Artificial Intelligence and Healthcare

Neal Bangerter

Associate Professor of Bioengineering at Imperial College London

Artificial Intelligence (AI) is the ability of a machine to think or perform cognitive functions we associate with humans. The confluence of developments in three broad areas drives its recent growth: advances in algorithms (sets of rules which are followed to solve a problem); an explosion of data; and an exponential increase in computing and storage power.

As far as algorithmic advances are concerned, Neal emphasised the seminal work of Andrew Ng and a team at Stanford University in 2009 which drastically reduced the training time for deep-belief networks (one form of algorithm) from several weeks to about a day. He further emphasised work by computer scientist Geoffrey Hinton (University of Toronto and Google Brain) on image classification using Convolutional Neural Networks (CNNs—another form of algorithm), and recent work by Google DeepMind’s AlphaZero team. AlphaZero learnt three complex games – Go, chess, and shogi – by playing itself, without any human instruction. In 2017, AlphaZero beat Chinese Master Ke Jie at Go, a centuries-old game considered humankind’s most complicated board game.

The explosion of data has been due to the widespread adoption of broadband and smartphones. Electronic device users now generate more than 2.5 quintillion bytes of data per day; and 90% of the world’s data were produced in the past two years. Every minute, YouTube users watch more than four million videos and mobile users send more than 15 million texts. Neal emphasised that the availability of such vast quantities of digital data are necessary for the training of artificial intelligence systems to perform tasks traditionally performed by humans.

Drastic reductions in the price of storing digital data have been central in enabling AI. By 2005, the cost of storage had dropped to $0.79/GB from $277/GB ten years earlier. Furthermore, Neal pointed out that the introduction of powerful Graphics Processing Units (GPUs) by companies such as Nvidia in the early 2000s provided much-needed computing power for training machine learning models. Further advances in processing power came in 2017 when Google introduced its Tensor Processing Unit (TPU): it runs its own machine learning models 15 to 30 times faster than current GPUs and CPUs (Central Processing Units). Google intends to make its TPU processing power available to the public via the cloud, which will further fuel development of AI.

Neal emphasised the importance of data itself, rather than computer models, in the development of AI. He referred to the work of Dr. Fei-Fei Li, Chief Scientist at Google Cloud and founder of the ImageNet visual object recognition repository. Dr. Li’s wildly successful ImageNet Challenge, where AI researchers submit visual object recognition algorithms to compete on accuracy and speed, is widely seen as a catalyst for the AI proliferation the world is seeing today. Dr. Li claims that “while a lot of people are paying attention to models, more attention needs to be paid to data. Data will redefine how we think about models.”

Although Neal thinks we are still “a long way from artificial general intelligence, where a machine could successfully perform any intellectual task that a human being can,” AI does have a significant role to play in certain areas – notably healthcare – over the coming decades. Neal set out a likely timeline. Over the next five years, data standardisation efforts (data acquisition, recording and cleaning) will be the main focus. This will form the basis for the proliferation, over the next ten years, of specialised deep-learning tools in healthcare. It will become commonplace for physicians to augment their own capabilities with AI-based tools. The regulatory environment will begin to develop to support AI-based healthcare decisions. Over the next 20 years, AI and deep learning will make healthcare more accessible and more accurate, helping contain costs in the developing world.

Throughout this phase of development, Neal pointed out that wearables and very active involvement by individuals in monitoring and addressing their health will become commonplace in the developed world, informed by AI-based tools using standardised data collection and near-constant monitoring.

In short, in the words of Curtis Langlotz of Stanford Radiology, “AI will not replace physicians. Yet, medical professionals who use AI will replace those who don’t.”


Phone: +44 20 7491 9111

Fax: +44 20 7872 3706

Issued by EFG Asset Management (UK) Limited.
Registered in England and Wales 7389736

Registered Office:
Leconfield House
Curzon Street
London, W1J 5JB

Authorised and regulated by the Financial Conduct Authority