How AI startup BlueDot spotted Coronavirus before the WHO | Hacker News Reader

How AI startup BlueDot spotted Coronavirus before the WHO

diginomica.com

18 pointsCzarcasm10 comments

The article makes it sound like BlueDot is some kind of advanced AI capable of sifting through the tea-leaves of news and alerting the world of a possible pandemic well in advance of official, governmental organisations.

But at least in the case of COVID-19, the truth appears to be that BlueDot was not that much faster than a properly functioning bureaucratic machine.

BlueDot: the article says BlueDot sent an alert of unusual pneumonia on 30 Dec 2019.

Properly functioning bureaucratic machine: WHO received info about a possible new outbreak of some sort from the Chinese government on 31 Dec 2019. Taiwan, a non-WHO member (and which is politically and socially inclined not to trust China on anything), reacted immediately to the news and began monitoring passengers coming from Wuhan on that day [0]. In Singapore, the health authority sent out an advisory regarding this potential new outbreak on 2 Jan - essentially the first day everyone got back to work after the new year celebrations [1] - and announced traveller screening measures. Hong Kong stepped up border screening immediately, too [2].

I dunno... I feel like the knowledge would not have been disseminated usefully faster in this case.

If anything, the important takeaway is not that we should have been alerted faster, but rather how important it is that government institutions are funded, function, and are staffed by competent individuals.

[0] https://wwwnc.cdc.gov/eid/article/26/7/20-0574_article [1] https://www.moh.gov.sg/news-highlights/details/precautionary... [2] https://www.scmp.com/news/china/politics/article/3044050/mys...

In terms of timeline apparently Alberta, a province of Canada, was already stocking up on masks and aprons in December in anticipation of the coming pandemic.

BlueDot doesn't use social media which is where there might be gains.

Here seems a good timeline of what happened, which Bluedot didn't win -

To me the important thing was social media discussing a media release and a central authority finding this was where it was at, no AI, just emergent intelligence from social media.

https://thenextweb.com/neural/2020/02/21/ai-sent-first-coron...

This is false. It was not conducting an intelligence operation across China. It is obviously playing the prophet card, as FiveThirtyEight did, in an attempt to claim authority on that which it has none.

Can anyone find any details on how often BlueDot produces false positives? Finding Coronavirus while giving out 20 false warnings a day would be a much less impressive result for example.

There are a couple of comments beneath the piece that suggest traditional news organizations (Reuters etc.) were reporting on it only hours after the 23:00 alert of BlueDot. If the AI cracked it early then they only beat reporters by hours - maybe that's significant, maybe not.

Perhaps Reuters is using BlueDot?

Could be, indeed.

I read the article/press release.

Just so people have the early timeline and other facts so far:

- The first patient that needed hospital care in Wuhan was Dec. 6, indicating infection around Dec. 1. (See timeline link below.) I suspect this started earlier in November, but the majority of Covid-19 cases don't involve the lungs. [0]

- a powerful unknown flu (same symptoms as Covid-19) was circulating in mid-Dec. throughout SE Asia. I was in Shanghai and aware of this around Dec. 15. I still don't know who I would contact if I had that kind of information next time.

(I'm American, so that makes me one of the very first Americans to know about this.)

- Bluedot parsed Chinese newspaper account and communicated around Dec. 30

- China informed WHO around Dec. 31

https://www.who.int/csr/don/05-january-2020-pneumonia-of-unk...

- one Youtube documentary says there's 2 virus labs in Wuhan, with one focusing on bats, FWIW.

- Wuhan International Airport is a 2 hour flight from Shanghai Pudong International Airport. The latter is a major international hub, #8 in the world by passenger volume.

https://en.wikipedia.org/wiki/List_of_busiest_airports_by_pa...

It looks like BlueDot just scans news reports, so is not a primary source. I read the article/press release and didn't learn anything new aside from the Top Corona Cities paragraph, all linked by airports.

What's interesting is that both BlueDot's Khan and myself followed Toronto SARS-1 in 2002/2003, and that experience motivated each of us to investigate SARS-2 very early. That should be a lesson for policymakers - staff with experienced people who've seen it before.

Regardless of whether this is really AI or not, had it told the right people quicker, that would be valuable since obviously there was a serious communications problem here, and an on-going problem for the Chinese government if the virus labs are involved.

So the AI is not predicting new events, but summarizing and communicating events of interest.

I think it would be worthwhile for somebody else to treat the BlueDot article/press release as raw material for more analysis. I have a feeling there is valuable information in it that needs more massaging, after discarding the PR component.

To some extent, what BlueDot is doing is what:

- people have used Google search or trending for

- US police fusion labs were doing with Twitter and Facebook API access, before those were reported to "have been turned off."

[0] Covid-19 Timeline (Starting Dec. 1, 2019)

https://www.nationalreview.com/the-morning-jolt/chinas-devas...

nationalreview is a right biased report[0], might not be qualified as a trusted source

[0]https://mediabiasfactcheck.com/national-review/

Reply on news.ycombinator.com