In This Episode: Thinking about thinking that might occur in machines — for the betterment of humanity.
024: Extending Thinking Beyond Humans
Tweet
How to Subscribe and List of All Episodes
Show Notes
- Help Support Uncommon Sense — yes, $5 helps!
- Notice that the sources which follow are both recent, and in multiple countries.
- Shortages in the Healthcare Profession: “New Research Shows Increasing Physician Shortages in Both Primary and Specialty Care” (Association of American Medical Colleges, April 2018) and “Global Health Workforce Shortage to Reach 12.9 Million in Coming Decades” (WHO, November 2013)
- Oregon Tetanus Case: “Notes from the Field: Tetanus in an Unvaccinated Child — Oregon, 2017” (Centers for Disease Control, March 2019)
- AI in Lung Disease: “AI Improves Doctors’ Ability to Correctly Interpret Tests and Diagnose Lung Disease” (Medical Xpress, September 2018)
- AI in Brain Tumors: “China Focus: AI Beats Human Doctors in Neuroimaging Recognition Contest” (Xinhua, June 2018)
- Diagnosing by Voice: “Looking to Technology to Avoid Doctors’ Offices and Emergency Rooms” (New York Times, February 2019)
- EMPaSchiz: “Towards Artificial Intelligence in Mental Health by Improving Schizophrenia Prediction with Multiple Brain Parcellation Ensemble-learning” (Nature, January 2019)
- New AI Research Centers in the U.K.: “How Artificial Intelligence Is Revolutionising Medical Diagnostics” (The Engineer, February 2019)
- I thought there should also be a source for “there are six different outbreaks of measles” currently in the U.S., and found out I was wrong! There are now eleven — in California, Colorado, Connecticut, Georgia, Illinois, Kentucky, New Jersey, New York, Oregon, Texas, and Washington. (CDC, February 28, 2019 [which page is subject to periodic updates].)
- No episode next week: I’ll be on travel.
Transcript
Driving home from the theater with friends Friday night, my wife started a conversation about how difficult it is for modern medicine to diagnose thyroid problems. Both she and her friend in the back seat have that to differing degrees, and neither one of them has received a real diagnosis. And then I, and the friend’s husband, widened it out from there.
Welcome to Uncommon Sense, I’m Randy Cassingham.
Diagnosis is the first step toward treatment, or even a cure, for any disease, but getting it right takes expertise, experience, and something doctors have precious little of: time. So unless the problem is objectively obvious like, say, a heart attack where the myocardial infarction can be located, measured, and quickly treated with well-established protocols, then it gets complicated fast. How about chronic pain, gut issues, alteration of mental status, and, well, thyroid problems?
Those are difficult at best to sort through, and therefore awfully expensive to fully diagnose in an era of a shortage of the very professionals we need to help us. The Association of American Medical Colleges estimates that by 2030, the United States will have a shortage of 121,000 doctors. And it’s nothing new: in 2013, the World Health Organization estimated that “the world will be short of 12.9 million health-care workers by 2035; today, that figure stands at 7.2 million.” Again, that was 2013. Now, we’re short somewhere between 7.2 and 12.9 million healthcare workers: not just doctors, but nurses and the technicians that run the machines to give the doctors the data they base their diagnoses on!
“Our ability to do things, for example, cure a headache or treat anxiety,” says Dr. Alex John London, the director of the Center for Ethics and Policy at Carnegie Mellon University, “often comes before our ability to explain why what we do works.” So doctors often end up treating symptoms without any real clue as to what the underlying problem is. Oh, and sorry: your 8 minutes of face to face doctor time is over. If you have more questions, make another appointment on your way out. Well of course we have more questions! But that doesn’t mean the doctors have answers for us.
And then we have groups trying to throw doubt on even well-established science. Do vaccines cause autism? No — but the online hype over a tiny minorities’ objections give parents pause: what’s the best thing for my child? So going online to “Dr. Google” brings all sorts of conflicting opinions, with authoritative sources mixed with ridiculous conspiracy theories given equal weight, without context to make clear what the consensus really is: that the risk from vaccines is so tiny as to be ignored compared to the absolutely clear harm that the diseases they prevent cause.
