Steps Ahead

Diversity in AI. How the technology transcends far more than just the industries it is benefiting.

Diane Castelino, Data Science and Research Lead at Mosaic Smart Data discusses her personal experience in the field of computer science both in education and in industry, and how diversity and inclusion have the capabilities to develop it into a truly global asset. 


Over the past three decades, I have had the privilege to witness Artificial Intelligence (AI) transition from the research lab into our daily lives. The recent exponential increase in computing power together with the availability of a huge wealth of data has enabled machine learning (a subfield of AI) into achieving super-human level accuracy in many areas such as image recognition, machine translation and speech recognition. Technology platform giants such as Google, Facebook, Microsoft, Amazon and Baidu have used AI to completely transform their businesses, leveraging their data from multiple sectors and are currently investing heavily in future research in AI.

Today, AI is driving a wide variety of important applications from diagnosing cancer to sensing and building digital twins of everything from bridges to human organs, thus contributing to a deeper understanding of our world. Glimpsing into the future, AI will be building autonomous services and factories which will not only bring enormous benefits but consequences that society must deal with. Given that AI has this potential to be completely transformative, scaling across every industry, every sector of our economy and every aspect of society, it is vital that the people involved in creating this technology are representative of our society as a whole. 

Throughout my career in technology, having spent time in both industry and academia, it is apparent that the field lacks diversity, particularly for the underrepresented. The term underrepresented spans not only gender, race and ethnicity but also socio-economic and geographical background. Ensuring that individuals who understand the technology are representative of society as a whole is proving to be vital in order to mitigate the risk of machine learning algorithms yielding unintentional bias on the basis of race or gender [1], [2]. To address these issues, the last few years have seen the growth of many non-profit organisations focused on inclusivity and diversity in AI and technology [3], [4], [5], [6]. Research is now underway into using AI to detect, analyse, prevent and combat human biases and discrimination [7], [8]. Since 2014, large tech companies have shown their commitment to increasing these numbers and publishing annual diversity reports [9]. Underlining the challenge companies face in recruiting minorities, the percentage of female graduates with core STEM (science, technology, engineering & mathematics) degrees is steadily growing, however, the split is still just 26%. This figure is also translated in the female STEM workforce, with women making up 22%. This shows that some work needs to be done to encourage women to both study these subjects, and transition into the workforce [10]. The purpose of this article is to describe my experiences as a woman in technology. By sharing my journey, I hope to support, encourage and inspire others to feel that this technology is truly exciting and open to everyone, regardless of your identity or background.

Real-world application

For the past six years, I have been working for Mosaic Smart Data, an analytics fintech company at the cutting edge of data science. Mosaic embraces research and innovation, knowing there are mutual benefits to working with universities; Start-ups can draw upon the wealth of expertise and innovation of world-leading academics whilst universities are exposed to the fast-pace and focus of working on real-world commercial problems.

From my perspective, had it not been for continuing research from academia over the last few decades, we would not have the machine learning algorithms that we rely on today. For example, deep learning, convolutional neural networks, recurrent neural networks, reinforcement learning and backpropagation algorithms were all developed in academia over three decades ago. Academia brought these methods into industry, educating the teams behind the world’s leading companies and encouraging the adoption of GPU’s. Thus the importance of research within industry is becoming widely recognised and more prevalent in industry [11], [12], [13]. Keen to stress both the importance and potential of AI to disrupt and transform the financial services industry, I started writing several articles [14], [15], [16].

In 2014, I came across an academic paper that described a class of statistical models designed to capture the behavioural characteristics of non-contractual customers [17], [18]; These were scenarios where companies were unable to directly observe when a customer had defected. This information could then be utilised to model and forecast the expected lifetime value of a customer. This inspired me to take these thirty-year-old statistical models from academic journals and apply them to the real-world settings, taking advantage of both the wealth of data and Python libraries available today. Believing that these models are remarkably effective enabling companies to have a greater understanding of what their customers as individuals are actually worth inspired me to write an article on this topic [19].

