What happens to you if the self-driving vehicle you are in gets into an accident? And what can insurance do for you?

One of the questions that always interested me regarding self-driving vehicles is how the insurance bit would play out. Apparently, in the trial for Uber self driving cars in Pittsburgh, passengers are likely unknowingly, waiving any rights to compensation. So the answer to my question is: you are on your own. And I personally thing that this is a gap insurance companies should fill.

(https://www.theguardian.com/technology/2016/sep/26/uber-self-driving-passengers-pittsburgh-injury-death-waiver)

Insurance for cars today is simple, simple enough that you can get a quote online based on a few questions. These questions are based around the factors that are thought to be important in determining your risk of getting into an accident, and the premium you pay depends on that risk.

Makes sense, right? People with higher risk pay more.

The science behind the premium you pay for your insurance is well established. A quick look at the questions that help gauge your premium shows the main drivers of risk to be demographics, past driving history, experience and track record, vehicle characteristics:

The question is: how do you estimate the risk of a self-driving car? Who is responsible in case of accident, the vehicle manufacturer, the sensor manufacturer, the company that provides the algorithms that translate sensor data to how the car behaves? Having a self-driving car from a ride-sharing company adds an extra layer of complexity.

Tesla has taken a different route insisting that, even though you might want to watch Harry Potter in your model S, you shouldn’t: (https://www.theguardian.com/technology/2016/jul/01/tesla-driver-killed-autopilot-self-driving-car-harry-potter).

The car makers did say that the autopilot that was likely engaged at the time of the crash is not a substitute for drivers paying attention, but more of a driving aid.

In fact, in May Tesla ordered updates https://www.theguardian.com/technology/2016/sep/11/tesla-self-driving-cars-autopilot-update because it was felt drivers were feeling too confident and not paying attention enough despite the crash in May. https://www.theguardian.com/technology/2016/jun/30/tesla-autopilot-death-self-driving-car-elon-musk

Other companies like Uber who do not produce the cars, seem to take a different approach.

Uber’s cars in Pittsburgh (just like nuTonomy’s in Singapore (http://www.theonlinecitizen.com/2016/08/25/singapore-to-have-the-worlds-first-self-driving-taxi-service-starting-from-one-north/)) have a driver who can take back control if need be. (http://www.bloomberg.com/news/features/2016-08-18/uber-s-first-self-driving-fleet-arrives-in-pittsburgh-this-month-is06r7on). Does that make the driver responsible?

In fact, Grab has just signed a collaboration with nuTonomy in Singapore (http://news.asiaone.com/news/singapore/uber-rival-grab-partners-driverless-car-firm-nutonomy-singapore#cxrecs_s). Apparently, outside the limits of the One North, the driver will be responsible.

I am not sure about the case in Singapore, but according to the Guardian, users of the Uber self-driving service in Pittsburgh have waived their rights to compensation. The driver is not responsible, and hence you cannot be covered under commercial usage terms. This might be something potential clients of the Grab service in Singapore might want to enquire about.

In any case, theoretically, how could the premium be calculated?

Machines are trained based on data. Data is obtained from recording real life experiences – let’s ignore simulations for the moment, since simulations are just another layer of the same thing. For example, I am quite sure nuTonomy will be capturing data from the Grab vehicles that will be plying the Singapore roads outside One North, hence under driver’s control.

It might be a good idea to make all the training data available to insurers so they can understand whose behaviour the machine learnt from. Also oversampling should be made clear. Then the insurers might have an idea of which type of profile the machine would be equivalent to. 

Transparency is key.

In a few years, you might be cursing at the self-driving vehicle you had a close shave with “You drive like a 55 year old ah pek!” in Singapore or “You drive like a college educated 40 years old married white man!” in the USA

(https://www.entrepreneur.com/article/242096)

Post Script:

I know the availability of training data is a very basic first step; there are more complications, since sensors might fail

https://www.theguardian.com/technology/2016/sep/14/tesla-fatal-crash-china-autopilot-gao-yaning

or be pushed too far:

https://www.theguardian.com/technology/2016/sep/15/autopilot-supplier-disowns-tesla-for-pushing-the-envelope-on-safety

or the machine still have more to learn:

https://www.wired.com/2016/02/googles-self-driving-car-may-caused-first-crash/

https://www.dmv.ca.gov/portal/wcm/connect/3946fbb8-e04e-4d52-8f80-b33948df34b2/Google+Auto+LLC+02.14.16.pdf?MOD=AJPERES

and be careful, humans learn too.

 

How the Manhattan Bombings could have been prevented with little intrusion into people’s lives

I value my privacy, and I don’t buy the argument that “it’s ok if you have nothing to hide”. But I am surprised how the alleged Manhattan bomber was not stopped earlier. Hindsight is 20/20, but simple use of analytics (yes, "Big Data", but just metadata) could have prevented this tragedy.

2 critical pieces of information came to light AFTER the bombing:

1.  The police was warned about the suspect TWICE since 2014
• http://www.nytimes.com/2016/09/22/nyregion/new-york-new-jersey-bombing.html?_r=0 
2.  He bought most components he used in his bomb off eBay.
•  http://abcnews.go.com/US/nyc-explosion-buying-bomb-parts-ebay/story?id=42243357

I am quite sure that the authorities have a clear idea of the raw materials that can be used to make bombs. Sure most of these are at least dual-use.

Citric acid is a well known preservative, but if you are not engaging in large scale food production, then it is unlikely that you’d need industrial volumes of citric acid.

Similarly, ball bearings are very commonly used in all sorts of equipment; if you are sitting on a swivel chair while reading this post, it is likely that you have ball bearings right under you. Again, if you are not engaged in manufacturing or repair of equipment with moveable parts, then you are unlikely to need ball bearings on a large scale.

Now when you combine citric acid, ball bearings and other ingredients, you end up with a potential recipe for a bomb.

What I am saying is:

1. The combination of individual dual-use items purchased over time should be enough to trigger an alarm, and this could be done by eBay itself.
2. Adding context is important (and this is where many “Data Science” endeavours fail since they do not incorporate “domain knowledge”), since these items are dual use, and there are more avenues for purchase on top of eBay. Hence, someone not engaged in food preservation ordering large amounts of citric acid should trigger a flag. Then a deeper search can be conducted.
3. A deeper search can be conducted once flags are raised, either by the combination of purchases, or by contextually strange purchases. What is key here is that only ‘metadata’ is required to paint the picture.

For example, the credit cards/bank cards/account details attached to the online ordering platforms can be tracked to the individual and transactions on all his/her cards/accounts checked for confirmation of the suspicious patterns. 

Furthermore, basic Call Data Records of phones associated with the individual can be accessed to show possible alternate locations such that more than just the residential address is included in the deeper search. And in case deeper analysis still is required, simple SNA run on the individual and his/her associates and their locations also included in the deeper dive.

All this is quite easily done if you have access to the metadata.

But you don’t need the metadata of everyone to be pulled out and analysed, only those flagged by their purchases, online behaviour, or as in this case, by people who suspected them.

And that is the most glaring part. You can argue that not all bomb recipes are known, and some purchases of dual use items might escape notice, and even combination of these or out of context purchases (after all, a farmer might decide to use fertiliser for bomb making, hence context is not exculpatory by itself). But once an individual is flagged, it would be a good idea to monitor such simple clues to what he/she might be up to, and avoid tragedies.