Wednesday, September 30, 2015

Brief foray into politics

I have another blog post queued up, but with today's events this "idea" is worth reiterating: end the death penalty.

I have no interest in, nor patience with, moralistic arguments about the death penalty. If we lived in a world in which you could know something with 100% certainty, then those arguments would be useful. But we don't. Every single source of evidence used in the judicial system has an error rate. Some sources (such as witness testimony) have alarmingly high error rates. There's no combination of error rates that, when compounded, yield 100% certainty.

The death penalty might be worth debating if we lived in a world of absolute certainty. We don't. Not even remotely close, in fact. Keeping an innocent person in jail isn't a great solution either but at least the possibility for some kind of restitution exists in the (again, remarkably likely) event that the evidence is flawed.

Update: The governor stayed the execution at the last moment.

Update 2: Looks like even DNA evidence can have high error rates.

Tuesday, September 29, 2015

Your tax dollars at work

These kinds of signs are commonplace while work is in progress. But after the building is up, street is fixed, school is updated, etc., the link back to proposition, amendment, or tax increase that funded the work is lost. I think this eventually leads people to question why they are paying such high rates and fees. Even though their everyday environment is constructed through a series of taxpayer-sponsored improvements, they don't see direct evidence of it.

So how about an AR app that shows a rendering of how something would have looked if it were not for public goods? This could be extended to other improvements that are less visible. For example, an app could point out reductions in particulate matter in the air or water, reductions in crime in an area that is linked to housing or health initiatives, or even reminders of why something hasn't been fixed. This could also be useful to remind folks about the extensive government funding behind everyday devices.

Monday, September 28, 2015

Tiny multimedia offices for all

I am a fan of introvert advocacy. In particular, I think that workplaces have placed too much emphasis lately on constant collaboration and not enough on deliberation. When most studies show that companies come up with better ideas when brainstorming is asynchronous, this is particularly worrisome. Furthermore, many workers in open office environments complain about noise pollution, making even simple tasks daunting.

Another large trend in many workplaces is remote working. Obviously, remote working can represent the furthest extreme of the individualist workstyle, which is of course not optimal at all times. Some casual communication and collaboration is surely necessary.

What if instead of offices companies provided workers tiny backyard (or garage) offices outfitted with basic communication technologies? Let's say each office has a four-state switch: off, focused, asynchronous, and synchronous. The "focused" case just notifies others (e.g., your boss) that you are working but are doing something that requires your full attention and shouldn't be bothered. In the asynchronous setting, you can post your work to a "wall" and see others posts as well, allowing you to maintain a lightweight awareness of what others are doing but not demanding too much attention. Finally, the synchronous setting would stream your audio and video, using webcams embedded in the office, to other sites. Stereos on the walls could also provide some "chatter" to simulate the water-cooler effect.

If this approach replaced offices, and all the commuting that goes with them, it would probably be cost effective. Also, having your own dedicated space would allow you to keep paraphernalia around for longer projects, something that is difficult to do in open office environments.

Sunday, September 27, 2015

Live streaming from drones for amateur astronomy

As I mentioned in the last post, I am not a huge fan of VR, but I could see live streaming panoramic video to VR headsets.

In honor of the lunar eclipse, it would be useful to stream video data from a drone calibrated for night time astronomical observations. The goal is for the drone to fly high enough above diffusive layers to give people a clearer view of relatively dim astronomical events. Perhaps live feature detection (e.g., meteor detection) would be useful as well for slightly delayed, near synchronous viewing. This would at least reduce neck and back strains for longer events. You could also program your drone to record an event over night and watch the "highlights" in the morning in a video captured and created automatically.

Saturday, September 26, 2015

Yesterday's tomorrows twitter robo account

OK, this one is snarky and simplistic. But over the years I've heard about so many tech trends that are "the next big thing" for so long. The latest is VR, which has been "the future of technology" for nearly all of the 15 years I've been active in computer science research.

