AI for disease diagnosis...hurry, please!

In 1970 in The New England Journal of Medicine, William Schwartz predicted that by the year 2000, much of the intellectual function of medicine could be either taken over or at least substantially augmented by “expert systems”—a branch of artificial intelligence (AI). Schwartz hoped that the medical school curriculum would be “redirected toward the social and psychologic aspects of health care” and that medical schools would attract applicants interested in “behavioral and social sciences and … the information sciences and their application to medicine.” But Schwartz's dream of smart medical technologies, for the most part, remains just that.

Eric Topol, however, is optimistic about the future of health care. In Deep Medicine, he anticipates that new machine learning technologies will improve the precision and accuracy of disease diagnosis, thus providing a better way to identify the best therapies. Like Schwartz, he hopes that the time freed up by these approaches will be devoted to reviving humane medical practices.

“Artificial intelligence” initially took a very symbolic approach to replicating human reasoning. Problems were described as logic puzzles, and inference was based on rules of symbolic logic. Problems were solved by executing a sequence of steps, and natural language was addressed by building dictionaries and grammars around words and parts of speech. Even images, often reduced to line drawings, were represented by discrete elements such as vertices and line segments.

Around 1980, a small group of computer scientists proposed an alternative: What if a large number of simple computational elements modeled (very roughly) on biological neurons were to be used instead? As Geoffrey Hinton, one of those pioneers, explained in a talk at the Massachusetts Institute of Technology in 2014, we were about six orders of magnitude short of the computational power required for such a feat in 1980. Only within the past decade have we developed the hardware and algorithms needed to achieve high-performance “neural networks.”

Topol describes the basics of so-called “deep” neural networks—“algorithms that permit software to train itself to perform tasks by processing multilayered networks of data”—by summarizing the kinds of problems for which these methods have been remarkably successful and reciting the litany of concerns arising from inscrutable decisions made by such networks (“baked in” biases, privacy issues, and the susceptibility of computer models to seemingly imperceptible changes to input data). “[They] still don't know exactly what features account for its success,” he writes about a Stanford computer program that matches the diagnostic success rate of dermatologists.

Most AI successes so far in health care have come from the application of image-interpretation methods in domains such as radiology, pathology, dermatology, and ophthalmology. Many of these strategies are restricted, for now, to the research literature, although the U.S. Food and Drug Administration has recently approved a handful of such systems.

Techniques for assisting what Topol calls “clinicians without patterns”—medical professionals who make assessments and formulate plans by integrating heterogeneous data from a patient's records, medical literature, and talking with patients and their families—are at even earlier stages of development. These include digesting the medical literature in general internal medicine, diagnosing atrial fibrillation in cardiology, identifying the best available treatment in oncology, introducing precision robotics in surgery, and interpreting subtle cues from online communications in mental health (to which he devotes an entire chapter). Later chapters examine how AI could enhance the operations of the overall health system, aid in basic scientific discovery, and help bring nutrition and diet into consideration.

Last, Topol turns to his vision of how AI can provide a virtual medical assistant to clinicians and how these technologies can lead to the resurgence of the empathy-based care that Topol—and many others—miss in current health care. “AI can help achieve the gift of time with patients,” and that extra time can develop empathy, which “is not something machines can truly simulate.”

The great contribution of this book is that Topol synthesizes the fragmentary views that we who work in this field gain from day-to-day reading into a cohesive vision of a future in which medical care is about human care. Alas, achieving that depends on much more than improved technological support for clinical medicine. Hopefully, the economic and administrative forces that have done much to frustrate other recent visionaries will not derail this new plan.

I hate to ruin the parade but I personally am scared to death of AI. Once AGI (Artificial General Intelligence) is achieved - and IBM predicts this by 2029, although many are of course skeptical as rightly they should be - then the jump to ASI (Artificial Super Intelligence) is pretty much inevitable and likely will be quick. This is not my professional opinion as I'm not a scientist. But it is that of many who are.

I recommend Nick Bostrom's book "Superintelligence: Paths, Dangers, Strategies". It's a sobering read by one of the best minds at work on this stuff. For a more general layperson's look, James Barrat's "Our Final Invention: Artificial Intelligence and the End of the Human Era" isn't bad. And there are others.

While AI will (and does now) bring many benefits to humanity, I have a deeply pessimistic view of our ability to safely control it - since we haven't been very good at controlling much else we've created (see: weapons, nuclear and otherwise; pesticide use; plastics; the internal combustion engine; autocratic governments, etc., etc.)

Just my few cents.

one problem with AI, is that it learns from humans and some recent experiments have shown that AI tends to pick up human biases as part of its "learning" process.

Who gets to provide the AI with its 'truth'? Surgeons or ROs or just questionable stats and studies?

hopefully the PC DX AI won't start recommending mastectomies for PC...