As I was putting this together, one of the news stories I saw was about a 6-year-old boy in Oregon who had the misfortune to be born to anti-vax parents. As little boys do, while out playing he got a cut on his forehead. The parents stitched up the wound themselves, but he didn’t get the basic treatment that’s totally standard in such cases: he didn’t get a tetanus shot. When I was around his age, I stepped on an old nail, and my parents took me straight to the doctor not because of the inconsequential wound, but to get the shot because something dirty had been shoved an inch into my body.
Well, the Oregon boy didn’t get it. Several days later he got something else instead: tetanus. It started with him being unable to open his mouth because his jaw was clenched tight — the old name for tetanus is “lockjaw”. He then had more general muscle spasms, his neck and back arching painfully. At least his parents called 911! Thankfully, the medics who responded recognized how serious it was and called for a helicopter to take him straight to a pediatric hospital. The poor kid was having trouble breathing because yes, even his respiratory muscles were affected. Thanks to the medics’ quick action, the boy was saved — barely. His body temperature, blood pressure, and heart rate went wild, and he was admitted to intensive care.
Only on day 44 was he able to be taken off a respirator, and he was able to take his first sips of water. On day 47 he graduated out of intensive care. On day 50 he was finally able to walk a little — 20 whole steps. On day 57 he graduated to a rehab center, and finally got to go back home to his so-called loving parents 17 days after that. By the way: total costs for the hospital care alone: $811,929. That doesn’t include the air ambulance, rehab, or any follow-up care, so let’s round it out to about a million dollars, most of which will be on the backs of others.
And you know what? Reports say his parents still refused to let him have any vaccines — not even a tetanus shot. No wonder the really smart and aware kids are starting to go behind their obliviot parents’ backs to get vaccinations on their own!
The tetanus vaccine was developed in 1921, and since then the number of tetanus cases has dropped to nearly none. This little boy’s illness was the first case of tetanus in Oregon in more than 30 years. Why? Because parents and doctors know how bad tetanus is and most kids get their shots so they don’t have to go through this kind of agony, or worse. These days, a lot of parents have never heard of, let alone seen, tetanus — or measles, or polio, or diptheria, or Rubella, or …well, the list goes on. They don’t know the terror their ancestors had to go through. From 1915 through 1997, the infant mortality rate dropped more than 90 percent.
But what does that really mean? If turn-of-the-century infant mortality rates were not reduced, then in 1997, about half a million children born alive in the United States would have died in their first year. Instead, due to all causes, that number was dropped to 28,045 infant deaths. That’s huge! The problem? The number is going to go up again if the anti-vaccine movement continues: as I speak there are six different outbreaks of measles, a disease that was essentially eradicated in the U.S. around the year 2000.
But this is a tangent: luckily, that tetanus diagnosis was pretty quick and easy. How are we going to improve diagnosis in general? More and more and more studies are coming out with information, new treatments being figured out and old ones discarded as inadequate or worse. Even if there were enough doctors, they couldn’t possibly even glance at it all, let alone remember the details of what they might see in their careers.
If only there was some kind of technology that could somehow store every single known fact about every known disease and treatment, and keep a running tab on what’s working for, say, a brain tumor in a male Asian 9-year-old who has a prior history of leukemia that was treated with a bone marrow transplant. Because that will almost surely be different from the brain tumor treatment for a female Caucasian 17-year-old with a history of appendicitis!
You know, like a computer! And that’s what I said in the car when Tim was talking about how he and his wife are struggling to diagnose her thyroid problem. He’s been driving her from doctor to doctor around the region, trying to get an answer. And a treatment. If only there was a computer system to help, if only there was a comprehensive data base available.
It’s difficult to sort the hype from the promise of artificial intelligence, so maybe the first ambiguous problem that should be put on A.I. …is to sort out the hype from the promise in the A.I. field!
But seriously, why not have computers do the jobs that require enormous amount of data processing and sorting? Better yet, they can take care of more than one patient at a time. A lot more. And we’re at the level now to do it. Let me give you some of the early examples.
