
13 minute read
OP News AW23: Embracing the Era of AI
from OP News AW23
by PerseADO
EMBRACING THE ERA OF AI
In recent years, advancements in artificial intelligence (AI) have propelled the field into a new era of innovation and possibilities. OPs and Friends of The Perse share their perspectives on the AI revolution in the context of their respective industries.

Neil Sardesai (2021)
I AM A THIRD-YEAR MEDICAL STUDENT at the University of Cambridge, intercalating in engineering. At The Perse, I studied maths, further maths, chemistry and biology for my A levels. At the same time, I also volunteered on the palliative care ward at the Arthur Rank Hospice Charity. My love of medicine grew and I realised that I wanted to pursue a career where I could combine my passion for problemsolving and my desire to help others.
I am fortunate to be working as a PRISE Research Fellow at Harvard University in the field of medical artificial intelligence. This opportunity has been a dream come true, as it allows me to explore my passion for bioinformatics, data science and machine learning in the context of medical research.
Under the mentorship of Dr Manrai, an assistant professor of biomedical informatics at Harvard Medical School, I developed machine learning models that improved the accuracy in estimating kidney function beyond the capabilities of current clinical equations. We also demonstrated the impact of removing race from kidney function calculations, illustrating how this adjustment would lead to increased diagnoses in Black adults and reclassification to more advanced stages of kidney disease. This would consequently allow for earlier access to specialist treatment. Furthermore, I am working on adapting these equations to better suit the British population, with the goal of enhancing diagnostic accuracy within the NHS.
Alongside this research, I am also working as a clinical data science intern at the Cambridge Centre for AI in Medicine. In this capacity, I am analysing temporal trends in the PaCO2- ETCO2 gradient (the difference between CO2 measured from the blood and from expired air) in patients with traumatic brain injury. I thoroughly enjoyed presenting our preliminary findings at the East Anglian Association of Anaesthetists annual meeting last year.
Artificial intelligence is a rapidly expanding field, which is increasingly being integrated into medical education and practice. As the healthcare landscape evolves, AI is recognised as a powerful tool to revolutionise patient care and medical research. I am fortunate that Emmanuel College, Cambridge has funded my research and allowed me to contribute to this area.
As I progress in my medical studies, I am grateful for the enriching experiences and opportunities that have shaped my academic journey and I am determined to contribute to the transformation of healthcare through innovative medical technologies. I am always happy to be contacted for potential collaborations, or to give advice on applying to medical school! Feel free to reach out to me via email or LinkedIn.
@neil-sardesai

John Strickland (1986)
AFTER ATTENDING THE PERSE all the way from Porson Road up to Sixth Form, I went to business school following a brief attempt at architecture. This led me to my first job, in Brussels, looking at electronic data interchange – how organisations share and exchange data. It was the year 1985, and the early days of the internet, when I helped that company set up a UK subsidiary in Cambridge Science Park. I have now worked in the data industry for over 35 years, more recently using AI as a tool to analyse exponentially growing sets of government and retail data.
Back in the 1980s, this internet thing looked as though it was going to be something, so I decided to set up an internet design-and-build-agency called Interesource. We built the first website for the London Stock Exchange before London even had access to the internet! The 14.4 kbit/s dial up modem turned into a 28.8-modem and, before you knew it, the 56-modem arrived which meant it only took 26 hours to download a film from the internet. The explosion of data had begun. I ran and grew that business for 12 years, including a spin off in Australia, before selling bits off. After a couple of philanthropic projects, I started looking for the next new thing.
The invention of the internet of things, in which machines began to talk to other machines, marked an exciting new era within data. I joined my present company, which takes data from multiple sources and visualises this with interactive dashboards. The data is then used to create budgeting plans and forecasts, applying machine learning and predictive analytics. This setting suited me – a private company with ambitious expansion plans that embraced my entrepreneurial spirit. I started in the retail vertical to help global giants, such as M&S, Burberry, Puma, Gucci and Decathlon, to drive digital transformation programmes. I then moved to the United Arab Emirates and am currently based in Dubai, directing transformation for prominent retailers and Saudi government entities.
