Albeit the public perception of the integration of AI into our healthcare systems, medical treatments and devices has improved, trepidation surrounding how disruptive AI will be to our existing health practices and treatments remains. Generally, patients are much more trusting of a human clinician than AI and yearn for the traditional doctor-patient relationship. To add to this, there are concerns over how ethical AI is and how secure our health data truly is.

However, with rising health concerns and spending, being able to harness AI’s potential responsibly and ethically will inevitably spell less time with human clinicians but could also mean improved outcomes.

Potential for discrimination – Studies have shown discrimination already exists, albeit often unintended, within our health services and culture against minority groups. If AI enabled devices are trained with inherently biased algorithms, the devices born out of that will inevitably be biased too. Studies have found reduced incidence in skin cancer diagnosis in darker skinned patients, when compared with light skinned patients. This is thought to be due to the AI algorithm having been trained on detecting cancer in predominantly lighter skin. We must ensure that datasets are not underrepresented in order to strive for and achieve equity in all medical device development. To do this we must adopt an equitable approach immediately and incorporate this in algorithm development and machine learning.

The proficiency of the data collected and used to train the AI algorithms must also be called into question. How can we be sure that the data used in these instances is of sufficiently high quality? The system needs to be transparent enough to ensure service users are reassured of the objectivity of the data used.

Data ethics – In terms of informed consent, the premise is for the individual giving consent to have a full understanding of what they are actually consenting to. In terms of AI enabled devices, it may be difficult for a ‘layperson’, so to speak, to have a full enough understanding to make an informed decision on their care. Issues surrounding data privacy have also been raised as a potential issue. Many suggest that AI enabled devices may contain patient identifiable data which will become vulnerable to data breaches, potential hacking or attacks on the AI network.

In terms of data ownership, the individual which the health data relates to ultimately owns the data. However, healthcare providers and companies own the right to access the data, and so often is the case, it will then belong to the highest bidder, to use it for advancement in their objectives. It cannot be guaranteed that this will always be in the best interests of the data owner, in other words, the patient.

Lacking in expertise – AI in medicine is a case of the virtual colliding with the physical. There is likely to be a lack of expertise which will contribute to disadvantages. For example, software engineers who are qualified to train algorithms are unlikely to be medically trained. Likewise, clinicians are not likely to be trained in software engineering. This may well lead to some technical illiteracy. This may in turn reduce efficiency and increase errors.

It is also suggested that original algorithms will eventually deviate due to the evolvement and variances in disease progression and patterns. It’s vital that we do not become too reliant on AI and ensure that it is regulated and monitored effectively to ensure it remains efficient. Currently, it seems there is a lack of standardised processes and regulation to protect against the pitfalls of AI.

Human factors – As mentioned, the algorithms guiding the AI devices will be trained and implemented by software engineers. There will always be an element of human error at play which could be unwittingly trained into the device. As well as this, human error on the clinician’s part, how proficient they are in using the AI enable device, could also have detrimental effects.

Ethics of clinical practice – rule orientated robots may well be more reliable but when faced with ‘human’ situations, requiring empathy, humans are more ethically reliable. Patients would much rather receive news about poor prognosis from a human clinician than from an AI platform.

Social acceptance – Albeit there has been increase in public exposure to AI, which has allowed its shortcoming to be discussed and addressed, most patients report to remain more trusting and accepting of being delivered a diagnosis by a human rather than AI. Concerns from the clinical sector include fears that they will be replaced by AI technology, but others fear clinicians will become too reliant on AI, and in effect, clinicians will begin to become de-skilled. Experts have warned clinicians must remain sceptical about AI outputs, they must not blindly accept AI results and instead continue to rely on their clinical judgment and experience to ultimately guide them.