It’s been a year since I joined the Alan Turing Institute. The Turing is the UK’s National Center for AI and Data Science and is made of 13 University Partners and a big network of 400 fellows, post-doctoral researchers, research associates, software engineers, data scientists, and operations staff. At the Turing, we undertake fundamental and applied research in data science and artificial intelligence which tackles big challenges in science, society, and the economy. We also organise public lectures, educational programmes, advise the government, and develop open-source software tools.
I was thrilled to join the Innovation team and work across Turing programmes to help translate science into technology. My role is cross-cutting and I think of it as connective tissue between academics, research software engineers and various partners. Even though I had previously worked at an academic and a corporate research lab, the Turing presented a new challenge for me, as it requires having a much more expansive and strategic view of the academic, government and business landscape of AI in the UK (and internationally).
Here’s how I describe what I do: imagine on one side there is the space of academic outputs and AI research and on the other side, there’s the space of the real-world challenges and applications. I am trying to find the function that maps one space to the other and this consists of the scope, people, funding, strategy, data, feasibility, and other factors that we need to consider.
This post is more of a recollection of my learnings, contributions, and thoughts during 2019. These are by no means just an individual endeavor- they are the result of teamwork.
So, I broke down my role into four functions namely orienteering, scoping, weaving and sharing.
Challenge identification and framing is something that I think about a lot especially when it comes to identifying missions that are important to both our academic community and society at large and areas where we can foster meaningful translational activity.
One of my favorite exercises in this was Turing’s AI for Climate Action stream of work. Last year, together with our Turing Fellows from the University of Exeter, we organised a research scoping workshop on Environmental data science. The aim of the workshop was to seed new collaborations and develop ideas for pilot projects on the theme of AI for Sustainability. In the weeks that followed I synthesized the findings, academic literature and reports to inform and write the scope of our funding call for projects on AI for Climate Action. The scope was aligned with UK policy objectives and provided directions for potential projects addressing the role of AI in climate modeling, reducing carbon emissions and building climate resilience.
The winning projects include downscaling weather models to better predict local precipitation events, enabling worldwide solar photovoltaic nowcasting via machine vision and open data, using satellites to understanding the drought resilience of farmer-led initiatives in East Africa, and managing the risk of abrupt greenhouse gas emissions from peatlands. I am happy that I helped shape this agenda and I look forward to seeing these projects grow! These will form the basis of a larger programme of work, about which you’ll hear more in a couple of months.
Throughout the year I also wrote and circulated internally various briefs regarding the landscape, strategy, data, and scope for a new programme on climate change and sustainability. I published a list of environmental intelligence datasets, which covers open data on earth observation climate, water, forests, biodiversity, ecology, protected areas, natural hazards, marine and the tracking of UN’s Sustainable Development Goals. This was inspired by my conversations and the work of Professor Gavin Shaddick and aims to inspire more people to jump into these challenges. I also published some thoughts on creating enabling environments for climate technology and produced some visualisations of climate change mitigation and adaptation technologies which I hope illustrate their breadth and depth.
At the Turing, I am involved in several scoping discussions. Scoping means unpacking the strategic goals, data quality/availability, as well as insights and research questions that are meaningful for all parties involved, and assessing how to best move forward with them.
I’ve found that frequently users confound a challenge that can be addressed by data science, with a proposed solution. Thus, the broader systemic context, informational requirements, and policy/ business goals and priorities need to be identified, understood and separated from possible hype-driven data scientific solutions.
In October 2019, for example, I had the opportunity to work with the Conservation Intelligence Unit of WWF UK to scope out a challenge for our 5-day datathon initiative, the Data Study Group. We used news data to develop an early warning system of infrastructural threats to World’s heritage sites -such as mining and oil/gas concessions. The timely provision of this actionable information is vital to enable WWF and the conservation community to engage with governments, companies, shareholders, insurers, etc. to help limit the degradation or destruction of key habitats. To give you a better sense, 10.5 billion tonnes of carbon are stored in World Heritage forest sites and the Intergovernmental Panel on Climate Change calculated that 17% of yearly CO2 emissions originate from forest destruction and degradation. So, we’d better protect them!
