SPOTLIGHT: SONIA INGRAM

Interviewed by Carl Austin

In our second Spotlight, I've been speaking with Sonia Ingram, Data Scientist at BJSS. Data science is a subject close to my heart. Having hired our first data scientist over 5 years ago now, it's been exciting to see the progression of the team. I very much enjoyed the opportunity to hear more about Sonia's background, interests and working experiences, and have only been able to do half of what we spoke about justice I'm sure.

Background

Sonia told me how her educational background is in pharmacology and biology. To be fair, I'm underselling that quite some way. Having taken a degree in pharmacology, the study of the action of drugs, Sonia went backpacking around the world for a couple of years, returning to work in medical communications for a London consultancy. Not really her cup of tea, she tells me how she returned to education, cranking up another couple of levels. A masters degree in transfusion and transplantation science, and subsequently a PhD in immunology and molecular biology followed.

"During my PhD there's obviously a lot data and statistical analysis, and I got interested in computational biology."

It was as part of her PhD that Sonia was exposed to a data and statistical analysis, gaining an interest in computational biology, which would become her gateway to data science. This interest extended to meetups on the subject, and it was at these that Sonia met her first data science researchers, and set her new career direction.

There are a number of bootcamps for PhD students looking to move into data science. These provide a hands-on way to improve your skills and help jump start a career in data science, typically including time working in industry. Sonia picked 'Science to Data Science', an intense training programme for those coming from a scientific background. 

It was during her time there that Sonia met the BJSS head of data science. She told me how this was a great opportunity for her to gain insight into data science in business and how to kick start a data science career. This change meeting turned into a short internship at BJSS, and Sonia subsequently joined BJSS full time. In her own words...

"I came for a week with BJSS, and the rest is history"

Life at BJSS

Since joining BJSS two years ago, Sonia has worked in a range of data science roles, unusually all with a single client, a retail bank. From data mining and graph data representations to ethical A.I. and audio transcription, Sonia has seen a range of interesting projects. 

"I think I've been lucky really" - Sonia says of the work she's undertaken while at BJSS.

Her current project is to transcribe voice call data and analyse the content alongside further textual information, identifying customers who are potentially vulnerable and may need additional aid. Data scientists are often thought of as working on experimentation and research, however the bank, like many BJSS clients, is productionising their use of data science. This often involves a different set of capabilities and Sonia's team consists of software engineers, platform engineers, architects, business analysts and everything else you would expect from a multi-disciplinary delivery team. Their goal is to deliver production software.

"We work in teams… I think that's what I really enjoy about all the projects I've been on so far. It's not just a Jupyter notebook that does a thing, it's going to shape the business ultimately."

Personally, I believe this is a critical aspect of data science that is too easy to lose sight of. The prize isn't proving the hypothesis alone, it's about turning that into something that delivers great value to an organisation in real-world use. 

We spoke a little about the things that working on enterprise-scale data science has taught Sonia. Things that might be front of mind to a software engineer are sometimes not as instinctive to many data scientists. Effective use of source control, maintainability, balancing efficiency with readability, and even documentation (certainly not one of my strengths!). These are all things that Sonia spoke about improving through working on production data science and within a multi-disciplinary team. She also noted how highly experienced data scientists often don't improve in these areas, working in research and without the support of delivery teams to work with.

It's not just Sonia learning from the team, other team members benefit from working closely with data scientists.

"We had a new head of testing come onto the team, and he was really concerned about how to test an ML model, as it wasn't something he'd ever done before. I think he talked to Sarah (Bryan, a BJSS data scientist and ML engineer) and she told him to treat it like any other software, not to be scared of it."

It's a learning experience for all involved.

5 Year Plan

I asked Sonia what was the greatest challenge she had faced so far. It turned out to be one she was working on as we spoke. On her current project she has been asked to put together a five year plan, an unusual request of a Data Scientist, who often work in shorter windows of thought.

This has involved considering the many ways in which the technology and data can further identify and support vulnerable customers, and how to effectively present a roadmap back to business stakeholders. Due to present to stakeholders only days after our conversation, I've no doubt it will be well received.

Responsible A.I.

"We were tasked with creating the 2nd version of the data science lifecycle for all ML projects at the bank."

A previous project Sonia worked on was as part of bank's responsible A.I. team, tasked with creating a standard data science lifecycle across the bank. This would ensure that aspects such as risk, governance, model monitoring, and algorithmic fairness were all part of a production strength operating model for the delivery of machine learning and A.I. solutions. 

"IThe main bits that I was focussed on were the model monitoring controls, and alongside that we had a project on fairness and bias in ML. So that was really cool, finding out about algorithmic fairness!"

We spoke about how impressed she was with the forward thinking nature of her client, to implement a responsible A.I. team. To place them in a position where they could influence the delivery of A.I. solutions across an entire retail bank. This is clearly an area of great interest for Sonia. I know that she has shared her experience and views with others at BJSS, including myself, on many occasions and has been driving initiatives to share this further. A noble cause.

Towards the end of our conversation I took the opportunity to ask for Sonia's view on intelligent decision support and full scale decision automation. 

"It 100% depends on the use case. Also if it's decision support I think it's so hard to get the exact level of support right … People think this is built on maths, and algorithms, and people trust the ML too much sometimes."

We spoke a little bit about the potential difficulties that can be caused by systems that both make and support decisions through the use of machine learning. Sonia gave a great example of automation bias that was new to me, but really highlights the difficulties of our bias, let alone that of the algorithm.

A system to aid judges setting bail for those on trial became too heavily relied on, with some judges imparting less of their experience and critical assessment in the process and falling back to relying on the algorithm. 

Sonia made an important point here too. Machine learning should ideally compliment the strengths of humans, rather than playing to their weakness. She explained how driverless cars are a good example of this as they stand. They play to our weaknesses by expecting us to be alert, ready to take control at any time, while at the same time, taking control themselves and leaving us nothing to maintain concentration or engagement. It is only natural for a person to switch off in that situation, leading to potentially fatal outcome if something does go awry.

I really enjoyed our conversation, and a something I took away was a comfort that the people who most care about the safe and ethical use of machine learning are the data scientists themselves (or at least this one!).