Tuesday 25 August 2015

A problem with driverless cars - jaywalking

Once pedestrians realise that driverless cars are able and willing (or at least programmed to) stop if the pedestrian walks out into traffic, there will be no incentive for pedestrians to walk to designated crossings to cross the road: they'll just simply step out in front of the driverless cars.

This is not so much a problem in the US where crossing other than at intersections is not permitted (and hence there are other disincentives to cross), but in much of the world no such regulation exists.

It is likely that this behaviour will not result in injury given the sophistication of the driverless car technology. However it is likely to result in traffic and a decrease in efficiency of the road network. As such it might be expected that crossing laws outside the US begin to follow the US model.

It is  notable that this behaviour is not likely to emerge until all cars on the road are driverless, which will likely take some time.

Saturday 15 August 2015

Temporary (inflatable?) insulation

In my (admittedly somewhat limited) experience buildings are designed and built for permanence with the features of the building remaining static throughout the year, despite the variation in the seasons. It occurred to me that it should be fairly trivial, and possibly quite beneficial to have (possibly automated) changes to the features of buildings to address seasonal changes.

One example I have given some thought to is the deployment of additional insulation to the exterior of the building to address cold winter conditions. A tent-like structure could be deployed around the building to trap (albeit not particularly tightly) another layer of air. It could be a large tent that sits entirely around the building, or the tent material would be fixed to the building rolled-up like a swimming pool cover for quick deployment. Possibly a full tent would not be required - an extra layer of roof could be unrolled over the existing roof, again from a swimming-pool-cover-like roll, unrolling into slots down the other sides of the roof.

Another possibility for insulation would be something inflatable, like a bouncy castle. The advantage of this is that the air in an inflatable structure is fully contained, providing better insulation. The inflatable structure could be fixed to the building such that it doesn't blow away in storms.

For buildings with plenty of land around them, an interesting solution to extra insulation would be to put a geodesic glass dome all around the building. But it occurs to me that with advances in robotics, it may be possible to create a self-assembling geodesic dome (perhaps using plastic cells, like the Eden Project). Of course self-assembling geodesic domes with plastic cells would also be very useful for this.

Centralization with localism's advantage of experimentation


One of the advantages of localism that is frequently cited is that the diversity of local approaches to a given problem and the relative ease with which new approaches can be adopted provide more opportunity for better solutions to emerge. It provides opportunity for experimentation and innovation, and is analogous to how the Cambrian Explosion resulted in the selection for better-adapted morphologies, or how a shotgun strategy (or throw stuff at the wall and see what sticks strategy) in business allows identification of niches and consumer preferences.

However, a problem that can occur with localism is that local approaches remain entrenched despite inferiority. Or put another way, once a superior approach has been identified in one locality, it is not implemented in other localities (whether such implementation is a push from a centralized authoritiy, or a pull from other localities copying).

So how can the benefit of experimentation be implemented within a centralized structure? Whilst generally working to the (hopefully optimal) centralised approach, local administrative structures would be able to apply for autonomy on a specific problem in order to test a defined alternative approach against a robust set of criteria. As a result of the central control, pushing out a new approach once an experiment has been concluded is relatively easy - the homogeneity of localities means that a single method of implementation is required for the new approach.

Monday 10 August 2015

If we had gone - route search

When driving have you ever wondered whether you would've been better off by going a different route? The data now exists to answer that question: real-time connected car GPS. All that's needed is a service that provides the ability to query that data.

Even when you're following the best route as indicated by your GPS, it's only a prediction based on current road conditions. It's not a retrospective that definitively answers the question of whether it would've been quicker to go another route.

The service would be an add-on to existing GPS services that would tell the user at the end of the journey whether the route they chose was the quickest.

The system would work by piecing together the data from other cars that go segments of the alternative route.

Sunday 9 August 2015

Storage is cheap, just record everything I consume (on the web)

In the long-run, once the technology catches up with the visionaries, I expect lifelogging technology to seamlessly be able to record a person's entire life into a personal database that is easily query-able. I'd be able to ask the lifelog (via my intelligent personal assistant) to remind me of every occasion that my friend Dave has said something funny. And the technology would be able to identify instances of my laughter that were antecedent to something Dave said (or perhaps an even more sophisticated understanding that Dave said something funny).

One of the things that attracts me most to the idea of lifelogging is being able to deal with those circumstances where you remember reading something, but don't remember the exact details, and don't remember where you read it. I can be surprisingly hard to find the original article with search engines, depending on the vagueness of your memory. And there's a great risk, with web content, that the content may have just disappeared.

One of the tools I use to record articles to address the "I've read something about this, but can't remember where" issue is Evernote. It's pretty trivial to clip a web page and tag it. But doing so requires action. It requires you to think "yes this content was good, I might need to refer to it later". Having to act takes time, and you might not always remember to do it. And further, your assessment of what's going to be useful isn't always that accurate.

So, given that storage is so cheap, should we now be auto-clipping: saving the full text and image (and video?) of every web page that is visited?

Possible problems with such an approach:

  • It might actually take more storage than I think, and cost more
  • It's incredibly wasteful of storage, unless the clipping service stores a single version, linked to multiple users
  • It may well breach copyright
  • Users may visit some websites that they don't want a record of them having visited

Natural language news quiz

One of the brain training apps that I've been playing around with requires that you read some text, and then answer questions based on it. It's testing (and hopefully improving) comprehension and memory.

But why can't I do the same with an article I actually want to read, rather than some pre-defined text for the brain training app?

How would such a system work? I envisage a browser plugin or RSS reader plugin that monitors the news articles (and other web pages) you're reading, then uses Natural Language Processing technology to extract from the article information that it can quiz the user on. Then, some time after the article has been read, the plugin pops-up a quiz for the user to test their comprehension and retention.

The user's quiz performance is tracked over time to see if quizzing improves comprehension and retention.

Obviously there's an assumption that the user is going to read the whole content of everything they open, which is definitely not a valid assumption. However, a sophisticated app may be able assess what the user has read, using the following techniques:

  • The extent to which the user has scrolled down in the page / the amount of text in the visible view
  • The amount of time spent on a page
  • Eye tracking with user-facing camera
The user would be able to customise the service in the following ways:
  • Frequency of quizzes
  • Blocking specific websites / domains from being quizzed on