DeepMind readies first commercial product

No need to rely on Chrome.

Currently in the US the largest health insurance provider is actually medicare/medicaid. When a patient with medicare goes to the doctor, the government says "This is what we will pay for that procedure, no exceptions." The hospital has a choice to either accept that rate, or to lose out on the massive medicare market.

Private health insurances plans have vastly fewer subscribers, and don't have the power to negotiate prices like the government does.

But regardless, why is health care and insurance a for-profit industry? It creates incentives to put profits ahead of people's health.

If you have insurance, the price tag will be X, and they will pay for some part (most?) of it.

If you don't have insurance, the price tag can easily be 100X and it's your responsibility.

If you have insurance, but you happened to pick up that can of pepsi from the left aisle (out of network) instead of the right aisle (in network), they won't cover it, and you're stuck with the 100X price. You won't know that before you buy the can of pepsi, of course.

If you try to ask your insurance if they cover the specific can of pepsi from the left aisle, they may or may not tell you. They may or may not give you the right answer, and if they tell you it's covered but then refuse to cover it, it's your problem (and again, the price will be 100X).

This is an important factor in how we choose to build products out of AI advances. If your product replaces a part of a service chain that is not either right next to the transaction or in a competitive part of the supply chain, it might only help incumbents aggregate more power and charge more rent. If you can deploy in a way that improves the competitive landscape, you not only distribute gains more but probably keep more leverage as a solution.

This is harder for AI than many technology areas because incumbents tend to have more data and regulatory protection on that data. Health care might be the boss level for that problem.

Every time I visit the doctors in the US there is a trail of paperwork produced. Outside of basic checkups (where you don't dare mention any niggling health issues because your instantly going to get billed consultation fees) there is always a back and forth between the patient, doctors office and insurance over what the doctors office is asking to be paid and what the insurance says it is prepared to pay. This is an enormous time wasting game - the doctors (or their medical group) ask for much more than the procedure cost to perform knowing full well that the insurance company will push back with a lower accepted payment and the patient will sat in the middle having to negotiate between the two.

Removing the pricing game would lower costs substantially (and vastly reduce the time wasted by regular Americans calling their insurance/doctors trying to sort out their individual billing messes).

Having lived in both the USA and the UK there is no question in my mind that the US health care system is a dysfunctional mess. In my experience the only place the US system has an advantage is that the healthcare providers are very (too) willing to send a patient for tests/scans to check for every eventuality and are happy to prescribe whatever medicine is 'best' for an ailment.

In the UK a fit person with a common cold would get sent home and told to not waste the doctors time, in the US you'll get a consultation and a dose of Z-Pack (and hey if I'm paying for my insurance I should use it when I get sick, shouldn't I).

I view the whole issue stochastically, with the immediate aim being to make (significantly) fewer errors than the current approach, which is having a human decide. I don't claim that I can design an experiment which could selve as an indication whether we are improving upon human judgments, but I think this should be the goal.

Reflecting upon my view, i think it comes from the experience of training ml-algorithms. You are always minimizing errors, but you goal is almost never to make 0 errors, because often your data is noisy and you are probably overfitting. I know the medical enviroments are more sensitive, but I can't really wrap my head around how we could design a learning algorithms that does not make any error and works on all abnormalies. I think it will always missclassify.

Rephrasing my argument: I think the approval should be given if an significant expected improvement over the distribution of real-life abnormalies can be detected and not over the uniform-dsitribution over all abnormalies.

EDIT: detecting out-of-distribution samples is hard and I don't think this is a solution and leads to a false sense of security.

1. If an MRI can be done 10x faster with the same results except in exceptionally rare cases, might that not still be a win? Order of magnitude reduction in time may translate to substantial reduction in cost and increase opportunity for applications. It seems like it is worth considering whether these benefits might be worth compromising the accuracy of the imaging.

2. How accurate are radiologists at diagnosing / detecting these rare abnormalities that validation might miss? If radiologists are actually pretty mediocre at this, might it be OK to make the scan slightly imperfect if the next stage (the human) is already very imperfect?

But if you are trying to bring this to masses one can start with with something "simple" like bones or some some relatively trivial organ. Not that MRI prices themselves are dropping much - gotta wait for room temperature superconductors.

Rather than saying "very dangerous", it would be safe to say that some individuals woudl not be well-served, but ideally, the overall health of the population would increase for a reasonable expenditure.

The point here is making people healthy, not providing some sort of validated theoretically clean design. If the rcts say it works, bring on the deep learning voodoo.

(edit: I, for the record, want to trust in average accuracy.)

You don't need high costs to charge a lot.

Adding more "cameras" to our MRI so we can do parallel imaging is expensive.

From the perspective of a MRI purchaser (say clinic or hospital director), maybe you can squeeze 1 or 2 more cases in a day, but it seems like a marginal improvement overall. I think cardiac is clearly the exception where you bring on board potentially new functionality with quicker acquisition.

For criminal cases though: the current judicial system is wayy too punitive. And an AI that would apply the letter of the law would likely criminalize society even more than what has already happened so far.

One happy scenario would be if the laws were more responsive and were changed to not be so punitive since the AI would have a high rate of conviction.

But then, you might have other failure scenarios. Rich people buying AI programmers and hackers to mess with the system.

Its a constant game of outsmarting the latest tech.

We are in for some truly exciting times.