So how about a Twitter account that quotes tweets about the "next big thing" and adds a citation of the first known time it got that moniker. #throwsomecoldwateronit

(This is inspired by a paper about a similar phenomenon in ubiquitous computing.)

Friday, September 25, 2015

Annotating longform docs on-the-go

I think wearable devices will see widespread adoption only when they can be operated away from phones. This is why I like the idea of wireless earbuds with onboard memory (such as the Bragi Dash). These devices would allow you to load up music or podcasts for a run or hike without having to worry about taking your phone with you (or hassling with wires or brittle Bluetooth connections). I actually never use music while I am on the trails (listening to music can be dangerous as it diminishes situational awareness), but I could see using them for podcasts when hiking through environments that don't hold my interest as well (like cities).

In fact, in that scenario, I would like to add a few features: text-to-speech, document layout hint injections, and annotations. The first two features are derived from SeeReader and are designed to allow one to convert a longform (written) piece into an audio document. The layout hint injections just mean that the system would read out not only the body text of the article, but also note when there is a figure that might be interesting. Obviously you wouldn't be able to look at it at the time, but in combination with an audio annotation feature you could "mark" parts of the document that you want to go back to later. So for example, the text might talk about the growth of fracking in northern CO and reference a map in the document that shows the appearance of drilling sites over time. Saying "mark" could create an annotation to that part of the document so I could check out the map when I'm back from my hike.

Thursday, September 24, 2015

Augmented equipment

The 94Fifty ball has been dismissed in some press as a trivial geek toy, but I think it might actually have some utility for young players trying to improve their game. The broader goal of augmenting sports to improve practice and training is something Jono Hey and I explored in our work on ping pong training.

Still, when it comes to adding sensors to sports equipment, I am more interested in improving either the game itself or fans enjoyment of the game. For example, video review during games is something that makes sense rationally but tends to slow things down and interrupt the flow of games. It would be better if the equipment itself was smart enough to detect automatically calls that currently go to review. Instrumented basketballs should be able to detect precisely when they were last touched, and with the addition of player-worn galvanic transponders, we should be able to detect right away who was the last person to touch the ball. This would eliminate all possession reviews as well as some goaltending reviews. The rest of the goaltending reviews could be cleaned up with an arc detector in the ball. Foot-behind-the-line detection (for 3 pointers and less regularly for free throws) should also be straightforward either with cameras or sensors embedded in shoes when there are visual obstructions.

Joey Crawford has his moments. But come on, nobody goes to games to see the refs. More ball less stall.

Wednesday, September 23, 2015

Better backcountry comm tools

When I am running a backcountry trail or bushwacking off-trail to some big trees (or, in New Mexico, bushwacking while supposedly on a trail), I tend to keep my phone with me, mostly to take advantage of off-line mapping tools. There are many mobile and smartwatch applications that track GPS position, mostly for logging purposes. These are cute, but the functions they provide are relatively unimportant. Consistent networking is (unfortunately, sometimes literally) the killer app for wilderness trekking. There are many satellite systems for SOS beaconing, but the goal is really to send and receive messages before you get into trouble (e.g., if you are canyoneering you'll want to know if there's been rain recently anywhere upstream).

Luckily, there do exist some 2-way satellite systems, such as the DeLorme inReach or the Spot Gen3. I think these kinds of systems could be improved incrementally to be more useful. In particular, this system seems like a perfect fit for push notifications to a smartwatch. It would be nice to be able to configure a service to listen to a particular Twitter stream (e.g., an NWS local) and push important notifications to your device and then onto your watch. It would be even better if the server service would have some basic filtering tools so that you could send only tweets that include "warning" or some other keyword. Even better would be if the NWS has a codified tweet protocol, but as far as I can tell they don't at the moment. This data could also be gleaned via regular web page scrapes from NWS local websites.

