Portal de Periódicos da CAPES

Artificial intelligence in health care

2020; Wiley; Volume: 56; Issue: 10 Linguagem: Inglês

10.1111/jpc.14828

1440-1754

David Isaacs,

Healthcare professionals’ stress and burnout

I have a confession to make: everyone is talking about artificial intelligence (AI), but I had no idea what it actually meant. So I set out to find out. My research tells me AI and machine learning or machine intelligence are terms used in computer science for 'intelligence' demonstrated by machines like computers, as opposed to the 'natural intelligence' of humans and some other animals and birds (Fig. 1). The concept of intelligent machines was considered and described by science fiction writers like Isaac Asimov before they became real. Mary Shelley, daughter of feminist Mary Wollstonecraft and wife of Percy Bysshe Shelley, started writing Frankenstein, the story of an artificial being, in 1815 when she was only 18. Douglas Adams invented an Electric Monk which believed things for you in order to save you the bother of believing everything the world expected you to believe.1 The word intelligent comes from the Latin intelligere, meaning to understand. Is a computer that beats a chess grandmaster actually intelligent or just well programmed? Alan Turing, the brilliant English mathematician who worked at Bletchley Park during World War II breaking German ciphers, whose work led researchers to consider the possibility of building an electronic brain, suggested a better question was not whether a machine was intelligent, but whether machines could exhibit intelligent behaviour. Incidentally, despite being a war hero, Turing was prosecuted for 'gross indecency' for the then crime of being homosexual and was 'chemically castrated'. In 1954, at age 41, Turing died after taking cyanide. In 2009, the British Parliament pardoned Turing and apologised to him for having persecuted him. In 1956, a young mathematician John McCarthy organised a workshop on thinking machines at Dartmouth College, New Hampshire. He proposed a 2-month, 10-man study of 'artificial intelligence', the first ever use of the term. People came and went at the workshop. Rather wonderfully, McCarthy lost his list of attendees. Enthusiasm for AI waxed and waned. The late 1970s when funding was withdrawn were called the 'AI winter', but that winter of discontent has matured into a burgeoning spring-time of AI use, including in multiple areas of health care. AI is more than just computerising tasks; it builds on and develops computing capabilities. AI is about how machines 'learn' the associations between things by being repeatedly exposed to data. They are agnostic as to whether or not those associations make a priori sense or are logical. AI is increasingly being used in diagnosis, treatment and research. In dermatology, AI systems get feedback data from scanned skin biopsies and gradually improve their diagnostic accuracy based on the data, although systems are not yet sufficiently reliable to be used as the exclusive determinant of which skin lesions need biopsy. Computer technology is being used in pharmacies to assist with drug dosing and even with choice of drugs. One use in research is to aid in sample-size determination. On a positive note, if reliable, AI applications have enormous potential to improve information use in resource poor settings. A literature search for AI yields lots of articles on robots. Although there is no universally agreed definition of AI, most experts agree that a robot that simply automates a task is not AI, because AI involves significant learning. Current robotic surgery has a master–slave basis; true artificially intelligent robotic surgery is still a way off. Surgeons are rapidly training themselves in using robot-assisted (minimally invasive) surgery to operate on adults with proven or suspected malignancy, gynaecological problems including tubal sterilisation, and prostatic hypertrophy. There is less use of robots in children, although there are reports of paediatric robot-assisted surgery being used to perform pyeloplasty in the USA2 and to treat solid tumours in the UK,3 albeit with little or no evaluation as yet of the advantages and disadvantages. The names are great. The main US robotic surgical system is called the da Vinci Surgical System, a fitting tribute to the 16th century polymath, vegetarian, liberator of caged birds, so beloved of paediatricians (Fig. 2). A robot soothingly named Calmer is better at reducing pain in neonates than 'facilitated tucking'.4 Robots are increasingly being used for the rehabilitation and education of children with disability, although research seems to be lagging behind practice. Robots have been used in locomotor training of children with cerebral palsy, spinal cord injury or after stroke.5 One might suspect a negative connection between robots and emotion, but this may be a mistake. In a hospital inpatient study, 54 children aged 3–10 years were randomised to be given an interactive, tele-operated social robot teddy bear, a tablet-based 'avatar' version of the bear or a plush teddy bear to cuddle.6 Every child interacted with a child life specialist, who physically operated the social robot or the avatar and gave the plush bear to the child. Whether or not one considers this a fair comparison, children with social robots or avatars were happier, as assessed by measures of 'joyfulness' and 'agreeableness', than children given a plush bear.5 Eat your heart out, Winnie-the-Pooh. AI raises many ethical implications. Will AI systems displace health-care professionals? How will AI systems avoid discrimination against minority groups? For example, facial recognition for job interviews may reinforce existing inequalities by discriminating against racial minorities or women. Stephen Hawking expressed concern that powerful AI systems could make dangerous mistakes. In 2014, billionaire entrepreneur Elon Musk referred to AI as humanity's 'biggest existential threat'. In 2016, Musk launched a neurotechnology company, Neuralink, aiming to develop implantable brain-machine interfaces linking human brains to computers. The short-term aim is to treat serious brain diseases but the long-term aim is 'human enhancement' to enhance mental capabilities.7 As we develop AI systems we need to be mindful that they may not behave in ways we consider ethical. We should try to anticipate those ethical considerations before, not after, we introduce AI. AI and machine intelligence seem to mean a whole host of things. They provide novel opportunities to improve health care, but also serious potential ethical challenges. And, with luck, a robot will write future editorials for this Journal and I can retire gracefully. Thank you to Dr Chris Elliot, Dr Anna Isaacs, Mr Tom Isaacs, Associate Professor Henry Kilham, Professor Jodie McVernon, Dr Ken Nunn, Professor Mike South and Dr Bernadette Tobin for helpful advice on this paper.