The Laboratory for Respiratory Diseases at the Catholic University of Leuven, Belgium, used quality data to train a computer to read pulmonary test results. They found that the system was more consistent and accurate in interpreting the tests, and making the diagnoses, than lung specialists. How much more accurate? Twice as good as their team of 120 pulmonologists from 16 hospitals.
In China, the Capital Medical University in Beijing built a system to diagnose brain tumors. They found it to be 87 percent accurate. Yeah, I know you were hoping for 99.9 percent accuracy, but I’d rather have the computer on my side: they compared the machine to a “team of 15 senior doctors,” whose accuracy was just 66 percent. Even better: it took the computer just 15 minutes to go through 225 cases.
A really interesting system being worked on right now uses A.I. to diagnose illnesses, like Alzheimer’s, heart disease, and sleep apnea from …how patients speak! It sorts through pitch, pauses, breathing, and hundreds of other factors. Rich Ross, health care research director at the research firm Gartner, explains, “The manner in which we speak and the word choices we make can be evaluated to accurately detect a growing list of clinical conditions.”
Which makes me think of people with mental conditions. There’s a lot of overlap, causing mixed symptoms. What treatments would help, or harm, the patient? We’re making strides there, too: computers can now diagnose schizophrenia with high accuracy. The system, called “EMPaSchiz”, or “Ensemble algorithm with Multiple Parcellations for Schizophrenia prediction” (say that three times fast!), is also 87 percent accurate.
All of that is surely just a start, and if you think about it, all the complexities just mentioned are pretty much true for any body system: overlap, mixed symptoms, differing treatment regimens. So isn’t it pretty much just Uncommon Sense to start working toward a more comprehensive A.I. solution to diagnosis?
Others think so: “You could look at almost any area of healthcare and see that advanced data science has an enormous amount to offer,” says Sir Mark Walport, chief executive of U.K. Research and Innovation, or UKRI. “This technology has huge potential right across the world of healthcare.” And what is the UKRI? The British government saw such promise in this field that they’re spending 50 million pounds — really, just a trifling sum considering the enormity of modern medical care — to establish five new research centers to get these systems up and running. “The potential here,” Wolpert says, “is to have a computer algorithm assisting humans in what are quite difficult diagnoses.”
How long do you think it will take before that 50 million quid, or about US$65 million, is paid back? My bet: less than a year from the systems going online in regular practice.
Are there worries about A.I. maybe taking over the world and enslaving humanity? To me, that sounds more like human insecurities talking there — doom and gloom science fiction. I consider that a very tiny risk compared to the enormous potential for relieving illness and improving life.
Funny, but that sounds just like vaccines!
In other words, going down this path isn’t just necessary, but smart. It really is Uncommon Sense in action.
The Show Page for this episode, which includes links to my sources, and a place to comment, is at thisistrue.com/podcast24
I’m Randy Cassingham … and I’ll talk at you later.
– – –
Bad link? Broken image? Other problem on this page? Use the Help button lower right, and thanks.
This page is an example of my style of “Thought-Provoking Entertainment”. This is True is an email newsletter that uses “weird news” as a vehicle to explore the human condition in an entertaining way. If that sounds good, click here to open a subscribe form.
To really support This is True, you’re invited to sign up for a subscription to the much-expanded “Premium” edition:
Q: Why would I want to pay more than the minimum rate?
A: To support the publication to help it thrive and stay online: this kind of support means less future need for price increases (and smaller increases when they do happen), which enables more people to upgrade. This option was requested by existing Premium subscribers.
I can’t wait for a reasonably accurate AI system that can diagnose & recommend treatment for EVERY common disease/condition & most uncommon diseases/conditions (success with the common ones will give Drs. confidence it can accurately handle uncommon ones). Most Doctors can recognize & treat most common diseases, but hit them with an uncommon disease or a condition they’ve never seen before or were never trained for, and they’re lost.