Through the Data Study Group, I found that the process of going from an abstract policy question to a data scientific question is more than just the data or the question. We are concerned with the quality, cleaningness, sensitivity, size, accessibility and the richness of the data. We also want to ensure that the question is challenging but feasible in 5 days, it can be tackled with the data at hand, is interdisciplinary and our partners are capable of using it to inform their decisions.
Early on, I was also involved in our urban mobility project with Toyota Mobility Foundation. This project taught me a lot about consolidating research projects into a challenge led framework with measurable outcomes, the importance of incorporating users early on in the academic research process (which is very obvious in the design world, but not so in academia) and the more nitty-gritty details of urban mobility. I also got to use my design skills to rework our proposition towards an extended collaboration as well as refine project sections related to the responsible use of data in cities.
The project comprised of multiple, interrelated components that tackle some systemic challenges in mobility: a) a platform that enables non-tech-savvy policymakers to manipulate data in order to predict traffic behaviour, b) a reinforcement learning system for traffic lights signal control which can reduce vehicle emissions by reducing delays and congestion c) and new mechanisms for urban planners to understand Mobility as a Service in their cities by creating novel data-driven models of passengers’ mobility choices.
I am currently working on several other projects of varied maturity and where I see my design and systems thinking skills are quite handy. These include some projects with the Met Office (weather and climate forecasting is a fascinating and complex topic and I’m thankful to the Informatics Lab for teaching us so much about it); scaling up and developing service blueprints and policy case studies for our air pollution project with GLA; and mapping the data and trust landscape for our modern slavery project. I will be able to share more about these in the coming months.
I think one of the strongest features of the Turing is its convening power; the ability to bring people together, create stronger knowledge flows and unleash participation from multiple networks — be they in academia, industry, or government. This requires an understanding of these networks, which we do via research mapping exercises, and by getting to know our researchers and partners. A lot of my time is dedicated to this. This understanding is important for aligning our resources and for supporting diverse and interdisciplinary collaboration and experimentation.
Throughout the year I worked with Turing academic leads to organise and deliver various research scoping workshops including one on AI & Data Science for the Environment, one on Molecular biology, and one on Responsible personalisation in the media sector. During these workshops, we want to identify things like:
- What scientific challenges does a particular theme present?
- Do these challenges require new approaches in machine learning and statistical modelling?
- What are the key research questions?
- What is the state of the art?
- What are the interdisciplinary challenges?
We map them in a y-axis (most to least tractable) and x-axis (short term 2–4 years and long term 5–10 years). We then ask participants to pick some of these research challenges and think about:
- Who needs to be involved (skills, expertise, partners)?
- What data is needed? Who are the data owners?
- What resources are required?
- What are the issues relating to privacy, security, and trust?
Additionally, as part of our Economics Programme, I worked with Nesta’s Innovation Mapping team to deliver HackSTIR, a Hack Week focused on applying novel data-driven methods to the problem of innovation mapping. Over the course of the week, the teams looked at a broad spectrum of challenges, including mapping trends in digital social innovation using social media data, predicting the success of research funding proposals and identifying policy themes from raw text. The datasets included social media, policy initiative databases, and funding proposals data. All workshop materials and Innovation Mapping tutorials are available here and I suggest you follow their work which is at the forefront of data-driven innovation policy.
Weaving collaboration in AI research and policy is important in a global context too, especially if we want to support the careful development and stewardship of AI systems. Our Trilateral Canada-UK-France workshop on Safe and Ethical AI is one such example. For this, I facilitated the development of several proposals that resulted from the workshop. I think similar international and transnational incentives and mechanisms should be designed by UKRI Councils alongside the Strategy Challenge Fund and the Global Challenges Research Fund. Different components of these mechanisms can be tested by various international public sector bodies in order to find better responses to the increasing domination of AI by the private sector.