Engaging with educational institutions

Mosaic’s collaboration with University College London was initiated in 2014, with Mosaic’s Data Science team mentoring postgraduate students in applying machine learning techniques such as generative adversarial networks and natural language processing to financial problems. Since this time we have also established partnerships with the University of Cambridge, providing technical talks to students such as the application of reinforcement learning to finance. I now hope to create an online Mosaic machine learning lecture series, harnessing the expertise within Mosaic to educate and support students from a wide variety of backgrounds from all over the world.

A strong relationship has formed with the University of Oxford, working alongside Professor Rama Cont, a world-class leader in the field of quantitative finance and who is the Chief Scientific Advisor to Mosaic. Sponsoring conferences on Machine Learning and Finance, and monthly seminars on quantitative finance has attracted a global audience to participate in a discussion on current research in finance [20], [21]. In my opinion, finance can be a particularly challenging application for machine learning due to the risk of misusing mathematical models, the complexities of data handling, overfitting to historical data and adhering to regulatory constraints. Therefore, sharing ideas, particularly with experts in the field, is vital to make progress in this area.

I am currently working on establishing a Research Lab at Mosaic, to provide a development environment for Data Scientists, Academics and Industrial Partners to develop and execute code securely within the Mosaic Environment. My goal is to encourage research in applying machine learning techniques to data; helping financial institutions understand their data through fact-based empirical evidence, rather than judgement and intuition and sharing ideas through publications to promote further research. I want to foster an open, collaborative environment, working with people from different backgrounds, each bringing their unique values and experiences into the company.

Over the last three decades, I have been fascinated to witness the growth of technology at such a rapid pace: products such as personal computers, the Internet, smartphones, cloud computing and graphics processors combined with AI have been completely transformative to a plethora of services. Outside of the technology sector, the financial sector has been the next biggest investor in AI, with fast growth in research and adoption forecasted over the next five years [22].

Looking ahead to the next decade

I am particularly excited about the combination of machine learning with quantum computing, allowing problems that are currently intractable to a modern supercomputer to be handled with ease on a quantum computer. However, this will open up a plethora of new challenges, such as privacy, internet security and cryptocurrencies [23]. Therefore, we need to ensure that this life-changing technology is of global benefit. We need to invite people from different sectors of life and society to come together to address these challenges.

I am proud of my background since it has given me a unique perspective and a wide range of experiences that I can bring to my role as Research & University Engagement Lead, the AI community and the future of technology as a whole.

Beginning the journey into computer science

My parents were born in Goa, India. Having suffered extreme poverty whilst growing up and then losing both parents at sixteen, my father had to leave behind aspirations for further education and seek employment. My mother did not have the opportunity to start school until the age of eleven and also lost both parents at an early age. After my parents married, they moved to Kenya, where my father returned to education through evening classes, working for many years for the Civil Service. In 1968, they decided to move to the UK through a government initiative at the time, in the hope to provide their children with the educational opportunities that they never had. Growing up in the 1970s, as the youngest of five children, I witnessed my parents struggle: seeking employment, experiencing racism, facing discrimination and working long hours on a low income. Despite the endless challenges they never gave up, always inspiring us to work hard. 

My first experience of programming came at ten years old when the early 1980s brought the arrival of the low-cost home computing era, the Sinclair ZX Spectrum in the form of an 8-bit processor with 48K of memory. Spending hours writing code in the BASIC programming language, I was fascinated by the idea that I could write a set of instructions that the computer would then carry out all by itself.

I felt that Maths, Statistics and Computer Science encapsulated all my interests at ‘A’ level, where I combined my studies with working sixteen hours every week at a restaurant. I was keen to study a joint degree in Computer Science & Statistics since even back in the 80s, I believed that statistics played an important role in enabling computers to analyse data. Today, most of the progress in AI is from using deep learning (a type of machine learning algorithm), which, although incredibly effective, are still limited in their explainability. Using probabilistic inference to model uncertainty together with causal reasoning is vital for the field to progress and is especially important in the financial industry due to regulatory constraints. 