I’m reminded of an episode of “Emergency” where an intern at a major LA hospital failed to diagnose a teenage boy with a case of polio because he’d never seen one before. The disease was so uncommon that his training hadn’t even covered it! A senior Dr. happened to overhear a discussion of the boy’s symptoms, and recognized the condition. Turns out the boy’s parents hadn’t had him vaccinated against polio because “everyone knows polio’s been eliminated”. With the (sadly) seemingly widespread incidents of non-vaccinated children similar situations are going to become more & more common.
Doctors can only be trained for so much & they can’t be trained for everything. With more & more kids going unvaccinated due to unfounded fears about vaccines, Drs. are going to start seeing diseases they were never trained for because the disease had long been considered eradicated. As a result, many patients are going to suffer needlessly despite their Dr’s best efforts. The public DESPERATELY needs such an AI and we need it NOW!
While AI systems can learn to do quite a bit, they are often terrible at extrapolation past their training sets. Since they can fail spectacularly, they should only be used as a guide. A few AI failures from 2017 are here.
There are other examples which could have been catastrophic (e.g., a system which learned to distinguish cars from tanks so it could target tanks but mysteriously thought every vehicle on the battlefield was a tank).
The problem with current AI systems is that they cannot explain why they came to a particular solution. This both limits how much we can learn from them and how testable they are because they silently absorb any bias in their training sets. I have been following this field since I spent several years working at Thinking Machines (an early AI and computing company) in 1984 and I think you are being overly optimistic. Maybe we will get there but I doubt it will be with the current crop of systems.
When reading all the glowing reports of success, do not neglect the file-drawer effect. In order to keep funding flowing, researchers rarely publish their failures but if you put the system into medical practice, we become the guinea pigs.
—
My conclusion wasn’t to put a comprehensive A.I. system in place tomorrow. Rather, it was “So isn’t it pretty much just Uncommon Sense to start working toward a more comprehensive A.I. solution to diagnosis?” Yes, we need to be damned careful, using the systems in parallel with a real doctor for a long while, and confirming the diagnoses that are made (whether by a human or computer) to learn from the mistakes — and there will be plenty, because there have been plenty all along. We’ve been the guinea pigs since medicine began! -rc
I agree with all of the arguments, it is a shame that so many children will suffer because their parents won’t see reason. But what about a vaccine against abortion? That would save an awful lot of beautiful, healthy babies’ lives. Just a thought.
I just learned about the book, Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again by Dr. Eric Topol.
It seems to make some of the same points you did and also covers some of the places where he sees problems and limitations.
—
Looks very interesting! Thanks for the tip. -rc
“finally got to go back home to his so-called loving parents”
Really Randy, how can you call them “so-called?” Why is it that you think anti-vaxers couldn’t be loving parents? It is possible for them to refuse vaccines because they love their children, and wish to protect them, even if the information they base their decision on is false.
In my opinion, parents should be educated, not shamed. They just may change their opinions if properly informed. In fact, I’m sure you have changed opinions when given new facts, I know I have.
—
Of course I have changed opinions on matters when presented with additional information: that’s what intelligent people with open minds do. Yet studies of anti-vaxers find that the more “education” that’s given to them, the more they resist and stick to their own warped “facts” …exactly as these parents did. The “education” is wasted. Do I think it’s 100 percent true of all anti-vaxers? Obviously not. But it’s demonstrably true in this specific case: it’s ideology over proper and loving parenting. I thus stand by my statement. -rc
“Are there worries about A.I. maybe taking over the world and enslaving humanity? To me, that sounds more like human insecurities talking there — doom and gloom science fiction.”
Really Randy?
The problem here isn’t AI itself. The problem is the other agendas business, the military and governments place above benefiting mankind. The twentieth century is full of examples of inventions and developments co-opted for profit, or for military advantage or for political expediency. As well, the same obliviocy you often highlight exists among the scientists working on these developments. They often exhibit too much hubris to admit they don’t understand what they are playing with. And, just like the anti-vaxers referenced in a previous comment, these people entrench their position when confronted with inconvenient facts rather than behaving in an intelligent manner. The simple truth is, we humans are not yet sufficiently mature to have AI available.
