The New World Order

No matter where you turn these days, there’s an issue or impact relating to the unprecedented infiltration AI is having on our everyday lives and work. Some is beneficial, some intrusive, and some is downright dangerous. AI is known to hallucinate and confabulate, making up stuff to fill in gaps in its desire to satisfy our requests.

Some parties are now ‘polluting’ information that they deem has been poached from their legitimate content. As a result, search engines are suggesting you add glue to a pizza to stop the cheese sliding off, or suggest mixing hazardous combinations of chemicals to clean your washing machine, resulting in dangerous chlorine gas formation.

Although AI is not new (it first entered the lexicon in the mid 1950s), modern computing power and the internet has allowed it to flourish. There are claims that all recorded text has now been parsed by AI systems, but this probably only applies to that available digitally. There must be plenty of texts, magazines and newspapers that were never digitised.

AI is being hyped to stratospheric levels and is becoming pervasive.

AI AND EYES

Careful consideration must be given to the critical issues surrounding AI that face our professions. Medico-legal and other ethical ramifications and questions regularly come to the fore. There are numerous debates and interesting discussions on social media and in professional forums.

Such has been the rate of change that our professional bodies, registration, and regulatory authorities are struggling to keep up.

What’s allowed and what risks could we face?

We need to be very careful as to what we implement, but at the same time we don’t want to get left behind. There are always early adopters who pioneer the future standards of care. Others sit on the fence and await the ‘evidence base’ before taking on new treatments, procedures, or processes. We wouldn’t have advancement without pushing the boundaries: Which then leads to developing an evidence base. We do, however, need to be properly informed.

Is it legal and ethical to use an artificial intelligence scribe to take record-keeping notes during consultations? The onus is on the practitioner to review such transcripts and to correct and edit them before being saved into a permanent record. Human scribes are not new (an ophthalmologist colleague has used one for decades) but AI versions will soon be in widespread use. I recently tested the Australian developed i-Scribe and was impressed by its performance (see a brief summary in the GenEye story on page 56).

To comply with Australian privacy regulations, it’s important to make sure that any patient data is stored only on Australianbased servers. This applies to cloud-based storage, or backups, or patient data in any form, not just AI systems.

CHAT BOTS

There are now diagnostic chat bots that can help you make a differential diagnosis on a case by simply uploading a few relevant findings and maybe a retinal image or optical coherence tomography (OCT) scan. Again, we must consider if this is safe and legal (are the servers overseas?). A retinal image is like a fingerprint that in theory could identify a patient.

We must understand that such diagnostic bots are not foolproof and are (currently) for interest and assistance only. Presently, the buck stops with us.

We must also accept that AI can do things that are beyond our capabilities. A recent study, ‘Gender prediction from retinal fundus using deep learning’, by Taha et al.,¹ showed that machine learning can successfully determine gender with 99% accuracy, simply from a retinal image. I’ve yet to meet a human that can do that 1% of the time! What else may such techniques discover hidden in retinal and other imaging? The study also pointed out that convolutional neural networks (CNN) achieved physicianlevel accuracy in numerous image-based health jobs, such as radiology, dermatology, pathology, and ophthalmology.

We’re also informed that in some states, it’s legal for a patient to take recorded transcripts of notes (surreptitiously) during a consultation via an AI-enabled device. Some suggest that recordings of consults should only be performed with mutual consent. Either way, surreptitious recordings by AI devices are already happening. The result is a record of what was said during a consult.

Bots can also summarise relevant findings and make recommendations. Much like a scribe. I see a resultant dramatic rise in medico-legal claims (and wins) by patients, as it is no longer a ‘he said, she said’ scenario. An example might look like: “He never warned me of dysphotopsia with the multifocal intraocular lens that he recommended. My life is ruined, it’s driving me crazy! This is what was said during the consultation. Sue him!”

Similarly, a patient might say: “They never warned me that I could lose my sight from swimming in my contact lenses! Now my left eye has no functional vision and looks scarred and disfigured. My girlfriend dumped me. I’m so depressed and in so much pain. Sue them!”

In general, we’ll likely see more and more artificial intelligence supporting us and aiding diagnosis.

FEAR? OR NOT?

My view is that AI is a tool, not a be-all and end-all of our existence. Patients increasingly want the human touch. I don’t like having to go through six keypad menus when phoning a business. I want to talk to a human. I don’t like support and chat bots that are replacing humans, where you have to type out your story over and over, yet they still don’t understand your specific set of circumstances. A human could solve the problem in a fraction of the time.

As always there’s a time and place for everything.

Take care out there.

Alan Saks is a retired optometrist. He is the Chief Executive Officer of the Cornea and Contact Lens Society of Australia, and a regular contributor to mivision.

Reference

1. Taha AM, Zarandah QMM, Abu-Naser SS. Gender prediction from retinal fundus using deep learning. international journal of academic information systems research (IJAISR) 6 (5):57-63 (2022). Available at: philarchive.org/archive/TAHGPF.