Working summer internships with GEC Avionics (now BAE Systems Avionics), I developed code for a pilot’s head-up display unit using the Ada programming language, popular for developing real-time and embedded systems. Following my undergraduate project on the design and analysis of algorithms, I was awarded a scholarship to study for a PhD in Computer Science.

As part of the High-Performance Computing Research Group, I started exploring techniques for tackling intractable problems where the design of efficient, yet effective algorithms was critical. During my research, I came across a paper that utilised ideas from artificial intelligence and over the next few years I developed a variety of techniques to incorporate intelligence into the framework of my algorithms, exploiting adaptive forms of memory and responsive exploration and the results had a profound impact on many important military and commercial based applications. 

To conduct my research, I used a Sun SPARCstation 10, which at the time was the most popular used for research and several years before the arrival of the multi-core processor. I used the powerful ‘C’ programming language which, today underlies most operating systems and programming languages such as C++, Java and Python. My first encounter with neural networks (later renamed deep learning), was also at this time, and although much progress had been made, they were not as competitive as other techniques until we had an increase in computing power and data.

Throughout my research, I never forgot about the importance of education. I found it extremely rewarding to teach undergraduate and postgraduate students mathematics, algorithms, C++ and other languages. At the time, I was completely unaware of being a female in a very male-dominated environment. I knew that I enjoyed technology and wanted to share my knowledge to help and encourage others into this exciting field. 

After completing my PhD, I was apprehensive about leaving academia for a career in industry; leaving a field that I not only enjoyed but also could make a contribution to future research. I felt that it was important that industry should take advantage of innovative ideas stemming from academia and could see mutual benefits if research was carried out in industry. In the late 90s, I worked as a consultant within defence and later finance, for Deutsche Bank and Lehman Brothers, developing code in C++ and later in Java which then became the preferred language for largescale development. I found that building an order management system as part of a team was quite rewarding in that it helped the equities markets transition from people on the exchange trading floor to fully electronic trading, something the market had been needing for some time. 


[1] AlgorithmWatch. [Online]. Available:

[2] “Computer Business Review,” [Online]. Available:

[3] “Women in AI,” [Online]. Available:

[4] “Black in AI,” [Online]. Available:

[5] “Women in Machine Learning and Data Science,” [Online]. Available:

[6] “AI4ALL,” [Online]. Available:

[7] “Diversity in AI,” [Online]. Available:

[8] AI.googleblog, “,” [Online].

[9] “Google Annual Diversity Report,” [Online]. Available:

[10] “Stemwomen,” [Online]. Available:

[11] “Google Research,” [Online]. Available:

[12] “Microsoft Research,” [Online]. Available:

[13] “DeepMind,” [Online]. Available:

[14] Diane Castelino, “TABB Forum,” 9 March 2017. [Online]. Available: Artificial Intelligence for Banks: The Revolution Starts Now.

[15] Diane Castelino, “TABB Forum,” 4 March 2016. [Online]. Available: The Power of Predictive Analytics And Machine Learning.

[16] Diane Castelino, “TABB Forum,” 15 June 2006. [Online]. Available: Big Data – Going Beyond The Hype.

[17] Schmittlein et al., “Counting your customers: Who are they and what will they do next?,” Management Science, vol. 33, p. 1–24., 1987.

[18] Fader et al., “Counting Your Customers the Easy Way: An Alternative to the Pareto/NBD Model,” Marketing Science, vol. 24, pp. 275-284, 2005.

[19] Diane Castelino, “Measuring the value of your customer,” 2014.

[20] “Conference on Machine Learning In Finance,” [Online]. Available:

[21] “,” [Online].

[22] “,” [Online]. Available:

[23] “,” [Online]. Available:

[24] “Office for National Statistics,” [Online]. Available: