Prime Minister Boris Johnson has embarked on a spending spree as he sets out to establish his credentials as the new leader of the United Kingdom. On August 8, 2019, he pledged a £250 million investment in artificial intelligence (AI) in the NHS. This money is destined for a new National Artificial Intelligence Laboratory, hosted within NHSX (the national agency for digital transformation in healthcare)- a new organisation set up to oversee the digitisation of the NHS.
Large IT projects in the NHS have had a poor historical record of implementation and delivery, besides being heavily overspent. AI, however, has potential promise, despite the hype that inevitably precedes any IT innovations in the health sector. There was, in fact, a 20-year period starting in the mid-1980s, when the medical AI movement came to a standstill. The so-called “AI winter” followed an initial period of high optimism in the 1970s. The messages of history should never be forgotten.
AI does not set out to replace doctors, but rather complements their work in an age of ever-increasing human-machine collaboration. Computers are after all unable to display compassion and empathy, and cannot read important non-verbal cues provided by patients. Clinical judgement and “lateral thinking” in individual cases can only be employed by humans at present.
AI can help improve diagnostic accuracy. This is important because of the continued high burden of errors in diagnosis of illness and injury. AI systems can actively learn, picking out patterns and correlations among the large amounts of data constantly being fed into the system. This may enable earlier and more reliable recognition of diseases, especially rarer ones, as the AI system matches the patient’s symptoms to already identified patterns of disease.
AI attempts to simulate or replace human learning, reasoning, problem-solving and the ability to correct oneself, all through the use of computers. This can be helpful, given that humans demonstrate many biases of thinking and reasoning- the so-called cognitive biases-especially when tired, under emotional stress, distracted or otherwise cognitively overloaded. The Turing Test foresaw AI by postulating a computer that was able to provide responses to a human interrogator that could not be distinguished from those made by another human.
Machine learning (ML) is used with structured data, such as scan images or genetic data, and natural language processing (NLP) is used with unstructured data, such as medical records and laboratory reports. AI systems can combine ML and NLP modules.
AI systems continue to learn on the job from unstructured information using natural language processing, which involves the conversion of free text into a structured and standardised format. This can in due course help diagnose or rule out various diseases by the automatic detection of patterns in the newly structured data. AI allows for learning from “big data,” such as that available from large hospital data sets of patients, by a process of data mining, or the extraction of useful information from raw data. Pattern recognition allows development of software applications to perform tasks without the need for explicit programming.
Doctors may distrust AI because the underlying logic which leads to a specific decision cannot be readily understood. The AI system processes information through a complex system of algorithms which are hidden from the user. This resulting lack of transparency of decision-making is sometimes referred to as the “black box problem”. There are also wider issues related to data security and with ensuring that machines can achieve equitable decision making. Ultimately, the question is who is to decide whether the AI system has made the correct decision?
The potential applications of AI to healthcare are wide-ranging and include the provision of virtual assistants to provide information and advice in conjunction with healthcare apps, and also in precision medicine, AI-assisted robotic surgery, and AI-enabled workflow and administrative tasks.
Several examples are emerging of the use of AI for diagnosis of disease. Moorfields Eye Hospital in London has partnered with DeepMind Health and University College London to develop a system to diagnose diseases of the retina in the eye . This is based on computerised analysis of images obtained by optical coherence tomography scans. The automated diagnosis of various types of scan images may become a routine procedure, complete with oversight from specialist radiologists.
AI is here to stay and to establish its role in healthcare. Given the private sector’s dominance in the area, it is inevitable that this will remain an area for public-private sector collaboration for the foreseeable future. The big players, such as Microsoft, Google, IBM and Amazon, have already entered into a Partnership on Artificial Intelligence to Benefit People and Society. The NHS has started to form partnerships with a number of AI start-up companies. The future for collaborative AI ventures in healthcare looks promising for the present. The issues of data privacy and of a fair pricing structure for commercial access to NHS data remain to be decided.
Ashis Banerjee (ex-NHS)