Squiboon
Lead Research Organisation:
University of Bristol
Department Name: Computer Science
Abstract
The ISCF Healthy Ageing Follow on Fund 2023 is part of the Healthy Ageing Catalyst Award programme, funded by the UKRI Healthy Ageing Challenge and in collaboration with the US National Academy of Medicine's (NAM) Healthy Longevity Global Grand Challenge. With Zinc's programme of support, the Follow on Funding will be used to build upon previous Healthy Ageing Catalyst Award funding, with the aim of commercialising research in the use of conversational AI in care settings to tackle isolation and loneliness.
Much progress has been made in recent years with the use of Large Language Models, such as GPT-3, and these are now seeping into the mainstream and leading to a fundamental shift in the interaction between humans and computers.
However, further research is necessary if these advances are to be of benefit to the ageing population. In addition to the accessibility issues of those with deteriorating eyesight, hearing, memory, or other impairments, there are new challenges resulting from biases and deficiencies in the pre-trained models.
We'll be exploring novel and inclusive ways of communicating, using best practices in accessibility, using simple graphical interfaces in combination with natural, conversational audio.
We'll be testing conversational AI in group settings, exploring how speech idiosyncrasies of older adults, sound quality affect error rates in speech-to-text conversion, and the ability of speech recognition models to tell the speakers apart (diarisation).
We'll be looking at the emotional content of conversations and testing computers' abilities to track the emotional arc of a group discussion.
Finally, we'll be gathering new audio, visual and textual datasets to custom-train the artificial intelligence to be more inclusive and representative of older adults.
These insights will all be used to underpin additional product development that has arisen from an iterative design process with our steering group of older adults, following the previous Catalyst Award. This product will be deployed across retirement communities and care homes, to explore the effectiveness of the product in tackling loneliness in communities, as well as to identify and address implementation challenges.
Much progress has been made in recent years with the use of Large Language Models, such as GPT-3, and these are now seeping into the mainstream and leading to a fundamental shift in the interaction between humans and computers.
However, further research is necessary if these advances are to be of benefit to the ageing population. In addition to the accessibility issues of those with deteriorating eyesight, hearing, memory, or other impairments, there are new challenges resulting from biases and deficiencies in the pre-trained models.
We'll be exploring novel and inclusive ways of communicating, using best practices in accessibility, using simple graphical interfaces in combination with natural, conversational audio.
We'll be testing conversational AI in group settings, exploring how speech idiosyncrasies of older adults, sound quality affect error rates in speech-to-text conversion, and the ability of speech recognition models to tell the speakers apart (diarisation).
We'll be looking at the emotional content of conversations and testing computers' abilities to track the emotional arc of a group discussion.
Finally, we'll be gathering new audio, visual and textual datasets to custom-train the artificial intelligence to be more inclusive and representative of older adults.
These insights will all be used to underpin additional product development that has arisen from an iterative design process with our steering group of older adults, following the previous Catalyst Award. This product will be deployed across retirement communities and care homes, to explore the effectiveness of the product in tackling loneliness in communities, as well as to identify and address implementation challenges.