Tuesday, September 22, 2015

Make IoT data fun

The Internet of Things has the potential to overwhelm us with notifications. We are all already bombarded with a variety of usually minor, inconsequential notifications from apps on our phones. Once everyday devices in our homes, offices, cars, and elsewhere begin streaming sensor data to the net this problem could become amplified by orders of magnitude.

Solutions proposed for the coming tidal wave of sensor data often take the form of different types of filters: from simple rules governing what types of notifications should be forwarded on to the user to more complex agents that learn patterns and attempt to detect the importance of notifications. Other solutions rely on displays (e.g., peripheral displays) to tailor the intensity of the notification to the importance of its information.

I propose that outside of absolute catastrophes (e.g., your house is on fire), notifications are simply the wrong model because, in short, people don't like them. There have been many studies of notifications, and most show that users ignore almost all of them (one study showed that they interact with just 6% of push notifications).

A better model for IoT should be apps that people actually like. I would design an IoT app for "down times" that people currently fill with games and social media. When you launch your "Home" app on your phone, instead of seeing a laundry list of notifications to deal with, just show cards. Maybe one card is about water, another power, another food, etc. Swiping away the card dismisses it and sends a small signal to the machine learning system in your app that that topic is less important to you. Tapping on a card opens up more details, or another stack of cards specific to that topic. The app could also learns not only what is or is not important to you, but in some cases also what a good or bad reading is. For competitive people, you could enter local competitions with your neighbors (e.g., who is using the highest % of renewable energy this week). Or, more importantly, show how much money you're saving, or could save, with various changes to your habits. I would show these monetary savings through interactive graphs, a la Brett Victor's explorable explanations.

Monday, September 21, 2015

Media capture feedback

At FXPAL we developed a system, ReBoard, for automatically capturing whiteboard content using a nearby camera pointed at the board. We also built several tools in the space of intelligent capture, including an app that would give you live feedback about the quality of captured documents.

For some reason, until now it hadn't occurred to me that combining the two ideas would be quite useful. Captures in the ReBoard system can be low quality for all kinds of reasons: someone is blocking the board, the light is too low, shadows are bad, etc. ReBoard has built-in mechanisms to deal with these as well as it can, but surely it would also be useful to add a post hoc step to evaluate the quality of each shot. This could be fed back to users in a small display near the board or on a back-end web UI (in the initial ReBoard system we used a Chumby for the in situ display).

This concept could be extended now that external capture (e.g., from drones) is becoming commonplace. Imagine giving a drone a high level command to capture high-quality media from an event. You could even specify certain requirements (the presence or absence of people, for example, or a mixture of different kinds of shots). The drone could capture media as well as it could then run this post hoc process to get near real time feedback about the quality of recent shots. If any are unusable, they would be culled and requeued for the drone.

By the way, I recently built a new version of ReBoard that works with a confederation of GoPro cameras. Email me if you're interested (carter at!

Sunday, September 20, 2015

The Penny Race

I used to do a lot of trail running races. I did decently well, finishing in the top 3 in my age group a few times. During a half marathon in Marin County I crested the final hill and was about to descend through an lovely second-growth redwood and fir forest. I was again doing well enough, well inside the first ten finishers. I was exhausted and in some pain, but thought I could probably soldier to the finish and drop only a spot or two. But then I looked around; early morning fog had settled into the grove in front of me, the giant evergreens morphing into ethereal specters, a familiar but wondersome transformation I have grown accustomed to after running through hundreds of redwood groves.

And my thought was this: I am in the midst of the best place I will experience in weeks and my goal at the moment is to get through it as fast as possible?

After that I took a left turn, away from the race and along a creek, jogging slowly. I've not done a trail race since.

Perhaps we should reconsider what it means to race, to compete. Races are easy things, simple to grasp because they are so easily measured. But what if we went after a different, more complicated metric? Something like flickr's "interestingness". Perhaps Dillard puts it better:

The world is fairly studded and strewn with pennies cast broadside by a generous hand. But- and this is the point- who gets excited by a mere penny? But if you cultivate a healthy poverty and simplicity, so that finding a penny will literally make your day, then, since the world is in fact planted in pennies, you have with your poverty bought a lifetime of days.

