AI: The Doctor That Never Tires?

We asked you for your opinion on the use of Artificial Intelligence (AI) in healthcare, and got the above-depicted, below-described set of results:

• A little over half of respondents to the poll voted for “It speeds up research, and is more methodical about diagnosis, so it’s at least a good extra tool”
• A quarter of respondents voted for “I’m on the fence—it seems to make no more nor less mistakes than human doctors do”
• A little under a fifth of respondents voted for “AI is less prone to fatigue/bias than human doctors, making it an essential new tech”
• Three respondents voted for “AI is a step too far in medical technology, and we’re not ready for it”

Writer’s note: I’m a professional writer (you’d never have guessed, right?) and, apparently, I really did write “no more nor less mistakes”, despite the correct grammar being “no more nor fewer mistakes”. Now, I know this, and in fact, people getting less/fewer wrong is a pet hate of mine. Nevertheless, I erred.

Yet, now that I’m writing this out in my usual software, and not directly into the poll-generation software, my (AI!) grammar/style-checker is highlighting the error for me.

Now, an AI could not do my job. ChatGPT would try, and fail miserably. But can technology help me do mine better? Absolutely!

And still, I dismiss a lot of the AI’s suggestions, because I know my field and can make informed choices. I don’t follow it blindly, and I think that’s key.

AI is less prone to fatigue/bias than human doctors, making it an essential new tech: True or False?

True—with one caveat.

First, a quick anecdote from a subscriber who selected this option in the poll:

❝As long as it receives the same data inputs as my doctor (ie my entire medical history), I can see it providing a much more personalised service than my human doctor who is always forgetting what I have told him. I’m also concerned that my doctor may be depressed – not an ailment that ought to affect AI! I recently asked my newly qualified doctor goddaughter whether she would prefer to be treated by a human or AI doctor. No contest, she said – she’d go with AI. Her argument was that human doctors leap to conclusions, rather than properly weighing all the evidence – meaning AI, as long as it receives the same inputs, will be much more reliable❞

Now, an anecdote is not data, so what does the science say?

Well… It says the same:

❝Of 6695 responding physicians in active practice, 6586 provided information on the areas of interest: 3574 (54.3%) reported symptoms of burnout, 2163 (32.8%) reported excessive fatigue, and 427 (6.5%) reported recent suicidal ideation, with 255 of 6563 (3.9%) reporting a poor or failing patient safety grade in their primary work area and 691 of 6586 (10.5%) reporting a major medical error in the prior 3 months. Physicians reporting errors were more likely to have symptoms of burnout (77.6% vs 51.5%; P<.001), fatigue (46.6% vs 31.2%; P<.001), and recent suicidal ideation (12.7% vs 5.8%; P<.001).❞

See the damning report for yourself: Physician Burnout, Well-being, and Work Unit Safety Grades in Relationship to Reported Medical Errors

AI, of course, does not suffer from burnout, fatigue, or suicidal ideation.

So, what was the caveat?

The caveat is about bias. Humans are biased, and that goes for medical practitioners just the same. AI’s machine learning is based on source data, and the source data comes from humans, who are biased.

See: Bias and Discrimination in AI: A Cross-Disciplinary Perspective

So, AI can perpetuate human biases and doesn’t have a special extra strength in this regard.

The lack of burnout, fatigue, and suicidal ideation, however, make a big difference.

AI speeds up research, and is more methodical about diagnosis: True or False?

True! AI is getting more and more efficient at this, and as has been pointed out, doesn’t make errors due to fatigue, and often comes to accurate conclusions near-instantaneously. To give just one example:

❝Deep learning algorithms achieved better diagnostic performance than a panel of 11 pathologists participating in a simulation exercise designed to mimic routine pathology workflow; algorithm performance was comparable with an expert pathologist interpreting whole-slide images without time constraints. The area under the curve was 0.994 (best algorithm) vs 0.884 (best pathologist).❞

Read: Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer

About that “getting more and more efficient at this”; it’s in the nature of machine learning that every new piece of data improves the neural net being used. So long as it is getting fed new data, which it can process at rate far exceeding humans’ abilities, it will always be constantly improving.

AI makes no more nor less fewer mistakes than humans do: True or False?

False! AI makes fewer, now. This study is from 2021, and it’s only improved since then:

❝Professionals only came to the same conclusions [as each other] approximately 75 per cent of the time. More importantly, machine learning produced fewer decision-making errors than did all the professionals❞

See: AI can make better clinical decisions than humans: study

All that said, we’re not quite at Star Trek levels of “AI can do a human’s job entirely” just yet:

BMJ | Artificial intelligence versus clinicians: pros and cons

To summarize: medical AI is a powerful tool that:

• Makes healthcare more accessible
• Speeds up diagnosis
• Reduces human error

…and yet, for now at least, still requires human oversights, checks and balances.

Essentially: it’s not really about humans vs machines at all. It’s about humans and machines giving each other information, and catching any mistakes made by the other. That way, humans can make more informed decisions, and still keep a “hand on the wheel”.