Dr Kwabena Nuamah

Image

Innovation Project

Creating a Compositional, Hybrid and Interpretable AI Framework

Three key pain points faced by most businesses across many industry segments are:

• difficulty in integrating data from diverse dynamic sources,
• difficulty in generating insights from such data, and
• the lack of transparency in AI models used to support decision-making.

Kwabena’s research plan over the next three years is to explore interpretable inference methods for question answering by learning to automatically compose algorithms.

This is motivated by the increasing demand for interpretable and explainable AI systems (systems whose internal decision-making processes can be inspected and understood by humans) in many domains including financial services, health, and law enforcement.

Although recent methods in automatic question answering leverage the statistical relationship between words in text, they are limited and unreliable in tasks involving multi-step reasoning, diverse data sources and complex quantitative operations.

Kwabena plans to create a framework for “whole system reasoning” with explanations where the automated system learns to reason not only about specific predictions task, but also about how it dynamically combines different intelligent operations, and how it does this under data access constraints. This will involve developing a new technique for combining logical and statistical methods and a unique class of hybrid (neuro-symbolic) inference algorithms where the symbolic and sub-symbolic inference mechanisms are equally vital and work together. This requires learning a formalism that can translate between symbolic and vector representations (and vice-versa) as well as meta-level inference over operations or programs. Additionally, he will explore human interaction in automated inference such that both the human and AI can learn from each other.

This research vision will lead to a richer range of interpretable solutions than is offered by current pure generative models alone, and the development of novel interpretable AI methods by combining different AI approaches. Current pressing needs for AI interpretability make this research direction practically relevant and timely, with the potential to have a significant impact in society.

Transcript

Michaela Turner: I’m Michaela Turner from Edinburgh Innovations and I’m joined today by Kwabena Nuamah, who is one of our new Innovation Fellows. Welcome and thank you for coming along today.

Kwabena Nuamah: Thank you very much, Michaela, I’m really excited to be here and very excited to be part of the first cohort of Innovation Fellows here at the University of Edinburgh.

Michaela: It’d be great to talk about your journey to date. So what has your career been to date and what’s brought you here?

Kwabena: I’ve been working in the software engineering space since about 2004 when I was at university and I’ve been looking forward to sort of exploring the use of technology in various application areas. During that process I got to work with companies like PwC, started a software company in Ghana called My Systems, but then I realised there was a lot more to explore and that brought me into research, especially in the area of artificial intelligence and I’ve been working in research in AI since then.

Michaela: So what drew you to the role of being an Innovation Fellow?

Kwabena: The idea of being able to take artifacts from research and translating that into an application that’s useful to society is the main motivation and driver for me.

Michaela: You’ve talked a little bit there about what you’re trying to achieve with your Innovation Fellowship. Is there more of a broader project that you’re looking to focus on within it.

Kwabena: So currently most of the research, or most of the activities around AI, in general, has been on how businesses adopt the technology within their specific domains. But then there’s a lot of talk also on how do we do this in a safe and reliable way. In a way that’s trustworthy, ethical and also inclusive.

Michaela: Building on from that, what do you think has been your innovation journey since you’ve joined the University.

Kwabena: We had an industry partner who was interested in exploring the use of our technology which was a system called FRANK. It stands for a functional reason for acquiring novel knowledge. That is, it takes a user’s question, applies different kinds of rules to decompose the problem into sub-problems that can be solved by other agents. Those could be functions, statistical methods or other AI systems. And pull all these together to give an answer but then make sure that the answer is explainable to a user.

But I realised very quickly that the idea also applies more broadly across different industries. And so I got into thinking, okay, how do we translate this beyond the research that we’ve been doing into something that can be applied and used by others in the business space.

Michaela: How has EI (Edinburgh Innovations) helped you on this journey?

Kwabena: So the first thing that EI helped with was ground the research into an application area, into something that’s actually useful for someone. So EI allowed me to start thinking about the customer, who would use such a solution if it were commercially available. But then also EI helped in providing resources to help build certain skills around that. So the commercial skills. Thinking about how would you build a business? What would that business look like? What are things you need to put in place for a business? A viable business that would be based on such a technology. But then also exposing you to other people outside the University who could add additional support, but also help you feed or sort of access sounding board to feed back to you what’s working or what would not work based on the ideas you’re thinking about.

Michaela: What were the major funding sources that you’ve used to get here and what different programmes have you found relevant to help you go on this journey, alongside the support from EI?