—
I don’t know what “Really Randy?” means here: I say it’s silly to demonize A.I., that fearing it is about human insecurity rather than anything real, and you disagree by saying A.I. isn’t the problem, it’s how humans co-opt and twist technology that’s the problem? Yep: that’s another way to put it. -rc
Let me preface my comment by saying I’m 100% in favor of advancing AI tech. Also, let me apologize for the length of my post but this is a critical issue and I’ve having trouble getting any traction on it as its difficult to explain. I’m looking for someone to help me spur discussion in this direction, so far not having much luck.
The real risk with AI is economic. It *may* not happen, but lets imagine what the world looks like if 100% of human labor is no longer needed because AI/robots do everything faster, better and cheaper, i.e. we’re talking about 100% unemployment. While some point to a UBI (universal basic income) as the solution, its only a partial solution that could actually help to bring about the worst case scenario.
Background: everyone talks about our free market economy but most focus on the consumer and capital (stock/bond) markets, but there’s a third *critical* market we mostly ignore, the labor market. When functioning properly, the markets work hand in hand to determine resource allocations, i.e. how we put different resources to use meeting people’s needs and wants. It works fairly well, but it only works because the consumer and labor markets counterbalance each other. In short, people have to be choosy with what they buy because they are essentially exchanging their limited supply of labor for that good/service. On the flip side good/services have to offer enough value so they can cover the cost of the human labor required to provide it. Those trade-off considerations make sure resources go towards things people want badly enough to pay other people to provide.
Technology up until recently has been driving the human labor costs lower by making people more efficient, i.e. they can do more so the per unit cost goes down while the value of a human’s time goes up — win/win (the stuff they make sells for less while they get paid more). This is what is behind our increasing standard of living, increased efficiency of human labor. But with automation, technology is shifting from making people more efficient to replacing them altogether. Its making enterprises more efficient primarily by reducing the need for people, less so getting more out of the people they have. Fortunately, so far, there have been replacement areas for displaced workers to go, but that may change soon.
Taking this to its conclusion (assuming its possible), 100% of jobs will be done by automation, I’m talking even CEO type jobs. Presuming AI/robots out-think and out-work humans presents three big problems. The first is no one will be able to earn an income through selling their labor. While bad its actually the smallest of the problems, it can be fixed by something like a UBI. The second is bigger and harder, without the labor market we no longer have a check on the consumer market. Instead of the natural limit of the value of their own labor, any check on consumption will be artificial, i.e. political. In a UBI world it’ll boil down to people voting on how big an income is “basic” — I don’t see that ending well.
The third problem is the biggest and hardest of all, but its also hard to explain, I’ve found that most people simply can’t wrap their minds around it. To try, I’ll back up a bit, we actually created the modern labor market when we outlawed slavery and feudalism. In essence we used those laws to force resource owners (“the rich”) to use money to pay for human labor. That change is at least as important as any other in creating the modern world and our high standard of living. By forcing resource owners to pay for labor, we forced them to sell stuff to earn the money to pay for it. Either directly or indirectly they earn that money by providing the goods and services we all want. They don’t do it out of generosity, they do it so that they can pay the labor costs for things *they* want. What happens when the rich no longer have to pay for human labor because they own AI/robot labor? Oddly enough, the rich will no longer need money, which in turn means they no longer need to sell us stuff. I’m not talking just about widgets and gadgets, I mean they will no longer need to sell us anything, including raw materials like lumber, iron ore, energy sources, etc. — and even food. To be specific, I’m talking real wealth as measured in ownership of resources (land, mines, factories, robots,etc.). This is where a UBI only solution really fails, what good is a UBI if there’s nothing to buy?
