Slashdot is powered by your submissions, so send in your scoop

 


Forgot your password?
Close
typodupeerror
×
AI Medicine Science Technology

MIT's AI Suggests That Social Distancing Works (venturebeat.com) 78

Posted by msmash from the closer-look dept.
In a preprint academic paper published in early April, MIT researchers describe a model that quantifies the impact of quarantine measures on the spread of COVID-19, the novel coronavirus. From a report: Unlike most of the models that have so far been proposed, this one doesn't rely on data from studies about previous outbreaks, like SARS or MERS. Instead, it taps an AI algorithm trained to capture the number of infected individuals under quarantine using the SEIR model, which groups people into classes like "susceptible," "exposed," "infected," and "recovered." This approach potentially achieves accuracy higher than or comparable to previous work, which could help to better inform governments, health systems, and nonprofits as they make treatment and policy decisions about social distancing. For instance, the model found that in places like South Korea, where there was immediate government intervention, the virus spread plateaued more quickly.

"Our model shows that quarantine restrictions are successful in getting the effective reproduction number from larger than one to smaller than one. The [model] is learning what we are calling the 'quarantine control strength function,'" said George Barbastathis, MIT professor of mechanical engineering, who developed the model over the course of several weeks with civil and environmental engineering Ph.D. candidate Raj Dandekar as a part of a final class project. "That corresponds to the point where we can flatten the curve and start seeing fewer infections."
This discussion has been archived. No new comments can be posted.

MIT's AI Suggests That Social Distancing Works More

MIT's AI Suggests That Social Distancing Works

Comments Filter:

  • 1. You mean physical distancing. We're all quite social still, as always.
    2.You don't mean AI. You mean universal functions. The degenerate of neural networks where *all* simulation realism was sacrificed for speed.

    And: Not throwing aerosol and droplets onto other people improves infection rights? No shit? You need truly pathetic in-vitro simulations to confirm that? Really?
    Have you actually checked that in ... you know ... real life?

    Does anyone have a bullshit generator for this?
    I wanna see what happens whe

    • The degenerate of neural networks where *all* simulation realism was sacrificed for speed.

      Ugh, stop it. Artificial neural nets aren't simulations of biological neural networks (when the purpose is AI), and biological neural networks are not synonymous with intelligence.

      • It's not his fault that he thinks what he sees in TV shows and movies is real.

        • It's not his fault only if he has serious psychiatric and/or cognitive issues. Other than that, if he's a person of even moderate intelligence, he should be able to access the vast resources of the United States government to inform him as to how viruses work.

          • I meant about so-called 'AI' being so much more than it really is, which is to say 'shitty half-baked software, over-hyped by marketing departments and the media'.
            Also my bad for forgetting the '</sarcasm>' tag. xD

    • 1. You mean physical distancing. We're all quite social still, as always.

      The Internet is not really being 'social', it's a lie they tell us to make us think it's the same as actually being 'social', and so-called 'social media' is the biggest lie of them all, it actually keeps people apart and encourges them to be anti-social.

      • For many of us, it's as social as we want to be. Tweak the experimental setting, and the benefits become evident. There are no absolutely good or bad behaviours. It all depends on your environment and the outcomes you want to produce.
  • That's good to know, let's do it!
  • Gee whiz, not having people breathe on each other reduces their chances of passing on an illness? Wow, such ground-breaking revelations from our AI robotic masters! /s

  • At this point there has to be enough data from enough regions of the world to know how well 'social distancing' works by comparing it to a comparable infection rate in a less dense area.

    • I agree. Why can't you just compare places that practiced social distancing to places that didn't (assuming some variables, of course). Sweden and Brazil have done little. Eight states here in the US didn't do much.

      Even this should provide fairly good views of how it plays out, and it looks like they are using cell phone data, at least for some of it:

      https://www.wired.com/story/tr... [wired.com]

      Also, it is flattening the curve. A lot of data should be similar, except one place peaked rapidly and the other will be ongoi

  • If the number of active cases drops, R0 must be less than one -- provided your testing program isn't cutting back.

