prior research had already shown that people used similar navigation strategies in a VR environment as they did in the real world. So, in order to better understand way finding and route learning in people with acquired brain injury (ABI - e.g. stroke and traumatic brain injury), BCHC asked us to develop a virtual simulation that patients and researchers could use – saving a massive amount of time on outdoor, providing greater control, and reducing risk. We created a network of identical streets lined with typical Victorian terrace housing, and gave researchers the ability to drag and drop distant landmarks (e.g. church spires, tower blocks) and nearby landmarks (e.g. pillar boxes, bus shelters) to create different route finding challenges. The researcher could then mark up the desired route with virtual arrows and let the patient learn the route from the landmarks, and then remove the arrows to see how well they could navigate with only the selected landmarks.

Theresa Powell, Consultant Clinical Psychologist at BCHC reported that they found Daden “very customer centred, checking with us at each stage in the development that it was exactly what we wanted.  None of our very ‘un’ technically phrased questions were ever too much for them and if our build requests went beyond the realms of possibility, they were always able to find an acceptable solution with us.”

The system has generated far more data than BCHC had expected. There are now two Doctoral students, two Masters students and two BSc students who have or are using the software in their research projects. Between them they have so far tested around 20 people with traumatic brain injury (TBI) and 30 to 40 controls for various projects.

And the findings so far? One project looking at the impact of contact sports on route learning showed that female American Football players performed worse on recalling certain types of routes than student controls. Another pilot project showed that people with TBI perform worse than controls when only landmarks in the distance were available and BCHC is continuing to gather more data to see if this is supported. Another new project is even combining the virtual route learning app with a fMRI scanner!