Ask HN: ML, DNN and Creativity in Battle vs. Covid-19?

[narogab]

Everything I see about Covid-19/coronavirus is old-school standard statistics - no ML, no DNNs. Doctors and statisticians screaming and warning to not do X unless we have proven X effective with classically statistically-controlled studies (which they confess, will take months to complete, write up, peer review and publish in the journal of their choice). But when we look around, having done over the last 5 decades literally tens of thousands of such studies, none of them seems to have any relevance!

Seems the medical community thinks success against Covid-19 can only be achieved with more masks, more gloves, more hospitals, more randomly-selected controlled double-blind studies and particularly more money for a (bigger) medical community.

I hate to be critical of the very people who are doing so much work, but I am sorely pressed to see how the group of dullards in charge (and I mean the entire medical and drug community) can possibly come up with a solution.

Example: look at this - this is a solution from the boneheads in charge:

https://www.jpost.com/International/Israeli-doctor-in-Italy-We-no-longer-help-those-over-60-621856

"Let the old and weak die! " What a novel solution!

Seriously, WTF happened to creativity? Where are the ML and DNN people? Where's the "artificial intelligence" (b/c there apparently isn't enough of the real thing to solve this problem. Why isn't more of the work done to date useful in finding solutions? Medical "science" has let us down.

[rvz]

The current tools that are used in clinical environments have been tested and proven to be safe for many years and are also compliant with healthcare regulations to ensure that they are safe to use because it involves human life.

You mention Machine Learning or Deep Learning based solutions are not used more to tackle the COVID-19 outbreak. The problem with using them is that the explainability in its decision-making is very poor and is as transparent as a blackbox at best. Thus it isn't useful for a clinician to give an explanation why this 'AI' came to that decision, which is very unsafe in a medical environment.

Explainable tools like decision support systems are what medical professional favour over the hype of DNNs, neural nets, or other AI buzzwords these days.

[tastroder]

(Vouching since this might be a good discussion to have.)

> I see about Covid-19/coronavirus is old-school standard statistics - no ML, no DNNs.

What you call "old-school" is also proven. The biomedical domain uses these models because they understand how they work, what abstractions went into them, and when they are reliable methods to lean on. Machine learning and DNNs are pretty useless without a specific task and good datasets (add large to that list for DNNs), very much like wet trials in the medical field.

If you look at the biomedical field in the last few months, lots of efforts already come together here that you might appreciate. Peer reviews of relevant papers are expedited, results are digitally disseminated around the world, cases are tracked with pretty unprecedented speed across bureaucratic borders.

> Seriously, WTF happened to creativity? Where are the ML and DNN people?

Where they belong. Some help people in the field, others focus on their own research, some will definitely try to put their specific methods to the test in order to facilitate other efforts.

I realize that many people these days want to help but please, this is not the time to get your covid paper on arxiv just because. Just like we should call out people profiteering from this pandemic economically, we should strive against adding pure noise. Note that this shouldn't hinder creativity but that is a completely separate thing. If you have a great idea to help, go ahead.

> Where's the "artificial intelligence" (b/c there apparently isn't enough of the real thing to solve this problem. Why isn't more of the work done to date useful in finding solutions? Medical "science" has let us down.

Medical science is doing what they can, the presumption that a buzzword can do it better than people that dedicate their life to this is neither help-, nor respectful.

[narogab → tastroder]

Thanks for vouching. However...

1. If the old-school stats work, why use NN/ML? Instead sample the big data and use a classical statistical model. This was stated to me by a business school faculty member, to my astonishment. I had no good answer. I had hoped that the new tech could do the work faster or see a pattern faster or see something that was invisible to classical stat.

2. In the medical science field I see lots of self-promotion (shades of Gary Larson's cartoon re the "Little Bang theory"): https://i.pinimg.com/736x/67/36/c1/6736c1b233ce99f0e992e3aa1... )

I don't agree that medical science is doing what they can. They seem more interested in protecting their turf. The most creative and ballsy are the Chinese guys collecting plasma with Covid-19 antibodies from healed covid19 patients. That is creativity! But right behind them are their naysayers screaming bloody murder.

An acquaintance disappeared 4 weeks ago; he's holed up in the mountains for the duration. I wish him luck but the rest of us need something more creative.

I am disappointed. My bet is that a solution will come out of left field, will be initially opposed by medical science and ultimately absorbed by the same.

Whereupon we'll be back in the same situation, waiting for the next bug to come along.

[tastroder → narogab]

>1. If the old-school stats work, why use NN/ML? Instead sample the big data and use a classical statistical model. [...] I had hoped that the new tech could do the work faster or see a pattern faster or see something that was invisible to classical stat.

That's just not these things work. Even if you had large amounts of data (which you don't), you still lack a good problem statement that neural networks can be trained towards, great test sets for these scenarios that impact human lifes, and a model that is explainable / interpretable. Otherwise you quite literally put human lifes on the line hoping your model learned something appropriate instead of using something proven and reliable instead.

Modern machine learning techniques have their place and I'm sure in the aftermath of this crisis we will hear a lot of stories where people put it to good use but it's not a case of "throw ML at it, ???, get vaccine".

> 2. In the medical science field I see lots of self-promotion

You'll find self promotion in any field, that does not really diminish the work people are doing overall.

> That is creativity!

There is a clear difference between desperate medical professionals taking a risk and misguided efforts to throw random tech at a problem without understanding it. The former might be considered creative, the latter just adds noise without helping anything.

> My bet is that a solution will come out of left field, will be initially opposed by medical science and ultimately absorbed by the same.

That's fine and I get that you might be disappointed in the status quo but there is just no reasonable assumption that deep learning will provide some left field miracle here. That's grasping at straws while people are actually trying to help. As for your last comment, I'm still hoping that we collectively learn something from this crisis.