Interesting Links for 22-04-2018

[andrewducker]

a 'diseased' film, a trip to Nigeria, dentistry, lasers, X-ray tomography, algorithms and some goo - recovering old tv shows

(tags: tv history Technology )

How plastic-eating bacteria actually work – a chemist explains

(tags: plastic bacteria )

Thermal Imaging Cameras Could Make Self-Driving Cars Safer

(tags: heat automation driving )

Just How Random Are Two Factor Authentication Codes?

(tags: numbers psychology security )

Dog's 'cancer' turns out to be four teddy bears

(tags: dogs cancer )

Meet the journalism student who found out she won a Pulitzer in class

(tags: journalism )

Donald Trump has always been worth less than anyone thought

(tags: usa politics money fraud )

One in three heart surgeons refuse difficult operations to avoid poor mortality ratings 

Goodhart's Law in action
(tags: doctors death OhForFucksSake )

Comments

[agoodwinsmith]

The plastic thing is very promising ... but ... (there's always a "but").

Apparently some cars now have fan belts and other little bitty fittings made from soy. Mice find them very tasty.

[danieldwilliam]

I think the article on thermal imaging cameras and self-driving cars is a good illustration of some of the reasons why I think that the wide-spread roll out of autonomous cars will take longer than we might like.

Uber has suspended their testing programme "indefinately" which probabaly means until they work out what happened and how to fix it and then have persuaded the authorities to let them start again.

And that's one bug in one system. There's a difference between the vehicles being able to get from A to B and the vehicles having proven that they can always get from A to B without hurting anyone.

[danieldwilliam]

The dodginess of Uber taken as a given for me it's more an example of the sort of remedial work autonomous car manufacturers will have to do when something goes wrong. If they have a serious accident then a likely outcome is that a) they have to integrate a new sensor, b) the regulator will pull their licence for a period, c) software safety cases will have to be re-done and d) the insurance companies will want to know what all about it.

You can't rush all of the those steps by throwing extra money at them.

And every time there is an accident then the political groups opposed to autonomous cars have another thing to talk about to legislators.

I'm sure there are lots of situations short of full-on, all-weather, all-situation, full-adoption autonomy that are economically valuable which will allow the manufactures to rack up plenty of live testing miles. I just think the process is going to be somewhat slower than optimists are expecting.

[andrewducker]

I'm not convinced that "having heat sensors" is such a big thing that we'd have to refit/retool things. It's a single step in a positive direction, but is it a bigger one than, say, anti-lock brakes? And we didn't insist that the introduction of ABS meant that everyone had to do remedial work. Or pull any licenses - nobody is having their license pulled right now for not having infra-red, it's just a suggested extra safeguard.

[andrewducker]

*Replacement comment for one I had to delete*

That one company which is renowned for cutting corners on safety has failed to operate in a safe manner and then been banned from testing (in Arizona) doesn't surprise me at all.

That doesn't mean that Waymo's self-driving fleet in Arizona is going anywhere, or that the 20,000 Jaguar SUVs Waymo have ordered aren't going to be used. (Or that their application to test fully-driverless cars in California will be refused)

Waymo have over 5 million real world miles under their belt now, without any deaths (and as far as I can tell, only one incident which was their fault). Their safety report is fascinating: https://waymo.com/safety/

Self-driving cars aren't perfect yet, and they're more limited by things like weather than humans are. But outside of those conditions they certainly seem to have reached the kind of level where they're as safe as human drivers. Unless they're designed by Uber.

[danieldwilliam]

Integrating another sensor is not a trivial bit of work. It's not huge in the grand scheme of things but I think the tolerance of it not quite working is going to be low.

You've got a whole bunch of testing to do and you have to work out what the autonomy engine's arbiter is doing with a new set of data. How is integrating a new data type in to its world model? Where do the boundaries of ambiguity lie, when the machine is not sure if that video image, thermal image and lidar image is a kangaroo or a person in a hat.

And for sure, maybe you don't need a thermal camera. Maybe you just need to have software built by some careful and reputable and firm. Or maybe you absolutely do need a thermal camera and you need a different thermal camera for when it is raining heavily and the heat signature of objects changes and you also need a new set of machine vision algorithms to cope with snow and you need to fix the bit of software that says "if you have hit something, reverse" to say, "unless you think you have run over a human being, in which case, don't reverse" and a dozen other things that might be trivial or might not be.

And it's all political too - there are a lot of people with a strong cash incentive to slow autonomy down. Everytime there is an accident there will be lots of people demanding extra testing, extra sensors, extra studies, extra licencing requirements, extra insurances, extra time spent understanding the issues.

Humans have accidents, machines malfunction.