And this is where I generally lose people, the rich won’t need money???? So perhaps an illustration. Right now, if Richie Rich wants a new yacht he has to pay for a lot of human labor. Not just the people who build the boat, but the lumberjacks, miners, refinery workers, truck drivers, accountants, lawyers, engineers, managers, secretaries, janitors, etc. that are needed to get all the resources in the right place, in the right condition/shape for the boat to be constructed. Now, in a world where that’s all done by robots, robots Richie Rich owns, he has essentially zero labor costs. If he also happens to own the forests and the mines and sources of the other raw materials he needs he has no external costs at all, that new yacht will cost him $0. It will still carry internal opportunity costs, but that puts resource allocations solely at his whim. Now, in this nightmare world, it won’t just be new yachts, it’ll be the food he eats, the house he lives in, the clothes he wears, etc., *nothing* he wants will cost him money so what’s he need money for? If he doesn’t need money he’ll neglect those things that earn him money — like his businesses that sell us stuff. Worse, his mines and many of his other resources are limited, why would he sell off the stuff he wants to use for his own purposes to gain something he doesn’t need (money). The resources owners could become very stingy with their iron ore and lumber, etc., reserving it for their own use. Sure, they may barter with other resource owners for those raw materials they don’t directly own but dealing with the general public, aka the masses, will be something they can simply walk away from.
The only solution I see that address all three problems is to find some way to save (or replace) the labor market. We *need* the functions the modern labor market provides: 1) providing incomes, 2) prioritizing and limiting consumption, 3) strongly encouraging everyone (especially resource owners) to participate in the general economy. The most direct path there, and the one that would be the most familiar to us, would be to require people to pay other people for the robot labor they use. This would require finding a way for the masses to exercise a right to sell labor even though it isn’t their own labor they are selling. I call it a right because today we all inherently understand that a person has a right to voluntarily sell their own labor, but in the fully automated world that is meaningless. That right needs to be rephrased as something along the lines of “the right to support oneself through the voluntary sale of labor they own”.
There are various way to approach that but I think the best approach is through requiring people to purchase the CPU cycles they use and granting every person an allotment of CPU cycles they can sell. The devil is in the details, but I think it can be done in a market oriented way. Especially challenging are things like increasing the supply of CPU cycles to support a growing economy while at the same time limiting the supply so that the price is high enough to support a “living wage”. We also have to make sure that the rich don’t wind up monopolizing such a market. I think we can set up markets that will address those problems, we just need to work through it and probably test it. That’s why we need to start discussing and working on solutions now, before its upon us. The push for a UBI is growing, but as I said its only a partial solution as it only addresses one of the three problems full automation creates. If we allow a UBI to be the only solution, people will be mollified until its too late.
If we don’t find a way to replace the modern labor market with something that performs all of its functions I fear we face a future society that looks an awful lot like feudal/slave based societies — but even worse. At least in those societies the rich still needed human labor so they needed their workers to stay healthy enough to work. Imagine that same world where the resource owners have zero concern for the rest of humanity. It wouldn’t even have to be out of contempt or even greed, it will happen simply because they don’t need the masses or have to interact with the masses. Their every whim or wish will be taken care of by the robots/AIs that wait on them hand and foot.
It would sure take a long long time to create a world economy run by AI. I doubt that robots could accomplish all the non-routine tasks now done by us mere mortals, so I will sleep well tonight knowing that my work will be there, waiting for me for many tomorrows :~]
People have been thinking about this. The short answer is tax robots!
https://qz.com/911968/bill-gates-the-robot-that-takes-your-job-should-pay-taxes/
https://taxrobots.org
https://www.nytimes.com/2019/02/23/sunday-review/tax-artificial-intelligence.html
My view is that robots can produce common goods (e.g., the stuff anyone can buy in a grocery or department store) or stuff exclusively for Richie Rich (robots owned by RR). Levying a much heavier tax on individually owned robots can help mitigate the risks you mentioned.
Another timely article from the NY Times. Researchers have found ways medical AI can be hacked. As for motivation, say a hospital is having trouble getting reimbursed for CT scans because of the behavior of the AI reading the scans. It would be very temping to manipulate the scans before the AI sees them to ‘fix’ that ‘problem’. Of course, the altered scan will probably end up in the patient’s permanent record but ‘hey, we got paid’.
https://www.nytimes.com/2019/03/21/science/health-medicine-artificial-intelligence.html
We really need to think this through rather than just allowing a rush to market to recreate a mess even bigger than the current EHR debacle. That’s going to take some uncommon sense.