    If active cases continue to increase, (a) R0 is greater than one... or (b) you're dealing with lag due to testing backlogs and delay of symptom onset.

    Signs point to (b) in most of the US. In most states the growth is moderating, and in a few states we've seen the first reductions in active cases. It's important not to jump to conclusions, but it looks like we're turning a corner.

    • If you stir all the samples into one big pot, you end up with a "regression to the mean" that isn't very informative. If you separate into subgroups by meaningful attributes, and each group is big enough to be provide statistical accuracy, you can draw likely conclusions. (That was in my Statistics 102 course.)

      • Re: (Score:3)

        by hey! ( 33014 )

        I don't think stat 102 likely covers time series data very well, but it is interesting to disaggregate stuff geographically. There's definitely states and areas within states where new cases are dropping off. It seems to take about three weeks from shutting down essential services to see the curves fall off the exponential track, and about four weeks to see active cases begin to drop.

        There's a lot of noise in the new case data because of the jankiness of our testing, but if you do a three day moving averag

  • Accuracy is what matter for the model (Score:5, Insightful)

    by larryjoe ( 135075 ) on Friday April 17, 2020 @07:28PM (#59960422)

    This approach potentially achieves accuracy higher than or comparable to previous work

    This quote from the summary is the only sentence that is interesting, but there is no substantiation of this claim. The paper has not yet been peer-reviewed.

    Furthermore,

    Raj Dandekar, a PhD candidate studying civil and environmental engineering ... has spent the past few months developing the model as part of the final project in class 2.168 (Learning Machines).

    So, a grad student who is not studying epidemiology or any other medical field and who is not studying machine learning or any other computer science or mathematics field took a grad class and developed this model as part of a class project. And he gets a slashdot article for his efforts. The power (or at least the PR) of MIT is strong.

  • Computers can only do what they are programed to do. There is no AI.

  • OK yes, we've known social distancing works since it was used in the Spanish Flu outbreak in 2018.

    More importantly, were these AIs programmed by people who themselves were convinced social distancing works? Because even if I agree factually it does, I'd be suspicious of bias creeping into the code.

  • Social distancing works as a way to slow the spread of viral diseases, true. But it also works as a way to increase sickness and death caused by depression, failure to treat non-covid diseases, and loss of insurance due to massive layoffs.

    When all is said and done, does AI tell us whether the benefits of social distancing outweigh the cost?

  • If you want a scientific analysis, you need to do statistics; trends, regressions, confidence intervals, all that tedious stuff. Trusting an AI is like trusting your neighbourhood gossip - it mainly depends on who it talked to last...

  • Why would ANYONE believe a computer model in the future? The US economy was destroyed by the flu. As of April 8th... https://www.youtube.com/watch?... [youtube.com] “Projected coronavirus deaths in the United States were lowered by 25% from 81,766 to 60,415 early Wednesday morning.” This is the Washington state model, the Institute for Health Metrics and Evaluation, Chris Murray model, IHME. This has been the lead model from the get-go. This model is why we’ve done what we’ve done, plus the mod
  • Social distancing doesn't work - how did 30% of MA resident get infected if it did work? https://www.foxnews.com/scienc... [foxnews.com]

  • Seriously, what was the point of this? Were they waiting for the results of the "is water wet?" experiment and thought they'd fill in the time.

    And as for "AI" - I ran the same (SEIR) model in a little 200 line Perl script and got the same sort of results (I was more interested in how much it might help than if it does). I suppose if I'd said it was AI then I might have got a million dollar grant. The SEIR model is iterative; it doesn't require any AI. In fact if you assume static conditions then it produces

  • Mathematician Grant Sanderson did a nice video on this at https://www.youtube.com/watch?... [youtube.com] .

    He made clear at the start and other times in the video that he was performing strictly numerical simulations, not specifically tied to the Covid or any specific disease. And that the videos are not to provide actionable information "I saw this on you tube so I should start/stop wearing mask" or whatever.

    Caveats aside the video is a good visual demonstration of exponential growth and how things..viruses, meme's, y

  • Well, duh?

Slashdot Top Deals

The truth of a proposition has nothing to do with its credibility. And vice versa.

Close