Blog

Why Bring Your Own AI (BYOAI)?

Read Time: 3 Minutes

Recently Trisotech posted a position paper entitled “Bring Your Own AI (BYOAI)“. Its observations are relevant to modeling and automation in healthcare. Personally, I think this is the way to make use of AI – at least for now.

Chasing the Next Shiny Thing

AI models are evolving right before your eyes, with many companies appearing then disappearing on a monthly basis. The environment is chaotic, and it is hard to know what to do, and it is tempting to try the newest platforms. This can paralyze project development as you struggle with deciding what tool to use, afraid that it will be obsolete next week. Often you look to see what others are doing rather than forging your own strategy.

Constantly changing environments are plagued by problems with version control and testing. Users become confused and may push back, especially after they hear about bias in algorithms or incorrect responses. They may feel forced to use something that they have no confidence in.

Overreaching

The natural tendency when using a powerful tool is to apply to a large problem all at once. For many projects this can result in subpar performance. Anyone using the current AI platforms can see a deterioration in performance once a certain task size is reached. Often you have no idea exactly what changed or why. Has a seemingly minor change in a prompt caused a vast change in output? Could you have run out of tokens? Are you competing against other users for access? When faced with something outside of your control you can hold a rabbit’s foot and pray for the best.

Who Is Going to Be Liable?

At some point something is going to go wrong – it always does. So just who is going to be left holding the bag? The lawyers are already lining up.

Do you think the AI companies are going to step up? The fine print of their user agreement tries as hard as possible to deflect responsibility. Traditionally clinicians have been liable and so carry liability insurance. What person is going to accept responsibility for a black box that they have no control over, do not understand, and offers no evidence for its actions. How can using a novel AI platform be considered standard of care? Will malpractice insurance cover this? You can be sure that your already expensive cyber insurance is going to cost a whole lot more.

So Why Does BYOAI Work in Healthcare Modeling?

Here are some reasons why I like the idea of BYOAI for modeling in healthcare.

1

You are not tied to a single AI platform.

Some platforms perform better in some tasks than others. A specific task can call whatever service works better. If a better one comes along then you can swap it in without changing the rest of your model.

2

You can control when and how the platform gets called.

You can limit its scope so that it returns focused responses. This makes the system easier to test.

3

You can make use of attended tasks.

These allow physicians to review and modify the model as it progresses in light of the patient’s clinical context. They can accept, reject or modify any suggestion based on the current situation and so can act responsibly in caring for a specific individual. Here AI functions as a useful ally with the clinician in control.

BPM+ Modeling with BYOAI gives you the best of both worlds. No one knows where AI will be going in the future, but modeling should be able to evolve along with the technology. You can confidently get started solving your problems now, rather than waiting for a future that is unpredictable.