Autonomous crash "needs to be put in context"
Forbes discusses the first fatal accident involving a Cooooooooooooolmobile operating in Cooooooooool mode. Hey, no problemo, dude! You just need to put it in context. Cooooooool itself tells us: This is the first fatality in 130 million miles of Cooooooool driving. Unperson cars have a fatality every 94 million miles.

As fucking usual, misusing stats. When you have a single incident in a small sample, you can't compare it with anything. Especially when the small sample involves people who are NOT regular or typical, people who drive in special places. (When you can turn Coooooool mode on and off, you're going to turn it off in tricky situations.) But EVEN SKIPPING THAT OBVIOUS FACT, 1/130 is well within the range of possibility for an event expected every 94 times.

Let's REALLY put it in context, using 'autonomous' software. I cobbled a quick Python program that generates a series of trials representing a series of cars. Each car has a prechosen probability of 94 miles before its first accident. The program puts stars by each trial where the accident happened after 94 miles. This is wildly simplified from reality, but I'm only trying to show that hitting the first event at 130 is EXTREMELY ORDINARY when an event is expected to happen every 94 times.

Results of 100 trials:

= = = = =

100 trials with probability 1/94

Car 0000 accident at 144 ***
Car 0001 accident at 166 ***
Car 0002 accident at 9
Car 0003 accident at 61
Car 0004 accident at 44
Car 0005 accident at 342 ***
Car 0006 accident at 225 ***
Car 0007 accident at 145 ***
Car 0008 accident at 45
Car 0009 accident at 74
Car 0010 accident at 122 ***
Car 0011 accident at 37
Car 0012 accident at 85
Car 0013 accident at 56
Car 0014 accident at 38
Car 0015 accident at 13
Car 0016 accident at 109 ***
Car 0017 accident at 187 ***
Car 0018 accident at 60
Car 0019 accident at 69
Car 0020 accident at 430 ***
Car 0021 accident at 110 ***
Car 0022 accident at 63
Car 0023 accident at 16
Car 0024 accident at 108 ***
Car 0025 accident at 63
Car 0026 accident at 30
Car 0027 accident at 9
Car 0028 accident at 89
Car 0029 accident at 29
Car 0030 accident at 5
Car 0031 accident at 15
Car 0032 accident at 26
Car 0033 accident at 304 ***
Car 0034 accident at 58
Car 0035 accident at 87
Car 0036 accident at 15
Car 0037 accident at 134 ***
Car 0038 accident at 69
Car 0039 accident at 192 ***
Car 0040 accident at 161 ***
Car 0041 accident at 128 ***
Car 0042 accident at 92
Car 0043 accident at 87
Car 0044 accident at 1
Car 0045 accident at 135 ***
Car 0046 accident at 94
Car 0047 accident at 81
Car 0048 accident at 187 ***
Car 0049 accident at 102 ***
Car 0050 accident at 265 ***
Car 0051 accident at 371 ***
Car 0052 accident at 15
Car 0053 accident at 4
Car 0054 accident at 135 ***
Car 0055 accident at 28
Car 0056 accident at 68
Car 0057 accident at 113 ***
Car 0058 accident at 108 ***
Car 0059 accident at 264 ***
Car 0060 accident at 24
Car 0061 accident at 167 ***
Car 0062 accident at 172 ***
Car 0063 accident at 165 ***
Car 0064 accident at 78
Car 0065 accident at 127 ***
Car 0066 accident at 141 ***
Car 0067 accident at 186 ***
Car 0068 accident at 2
Car 0069 accident at 148 ***
Car 0070 accident at 57
Car 0071 accident at 167 ***
Car 0072 accident at 44
Car 0073 accident at 125 ***
Car 0074 accident at 105 ***
Car 0075 accident at 114 ***
Car 0076 accident at 94
Car 0077 accident at 4
Car 0078 accident at 218 ***
Car 0079 accident at 171 ***
Car 0080 accident at 104 ***
Car 0081 accident at 74
Car 0082 accident at 164 ***
Car 0083 accident at 58
Car 0084 accident at 6
Car 0085 accident at 229 ***
Car 0086 accident at 114 ***
Car 0087 accident at 63
Car 0088 accident at 31
Car 0089 accident at 138 ***
Car 0090 accident at 13
Car 0091 accident at 2
Car 0092 accident at 132 ***
Car 0093 accident at 55
Car 0094 accident at 59
Car 0095 accident at 88
Car 0096 accident at 110 ***
Car 0097 accident at 77
Car 0098 accident at 15
Car 0099 accident at 69


= = = = =

In this particular run of 100 cars, 45 of the cars had their accidents AFTER 94 miles. As it happens, the very first one was over 94, and its accident was at 144 miles. Not too far from 130, though that won't happen on the next run of 100 trials. (On the next run, the first car had its accident at 1 mile!)

Would any 'reputable source' ask us to put the Pinto's tank fires in context? NOT ON YOUR MOTHERFUCKING LIFE. They would skew the stats the other way because Ford is MAXIMALLY UNCOOL for obvious fucking reasons.

= = = = =

Other articles quote a different 94, by a nice coincidence: "Human error is responsible for 94% of collisions." Again missing the point. How many crashes are PREVENTED by an ALERT human, when faced with a drunk or texting or Harry Potter-watching (Jesus H Fucking Christ again!) human?

Later: Coooool Inc is blaming the dead driver for "not being alert". Say, I've got an idea. Since autopilot requires you to be alert, maybe we could make the Cooool Experience even Cooooler by adding a control device to occupy your muscles and help to keep you alert. Perhaps a lever or big knob to control the steering, and levers or foot-activated modulators to vary the speed of the Coooool Auto. This would turn the autonomous driving experience into VR! I know it's radical, but wouldn't it be oh so Coooooooool?

Labels: Asked and answered, Blinded by Stats, Carbon Cult

  ¶ 2:14 AM