It is difficult for us to understand the vast volumes of data humans produce and will exponentially continue to produce. Currently, the world’s data is doubling every two years and is predicted to double every 15 hours by 2024, according to Forbes. That’s a lot of data to be analysed and this is where AI steps in to help make accurate predictions. The commercialisation of generative AI has the potential to be a significant tool in data analysis in the near future, although we are not there yet.
All of these technological advances would have seemed like pure sci-fi when I was a student at The Perse. Thinking back on my school days, I remember walking over the parade ground to the old sheds where there were a couple of BBC punch card computers. I also remember the three-minute Shakespearian speeches in the junior mummery, buying Fitzbillies Chelsea buns at breaktime and attending the Second to Fourth Form discos. The Perse also introduced me to international travel, which I still love to do and hopefully passed on to my two girls. Commander Sumnall at the Prep once told me, in only the way he could, that you could always recognise an Old Persean as they are always performing. What sage advice.

Thatcher Ference (2020)
I COMPLETED MY A levels at The Perse after studying at Harrow Beijing and Phillips Academy Andover in the US. I am currently completing my first university degree in machine learning and artificial intelligence, and I am also a co-founder of a start-up company applying AI in medicine.
The apparent rapid advancements in generative AI have certainly captured the public’s imagination. Ironically, however, these advancements were not due to a technological breakthrough in the underlying algorithms, which have largely remained unchanged for several years. Instead, it was the intuitive leap that humans must guide the AI to “generate” something useful. In this case, the technique of reinforcement learning with human feedback created a game that was scored on how well humans thought the generative AI sounded like another human. The algorithm then learned the rules of this game and how to converse in natural human language.
This revolutionised the user experience because the algorithm could generate responses as if the user were communicating with another human. However, the ability to generate original responses to questions is both generative AI’s greatest strength and its biggest weakness. Sometimes, the algorithms generate inaccurate or even false information. This is euphemistically referred to as “hallucinating”. Although this could be a strength in generating new stories, song lyrics or other creative ideas, it can be quite dangerous when the false information can cause harm – particularly in science and medicine.
My research focuses on encoding biological cause and effect into AI algorithms. Encoding biological cause and effect creates algorithms that reflect the biology of how common diseases develop. As a result, these algorithms can both predict outcomes and, for the first time, prescribe specific actions to change those outcomes. These algorithms are designed to explain why a person is at risk, how to reduce that risk and how much they will benefit from specific interventions to reduce risk.
These algorithms are also trustworthy because they can be empirically tested against the observed biology using randomised evidence. Embedding the algorithms in a generative AI “wrapper” creates a human-like chatbot that can guide people on how to personalise the prevention of common human diseases. The user experience is transformed, since it is as if they were able to communicate with a human to guide them along the way.
This research is now the focus of a biotechnology start-up company that aims to show how generative AI can be used in a safe and trustworthy way in medicine. Our goal is to create a humanlike bot which will be the world’s greatest expert in how common diseases develop and how to prevent them. We intend to make this bot freely available around the world to help empower people to extend their healthy lifespan. This could be particularly important in the developing world that lacks an adequately trained healthcare workforce.
The quick progression in AI technology has happened in just the last three years since I left The Perse. New tools are being developed every day. For that reason, I would strongly encourage current Perse pupils to join the AI revolution and help shape its future. Even if you do not encode the algorithms, these tools can help empower you to solve problems in any field you choose to go into. Indeed, these tools will empower you to help guide the AI to accomplish remarkable things that are only limited by your imagination. In this way, you can live out the Perse ideal: “To love learning and strive for the greater good.”
INTERVIEW with leader in AI technology David Hogan

David Hogan is a Perse parent and Vice-President for Enterprise in the EMEA (Europe, Middle East and Asia) region at NVIDIA.
What do you do?