Let me start with a bit of tedious moderator business. Rather than quoting in its entirety an article that you suspect might be copyrighted, it is better to quote an excerpt and include a link for the full piece. In this case that link would look something like this:

Deep learning could give doctors of the future more time for the human aspects of health care from Science Magazine's "Books Et Al" 26 March, 2019.

If you are not good with creating links (and they can be tricky) you can give a reference like a footnote in a high-school term paper. A search engine such as Google can find an online article from its title in a fraction of a second so there's usually not much practical difference between a reference and a link.

Now that that's out of the way...

In my career as a computer jockey I have had occasions to work with AI systems on a number of occasions and in my experience they fall into two overlapping classes -- systems that are simple enough that they generally do what you expect because that's what you told them to do, and systems that are complicated enough to produce surprising (and sometimes inexplicable) results. Rules engines tend to fall in the former category. Theorem proving systems are particularly buttoned down but forward-chaining engines are a bit more likely to wander off into the weeds. Neural networks often fall in the second class, especially when they are used to reason in complex domains.

At a conference I recently saw a presentation of a system that combined a little bit of both. The application controlled the dispersal of food pellets at a commercial fish farm. It consisted of a neural network that had been trained to watch a video feed and recognize fish and food pellets and a rules engine that regulated the rate of feeding. The neural network reported how active the fish were and how far the food pellets were sinking before they were gone and the rules engine would control the rate of feeding so that the food pellets didn't sink past the fish and would stop feeding when the fish stopped eating.

It worked fairly well. The neural network drew outlines around the fish in one color and the food pellets in another color. Occasionally it would miss a fish here or there and it had trouble distinguishing fish food and fish poop but it worked well enough to provide data for the rules engine to work with. It was an impressive application of AI technology and it, in effect, replicated the work done by a conscientious seven year old shaking tetra flakes into a fish bowl, only on a massive scale.

“I've seen that movie, actually several of them. We'll all be gone from PCa, global warming, little green aliens or idiotic politicians, long before AI robots do us in ;-)”

My sentiments precisely. If anything, there’s a tiny chance AI could get us out of our messes. AI could also view us as the hazard/plague that we are and treat us accordingly. But we likely will do it ourselves

I'll never get into an AI argument (okay, discussion) with a computer scientist and expect to come out with anything other than confusion. And quite possibly egg on my face. That said, I don't think the dangers from AI come from Arnie with a titanium shell but simply programs that can outthink us at who knows how many orders of magnitude. Viz. the difference between us and flatworms.

There's no question that narrow AI as it's been and continues to be developed has been a great boon to mankind. Although just the other day I was trying to reach the hospital's urology department and the pleasant female voice shunted me to neurology instead.... And I was being articulate!

And yeah, we've always been great at screwing ourselves. I'm glad I won't be around for 2100 and fear for the grandkids (although if Ray Kurzweil is right - and I don't think he is but it would be cool - we'll be able to upload our consciousness by 2045 and live forever). But many pretty decent minds in the AI field predict ASI a whole lot faster than some of those other man- or climate-induced cataclysms. (I make an exception for a nuke war; that could happen any day)

I deeply hope that when the day comes that a machine is smarter than the smartest human that it will look kindly upon us mere mortals. But given how avaricious our species is, and the fact that there are prolly hundreds if not thousands of governmental and private companies racing to get to the AI prize, I fear that greed and power will trump caution. (note: using that last verb in its original meaning, not going political)

AZ Guy said...
There will be people afraid of AI just like every other great technologic advancement. But just like every other breakthrough it is coming and will serve mankind more for the positive than the negative. So long as we can enter enough accurate data linking diagnoses to treatment outcomes, men in the future will have easier decisions to make regarding what choices to make.

Nailed it!

I've introduced some AI in my workplace. All it is is a tool to improve operational efficiency in a "big data," or at least a data-rich, environment which can accelerate human judgements...I've been using the term "augmented intelligence" in an attempt to reduce the likelihood of stupid people saying stupid things about AI.

Post Edited (Blackjack) : 4/5/2019 12:59:48 PM (GMT-6)

I'm interested to know what some of those stupid things might be.

Nailed it!

I've introduced some AI in my workplace. All it is is a tool to improve operational efficiency in a "big data," or at least a data-rich, environment which can accelerate human judgements...I've been using the term "augmented intelligence" in an attempt to reduce the likelihood of stupid people saying stupid things about AI.

I'm interested to know what some of those stupid things might be.

Agreed. Computers are only malevolent in Hollywood. Otherwise they just follow their programming. And I'm no scientist. But Bostrom's book that I mentioned earlier is fairly detailed in its description of how that programming might or might not get us into trouble.

For example, a robot/AI taught not to allow humanity to come to harm - either by its own agency or humanity's - might deduce that the simplest and best way to accomplish this would be to build a secure room for every person on the planet, closed off from any worldly harm, and supply that room with adequate food, water and AI medical care when needed. Voila: law obeyed.

As 81GyGuy said, how many laws would we need to be safe to pursue, as our great Declaration of Independence says, life, liberty and the pursuit of happiness?