Throughout the year I tried to share my learnings and I wish I could do this more and better.
Some highlights include being a mentor at UNDP’s Data Innovation Lab. There, I worked with local UNDP offices to map systemic barriers to data, and help them develop and refine a project design that includes responsible and meaningful use of data science methods and tools. The purpose of this initiative was to also explore and test new ways of using data for development and co-create the discourse that will shape the next generation of data innovation practice.
This was really refreshing and illuminating as I got to understand UNDP’s challenges better. I highly recommend following the work and guides produced by the UN Global Pulse Labs as well as the emerging work of the UN Innovation Labs. Inspired by this and by my conversations with Cassie Robinson, Sebastian Vollmer and Rayid Ghani I also reflected on what AI for Good means.
Moreover, for a couple of months, I participated in Climate Colab, an online distributed lab by the MIT Center for Collective Intelligence. The goal of Climate CoLab is to harness the collective intelligence of people who suggest local, national, and global plans to climate change. The final report of this lab will be out in a couple of months.
Recently I also participated in a panel on AI and Climate Justice at Neurips conference. This was part of a workshop called “Minding the Gap: Between Fairness and Ethics”. I need a whole another blog to reflect on Neurips and my talk there, but I’ll leave you with Meredith Whittaker’s remark that perhaps we don’t need more academic papers on AI fairness, more than we need tech workers to organize themselves in the workplace.
Last but not least, I was invited by my anthropologist friends Alexa Hagerty and Meredith Root-Bernstein at the Greening Cities symposium, a conference about the ways that we understand nature in cities. I talked about the use of data and machine learning in green and blue urban infrastructures and some of these thoughts can be found here. Through this network of environmentalists, I came across Emanuele Coccia’s work “The life of plants” which is a beautiful treatise on the metaphysics of co-existence and mixture.
The year ahead
There are many other engagements and failed proposals and ideas that never saw the light of day. The Turing can be chaotic but also a really exciting and interdisciplinary place to work. It provides the context and stimulation to think through questions such as:
- How do we better map and understand the AI skills and interests of our community?
- How do we offer directionality in this space? How can we best pick up signals and emerging themes and provide our network with nurturing environments to thrive?
- How can we achieve a more integrated portfolio of projects and increase the knowledge transfer inside our portfolio, but also across the Turing network and across the UK?
- How do we close the loop between methodologically driven AI challenges and challenge-driven methodological work?
- How can we have better public dialogue and public engagement regarding the development of AI systems?
- What is the AI world we would like to see and how can we help shape this narrative?
- How can we better define and capture the impacts of these sociotechnical systems in different sectors and applications?
- How do we ensure diversity and equality of participation in the teams we create?
I also mull over the role of systemic design in this landscape and my own practice. Designers work at the level of the human (and recently the post-human) condition. We aim to bring change in the world by making products, environments, and systems. We like making complex things clear, ‘invisible’ things like power and values visible, finding patterns and interdependencies and exploring/expanding the opportunity and action space. This isn’t exactly how the academic world works, though.
In 2020, I want to bring these closer to my work, refine my practice as an applied research designer, ask better questions, help more people, focus my energy on long term ambitious challenges as they interface with AI & data science (such as climate change) and be less impatient. I got lots of work to do!
Can’t be more thankful to Christine Foster (especially), James Hetherington, Aida Mehonic, Hushpreet Dhaliwal, Adrian Weller, Nico Guernion, Sebastian Vollmer, Gavin Shaddick, James Geddes, Anjali Mazumder, Adrian Smith, David Leslie, Jules Manser, Jon Rowe, Theo Damoulas, Mark Briers, Jon Crowcroft, Radka Jersakova, Anastasia Shteyn, Amit Mulji, Neil Lawrence, Catrin Evans, Flora Roubani, Yiannis Kosmidis, Mark Girolami, Darren Grey, Steven Reece, and Maxine Mackintosh, for supporting and inspiring me in their own unique ways!