I work for one of the world’s leading AI companies, NVIDIA, enabling customers across the EMEA region to build and scale AI projects. For example, we work with healthcare experts to look at how AI can move patients to the right point of care more quickly, given factors like previous medical history. NVIDIA has been behind many of the major advancements in AI in the past decade and is the platform which powers the generative AI revolution.
AI plays a role in all the different industries you can think of, including automotive, retail, logistics, financial services, entertainment and what we call Smart Spaces, where AI is used to create safer cities and buildings. It is a broad field and interesting to see varying degrees of adopting AI across the world.
What is the current state of AI and what recent advancements are most promising?
This is the moment when AI has become understandable and real. Before ChatGPT was launched in December 2022, the only real experience the average person had of AI was voice recognition from technologies like Alexa or Siri.
Generative AI models, such as ChatGPT, are large language models (LLMs) trained on what is written on the internet. Based on the learned information, it can determine relationships and therefore understand patterns which it has never been taught directly.
This is why LLMs are transformative compared to “traditional” AI models, which require large amounts of training to recognise patterns and progress sophistication. Since LLMs are already available and you do not have to build them yourself, anyone can use or enhance the models by training them on specific data for a particular domain.
ChatGPT has been called the iPhone moment for AI. When the iPhone was launched there were maybe 100 applications, and now there are millions. Similarly, AI will underpin all major industries and any application or software you use in the future will be AI-enabled. In the next couple of years, the user of any software will get a more personalised, intelligent and, hopefully, more helpful experience. However, you may not necessarily recognise that AI is behind it.
As AI evolves, how can the challenges of AI be addressed effectively?
There are already good regulatory frameworks that apply in many areas of life, the majority of which can be applied to an AI when it replaces a human activity. If the human needs a licence to drive a car, so should the AI. Like all technology, AI can be misused and regulations are important. However, they shouldn’t stifle investment and restrict innovation. Regions like Asia and North America are already ahead in Al development compared to Europe because the attitude towards AI here is “we must control it”.
It is key to think of AI as a tool. Right now, AI is nowhere near the level of complexity and capability of a human and it is unlikely to ever reach that point. However, AI is excellent at doing certain tasks humans struggle to do with a high level of efficiency and accuracy. For example, computers are much better at mining through lots of data or repeating the same task many times.
Improving accuracy and reducing bias in AI models needs to be continuously worked on. NVIDIA is working on what we call “guard rails”, where a second AI model checks that the foundational model doesn’t exceed what it should do i.e., it doesn’t give inappropriate answers and therefore creates a level of safety. It is not perfect, but still good at filtering out unwanted outputs.
What impact do you think AI will have on education?
Critical thinking, emotional intelligence and creativity are the skills of the future and schools need to pivot away from fact-based learning. Here are some ways LLMs can be powerful tools in education:
Personalised learning: By analysing individual strengths and weaknesses, an LLM can generate customised lesson plans and resources that cater to each pupil’s needs.
Interactive tutoring: LLMs can serve as interactive virtual tutors, providing immediate feedback and explanations to pupils’ questions, helping to reinforce concepts and encourage active learning.
Language learning: LLMs can be used to develop language learning platforms that provide real-time translation, grammar correction and vocabulary suggestions.
Content creation: LLMs can generate educational content such as quizzes, exercises and study materials. This can save time for educators and ensure the content aligns with the curriculum and learning objectives.
Research assistance: Pupils can use LLMs to find relevant and reliable information for their projects and research papers. The AI can assist in summarising articles, citing sources and fact-checking.
Critical thinking and problem solving: LLMs can be used to develop scenarios and case studies that challenge pupils to think critically and solve complex problems, fostering higher order thinking skills.
Gamification: Integrating LLMs into educational games can help to motivate pupils to participate actively in the learning process.
Accessibility: LLMs can be adapted to assist pupils with disabilities, providing personalised support and accommodations to facilitate their learning journey.
It is crucial to consider ethical and privacy concerns when implementing AI technologies in education. Ensuring data protection, minimising biases and prioritising student wellbeing must be paramount. It is essential to stay updated on the latest advancements and research to make informed decisions about incorporating LLMs into pupils’ education effectively and responsibly.