• Public transport with masks is not a COVID super-spreader (TorontoStar)
• A window of opportunity for road user charging (CentreForLondon)
• 4 storey London underground car park to shopping conversion (IanVisits)
• What links Florence Nightingale & Isambard Kingdom Brunel? (RailInsider)
• What is a tramway really worth? (Trams&UrbanTransit)
• Unofficial 2020 map animation update on LA’s 15 metro projects (CoMotion)
• AV racism: Driverless cars may not detect darker skin pigments (StreetsBlog)
Read our most popular articles:
- Schrodinger’s Cab Firm: Uber’s Existential Crisis
- You Hacked – Cyber-security and the railways
- On Our Line Podcast #8: Talking Uber, Lyft and Mobility disruption
And some of our other sections:
Feel we should read something or include in a future list? Email us at [email protected].
Reconnections is funded largely by its community. Like what we do? Buy us a cup of coffee or visit our shop.
That Nightingale/Brunel story needs much wider circulation.
Brunel’s hospital design reminds me inescapably of his GWR goods sheds.
I hope if road pricing is implemented, it is much more sophisticated than a fixed distance rate, which was mentioned at the end of the CentreForLondon article. Such a charge would be as indiscriminate in its effect as a fuel tax, more regressive, and harder to collect. If road user charging is to justify its high collection costs, and other difficulties, it has to have some kind of context sensitivity to do a lot better than that.
I thought the issue on machine learning that is raised by StreetsBlog was well known. This issue is that when “training” a machine learning program, it is suboptimal to give it the data in the frequencies it occurs in everyday life. If you do that, the computer fails to gain the “wider intuitions” that humans have from their ability to generalise from common situations to rarer situations. Rather computers have to be given larger training on rarer situations, else they will misjudge them in a way humans would not. I came across this almost 20 years ago as computers were learning to play backgammon, and very rapidly overtaking them. What made a good program was not just the code, but also the data used for the “training” process. Players learned that they could beat early strong programs by leading them into certain weird positions that would almost never occur in normal play. Human players of even moderate ability could see, on basis of general playing experience, that these were terrible positions, and would stay well away from them. But the computer did not encounter these weird positions in their training data, and did not know to avoid them. So the computers needed to be explicitly trained on the weird positions also, to match human intuition that these were terrible positions, and stay well away from them.
Ivan: Although a flat rate was mentioned in the article, it was closely preceded by a paragraph suggesting differently (including the words: “Our report envisaged a system of dynamic charging that would vary depending on road conditions and alternative options”). The flat rate mention was, I thought, just put in to indicate what order of magnitude of charge would be required to approach funding of TfL’s deficit.
Fascinating report about backgammon, though!