Kwabena: One of the things that I got to benefit from, from EI, was the Venture Builder Incubator. It was like a starting point. That opened up the opportunity to look at other sources of support. And EI was very instrumental in that, for instance, they you know informed me about the IKEA programme, which helps also in translating academic research into business or commercial products. And through EI’s help I put together a good proposal and got accepted into the IKEA programme which is quite competitive.

Michaela: Can you tell me a little bit about the project you’re working on now and also about what impact are you hoping it’ll have in society as well?

Kwabena: The idea is how do we build systems that are a bit more compositional. That is, you don’t need to depend on one large system that does everything, but rather can depend on a variety of subcomponents which can be orchestrated in an intelligent way to result in an output that’s explainable but also very transparent to the user. So in a practical sense, it’s about giving choice and flexibility to businesses so they can adapt whatever systems they have to new ones, but then be able to leverage different AI systems rather than having to be forced to use just one.

Michaela: What do you feel the impact on society might be from this?

Kwabena: I realised that across government agencies all the way down to private enterprises, there’s a huge demand for systems that are explainable. The users of the system don’t want to feel out of control of the tools they’re using. So as much as they can get you know high throughput, increased efficiency, that fundamental need for the human to be in control of the decision that’s been made, it’s still quite paramount. So I think the work we are doing hopefully helps to bring some of that control back to the user, while still leveraging all the capabilities that we can get from AI systems.

And for all the executives we spoke to across different financial services organisations, education sector, health, there was that constant “but” whenever I asked are you using AI or are thinking of using AI, it’s always, yes, we want to use it. We are keen to explore it but we still need to figure out how we do this in the right way. We still want to be able to control how it is used. We still want to be able to understand how it works. So we think this line of research would have huge impact on how companies adopt AI and AI related technologies.

Michaela: What’s your ambition for your innovation project? What are you hoping to do in the the longer term with this project?

Kwabena: So our goal is hopefully to spin out a company using the innovation we started off with our FRANK project, but then there’s a lot more to explore in that space also. So the new research direction that we are looking at would help to uncover new solutions, hopefully with new technologies that are constantly being developed. But more broadly, thinking through how we help end users of such AI systems to adopt them in the right way by providing tools that are transparent in how decisions are arrived at things, tools that also allow them to see any kind of bias in the systems, and also help in the ethical use of AI technologies.

Michaela: What most excites you about the next step in the journey?

Kwabena: I think for me, it’s the fact that you’re able to think of the problem in a slightly different way. That’s the benefit that you get from research. You can think beyond what’s currently available, what’s currently being done. So instead of building big language models, can we use the research that we’ve done, that we’re doing in automated reasoning to create smaller inference systems, reasoning systems that can leverage as many different language models as possible. That’s quite exciting in the sense that you’re creating new things from things that are already available.

Michaela: What would you tell somebody at the start of this journey and start of their career? What advice would you give them?

Kwabena: It’s very easy during research to think solely on the research project and getting the outputs that are expected as deliverables from that project. But taking a step back and understanding the wider context in which the research is being done is, I think, as vital as the research itself. But then also to understand that the impact that comes from research might not necessarily just be in the papers that are published. It could be in an application of that technology in a real world scenario. So having conversations with the translational teams within the University, EI in particular, as early as you can. I mean it doesn’t matter which stage of that project you are on. The fact that you have an idea that I think could be applied in a commercial context, that’s a good state at which to speak to someone from Edinburgh Innovations.

It also positively impacts the actual research you’re doing because then you have more tangible motivation for doing what you’re doing.

Michaela: Thank you again for joining us. It’s been really great to have this conversation and hear more about your journey and I’m excited to support you in the next steps.

Kwabena: Thank you very much, Michaela. It’s been a wonderful experience being part of the Innovation Fellowship. Thanks so much for the opportunity to be part of this first cohort of Innovation Fellows and thanks also for all the support that has been provided from EI over the past couple of years. It’s been really helpful in shaping a lot of what I’m currently doing and things I’ll be doing in the future. Thank you very much.

Research Experience

• University of Edinburgh, School of Informatics, Artificial Intelligence and its Applications Institute

  • Research Fellow
  • Postdoctoral Research Associate
  • Doctor of Philosophy (PhD in Informatics)
• Chongqing University - Master of Engineering (Software Engineering, with Honours)
• University of Ghana - Bachelor of Science in Mathematics and Computer Science
• Brainnwave Ltd - Data Scientist
• Cognate Systems Co. Ltd - Co-founder and Software Engineer