The Penny Race. Teams of people bring their mobiles and set off to capture photos and videos of interesting things within a given period of time. Explore as far or as close as you wish, but return to the start by a certain time to show off your compiled multimedia document. Teams vote on a winner: the best is treated to a round of frothy ales.

Saturday, September 19, 2015

Reverse tolls for bikes

Bikes have many advantages over cars; they are:

  • Better for the environment
  • Better for your physical health
  • Better for your mental health
  • Better for your wallet

Yet while commuting by bike is growing in the US, it still represents only 1% of all total commutes. Why do cars dominate? They are so widespread mostly because of an almost total lack of taxation on market externalities (and non-existant regulation at dawn of the 20th century. For more on that read, "Crabgrass Frontier", though be warned it can be incredibly depressing). While of course some state and federal taxes exist, they come nowhere near the actual cost of cars (which, let's not forget, might be the main cause of the destruction of the Earth as we know it if the worst-case scenarios for global warming come to pass).

Electric cars are sometimes seen as a remedy, but this is unlikely since they are plugged into a grid that is majority coal and natural gas (which, because of fracking, is as bad for the environment as coal).

A clean grid might help, but clean fuels are unlikely ever to cover the current total usage. Furthermore there is the immense construction costs of the cars (and infrastructure). Finally, because of Jevons paradox, it is likely that any increase in efficiency in self-driving cars will be offset by an increase in use.

The only efficient solution is to reduce use, which we can do with bikes (along with all of the other benefits noted above).

So how do we get more people to ride bikes? A recent episode of Real Sports on HBO documented the rise of a biking culture in Copenhagen. The solutions there, building out more infrastructure key among them, are certainly worth considering. On the other hand, many startups in the US try to encourage good behavior by giving people stars or ratings or other extrinsic motivators. I think those are mostly childish nonsense. Give people cash. Increase taxes on cars (slightly, so there aren't riots in the streets) and apply a reverse toll for bikers (pass a certain point on your bike twice a day during commute hours and you get a check).

It turns out that a Norwegian town tried this out, but only very briefly. To make it really work you need widespread sensor/reader deployments (funded by the aforementioned auto tax hike).

It's not much, but the only way to avoid utterly destroying the planet is to reduce consumption.

Update: The VW scandal provides more evidence that cars are never likely to be truly eco-friendly.

Update 2: Good job, Google.

Friday, September 18, 2015

At least 1 idea/day. Here we go.

I am going to try to get down one idea per day using this blog. "One idea per day" comes from the Seinfeld "don't break the chain" productivity approach.

In the spirit of this effort, the first idea is a blog extension that automatically creates a "chain" visualization. There exist many apps for creating such a calendar generically, but it is always better to have the visualization grow automatically if possible. This could be useful for any event that can be sensed. Apple Watch does something similar with "Activity history", showing a calendar of activity events.

What if you could just attach a sensor to something that you wanted to make sure you did everyday? Let's say I want to learn an instrument. I attach a small sensor to it that is tuned to detect when it is played. I use a mobile app to set my goal every day for a particular sensor (e.g., "play for 20 minutes") then attach it. This would automatically create a "chain" for me on the app.

The problem is, of course, that practice does not make perfect. Perfect practice makes perfect. So how can you tell if you're doing well? Perhaps the sensor has a mic that pics up audio and has some basic analysis abilities to detect quality. If those don't work well enough, clips could be fired off to experts (that is, crowdsource the problem).

Could such a basic sensor be used to track other things in this way? Attach a motion sensor to a bike, to shoes, hats, etc. In this way you could solve the problem with many activity trackers: they fail to track anything that doesn't look like a step. Just attach a sensor button to things (your swim trunks, kayak paddle, etc.) and you'll get at least some data for every activity.