Is One Concern responding to emergencies, or sucking up a quarter million dollars a year in funding to paint a map using "AI"? That's a lot of actual boots on the ground, saving people.
I think it's reckless to make claims like this even with genuinely useful technology because it triggers the "Arrogant engineers think tech is the answer to everything" response which leads to people in other fields dismissing things out of hand that could actually be worthwhile.
For one, it’s great to see startups tackling hard problems like this: generally I believe there is huge benefit and opportunity for startups in providing government services like this. It certainly helps to dispel the feeling that our brightest minds are working on ads.
Secondly, this really seems like the initial phases of product development and fit — agencies would be foolish to simply rely on a startup like this to coordinate 100% of disaster response, I’m sure FEMA has similar models. With startups, especially ones tackling more “serious” domains (for lack of a better word) I’m willing to have a longer leash and proceed with caution.
> Ivan Porto Carrero, who oversaw a team of engineers at One Concern, said he was fired in June after speaking out against what he viewed as a company culture of dishonesty. He said the usual start-up attitude of “fail fast and try something new” was inappropriate to apply to disaster response because, “If you fail fast, people die.”
I can already see how something like this will further increase inequality due to unaccounted biases in the training data, for example worse no emergency response in black neighborhoods because there is less data about them.
The question is whether this bias is better than just relying on government agencies, which we know are absolutely infested with racists and fascists.
Turns out, humans are also pretty good at probably being almost right. Luckily though, they've figured out how to build systems with higher fault tolerance than their own.
It's easy for an ML model to do great in the average case but horribly in the worst case, so you're correct if "fault tolerance" is a simple loss function to be minimized, but wrong if "fault tolerance" actually means what we care about. That's much harder and not something we uniformly know how to do with ML. Like the costco example in the article.
To do that with ML, they'd need a loss function that captures what we care about and data it sounds like they don't have.
We do know how to build systems with high fault tolerance the old fashioned non "AI" way.
My experience with disaster response is a month in Louisiana helping to operate the Rapides Coliseum shelter managed by the American Red Cross after hurricane Katrina in 2005.
Those who are resourceful, can solve problems creatively, and take initiative will lead a disaster response while those granted rank and position tend to spin wheels and control a situation they don't understand. Disasters aren't responded to by staff employees with adequate training. If it's a major disaster, people who have never worked together nor acted in such a capacity are loosely organized together and used to channel resources and effort to victims and their families. Technology absolutely plays a vital role in disaster response but at times it can be a major hindrance. I recall when agents of FEMA came to the coliseum, they asked where the computers and network were. The location did have them, unlike others, but over time they went down and no one has available to maintain them. I explained to the agents that they will have to use forms and manual entry the old fashioned way. They just looked at each other and left. Several days passed. Then, FEMA returned with a mobile command center, featuring computers and satellite communication.
The experience didn't surprise me but made me realize how dependent people are on technology and how unprepared they are to handle circumstances they haven't trained for.
"In classes that fall, Mr. Wani and other graduate students turned the problem into a project. Using crowdsourced ground-shaking reports from previous earthquakes, they trained a computer system to predict the areas of greatest impact so that responders would not have to rely on 911 calls."
After a major earthquake, I seriously doubt competent emergency dispatch operators will be proactively sending critically-understaffed ambulances, fire engines, and police to a location identified by old, crowd-sourced cell phone shake data cobbled together for a grad-school project.
Not that emergency systems IT is not without challenges: old (often third party) technology, overly stove-piped administrative domains, and just plain old out-of-date technical skills. In other words, there's lots of low hanging fruit for bright young programmers looking to serve their community!
People should probably watch Silicon Valley before giving interviews. Repeating satire verbatim isn’t a good look.