ARC Cohort 2022

Senseful AI

23rd September 2022

We chat with Dr William Short from the University of Exeter about his venture Senseful AI, a new search engine that allows you to search text by concepts and conceptual relations. This project is part of the ARC 2022 Cohort.

Q: What is your academic background and what are you currently working on? 
I am a classicist and philologist by training. My special areas of research have been the role of metaphor in the Latin language and in Roman cultural thought. So that means that my research rests at the intersection of linguistics, anthropology and the cognitive sciences. I am currently working on a project exploring how Roman culture conceived of courage and cowardice metaphorically. 

Q: How did you come up with your venture idea? Why did you decide to focus on this idea? 
Senseful AI comes out of a tool that I developed to support my research. In my research on Latin metaphor, I found myself facing a very specific research bottleneck – a problem of how to identify figurative relationships between concepts, across textual sources of different kinds, independently of those concepts might be expressed linguistically. However, because no appropriate tool existed, I had to build one myself. And as it turned out the tool I developed to solve my own problem has applications far beyond this particular use case. In fact, from prior research and from the conversations I have had with folks in many different sectors, it is clear that the problem this tool is trying to solve is a long-standing and widespread problem that plagues information retrieval in many domains.

Senseful AI is a search engine that allows you to search text by concepts and by conceptual relations. The basic problem it tries to solve is known as the ‘keyword search problem’. This is the idea that whereas human beings think in terms of concepts and relations between concepts, search engines are limited to literal string matching – placing the burden on us to guess how the information we want to find is actually expressed in words. If you are searching the Internet with Google – or even some more limited database such as a library catalogue, a collection of academic journals, or an online health information resource, often you have to work hard to come up with the right wording that will match relevant information. This can mean sitting in front of your computer trying multiple alternative expressions, and perhaps never finding what you are looking for. Probably all of us have experienced that frustration of keyword searching!

Q: What does your venture aim to achieve and how does it tackle the issue? 
The tool that we are building aims to take some of the guesswork out of searching by understanding the concepts and relations between concepts that we human beings and speakers of language understand, and by recognizing that we approach texts with certain expectations and certain suppositions about how they convey meaning to us. We think about the world in terms of concepts and relations between concepts, not in terms of keywords – and the search engine we are building permits us to find information that is relevant to us in ways that more closely match the way we think.

Q: Who would be your ideal user or customer?  
Any data provider in industry, government, or the public sector offering their clients or customers access to large, unstructured textual data. This could be a fintech business that hosts company profiles, reports, and prospectuses, with users who want to find information across many different documents, where there is not necessarily standardised terminology, and when they do not necessarily know beforehand how that information might be expressed. This could be academic research aggregators, news aggregators or other media publications, whose users again need to find relevant information within highly unstructured documents that may be heterogeneous in the way that certain concepts are conveyed in words. Or this could be legal organisations – for example, a law firm doing due diligence in the trial discovery process, having to sort through and find relevant information within a huge document trove. With keyword search, this could be hugely inefficient and time-consuming. Semantic search, on the other hand, can make information-finding more efficient and more effective.

Q: So at the moment, what stage are you at? Is the tool complete?
I developed the core technology to solve a research problem for myself, so a naively implemented prototype exists. However, this prototype is really just a proof of concept and will need to be scaled up to meet enterprise needs in terms of its performance, maintainability, and so on. At this stage we are seeking funding to transform the prototype engine into a market-ready minimum viable product.

Q: How is the ARC Accelerator supporting you in bringing your venture to life?
The ARC programme has been critical especially for driving the market validation activity we have been conducting. This has meant many, many conversations with stakeholders – potential customers, partners in industry or academia, competitors. I think ARC has been particularly good at getting me out of my comfort zone, to engage not only with sectors where I feel I already understand the challenges and opportunities, but also those where I might not have conceived Senseful AI applying. What’s more, beyond the commercial opportunities the ARC programme has helped reveal, this has led to collaborations feeding back into my research.

Q: What have you learned through the programme that you will bring back to your work at university or research more generally?
One insight from the market validation activity that I think will inform my academic research is to always consider your audience – to think about their needs and interests, or anything that falls in the realm of ‘ethical persuasion’. This is something that I think we can easily lose sight of as academics, especially when we are deeply embedded in our research and can begin to take for granted the interest, utility, or value of that work. I often think of Simon Sinek’s ‘Golden Circle’: start with the why, then the how, and finally the what – almost exactly the opposite of how we are trained to go about constructing academic discourse. But the ‘why’ needs to be front and centre even – or especially – in these contexts.

Q: Where can we go to learn more?
You can get in touch via the email address


Photo credit: Firmbee via Unsplash