That moment when you realise, to your immense surprise, that the research you’ve spent most of the year on might actually demonstrate the thing you set out to test after all. 😲
Screw you, null hypothesis.
Parachute use did not reduce death or major traumatic injury when jumping from aircraft in the first randomized evaluation of this intervention.
However, the trial was only able to enroll participants on small stationary aircraft on the ground, suggesting cautious extrapolation to high altitude jumps.
As always, when the BMJ publish a less-serious paper, it’s knock-your-socks-off funny. In this one, a randomised trial to determine whether or not parachutes are effective (compared to a placebo in the form of an empty backpack) at preventing death resulting from falling from an aircraft, when used by untrained participants, didn’t get many volunteer participants (funny, that!) until the experiment was adapted to involve only a leap from a stationary, grounded aircraft with an average jump height of 0.6 metres.
Silly though this paper is, its authors raise a valid point in the blog post accompanying their paper:
That no one would ever jump out of an aeroplane without a parachute has often been used to argue that randomising people to either a potentially life saving medical intervention or a control would be inappropriate, and that the efficacy of such an intervention should be discerned from clinical judgment alone. We disagree, for the most part. We believe that randomisation is critical to evaluating the benefits and harms of the vast majority of modern therapies, most of which are unlikely to be nearly as effective at achieving their end goal as parachutes are at preventing injury among people jumping from aircraft.
However, RCTs are vulnerable to pre-existing beliefs about standard of care, whether or not these beliefs are justified. Our attempts to recruit in-flight passengers to our ambitious trial were first met with quizzical looks and incredulity, predictably followed by a firm, “No, I would not jump without a parachute.” For the majority of the screened population of the PARACHUTE trial, there was no equipoise—parachutes are the prevailing standard of care. And we concur.
But what if we provided assurances that the planes were stationary and on the ground, and that the jump would be just a couple of feet? It was at this point that our study took off. We set out in two groups, one at Katama Airfield on Martha’s Vineyard and the other at the Yankee Air Museum in Ann Arbor. One by one, our study subjects jumped from either a small biplane or a helicopter, randomised to either a backpack equipped with a parachute or a look-a-like control. As promised, both aircraft were parked safely on terra firma. The matchup was, unsurprisingly, a draw, with no injuries in either group. In the first ever RCT of parachutes, the topline conclusion was clear: parachutes did not reduce death or major traumatic injury among people jumping from aircraft.
But topline results from RCTs often fail to reveal the full story. We conducted the PARACHUTE trial to illustrate the perils of interpreting trials outside of context. When strong beliefs about the standard of care exist in the community, often only low risk patients are enrolled in a trial, which can unsalvageably bias the results, akin to jumping from an aircraft without a parachute. Assuming that the findings of such a trial are generalisable to the broader population may produce disastrous consequences.
Using humour to kickstart serious conversations and to provide an alternative way of looking at important research issues is admirable in itself.
Fantastic lightweight introduction to bacteriophages and how they can potentially be our next best weapon against infection as we approach the post-antibiotic age. Plus an interesting look at the history and the discovery of bacteriophages!
I’ve generally been pretty defensive of Microsoft Edge, the default web browser in Windows 10. Unlike its much-mocked predecessor Internet Explorer, Edge is fast, clean, modern, and boasts good standards-compliance: all of the things that Internet Explorer infamously failed at! I was genuinely surprised to see Edge fail to gain a significant market share in its first few years: it seemed to me that everyday Windows users installed other browsers (mostly Chrome, which is causing its own problems) specifically because Internet Explorer was so terrible, and that once their default browser was replaced with something moderately-good this would no longer be the case. But that’s not what’s happened. Maybe it’s because Edge’s branding is too-remiscient of its terrible predecessor or maybe just because Windows users have grown culturally-used to the idea that the first thing they should do on a new PC is download a different browser, but whatever the reason, Edge is neglected. And for the most part, I’ve argued, that’s a shame.
But I’ve changed my tune this week after doing some research that demonstrates that a long-standing security issue of Internet Explorer is alive and well in Edge. This particular issue, billed as a “feature” by Microsoft, is deliberately absent from virtually every other web browser.
About 5 years ago, Steve Gibson observed a special feature of EV (Extended Validation) SSL certificates used on HTTPS websites: that their extra-special “green bar”/company name feature only appears if the root CA (certificate authority) is among the browser’s default trust store for EV certificate signing. That’s a pretty-cool feature! It means that if you’re on a website where you’d expect to see a “green bar”, like Three Rings, PayPal, or HSBC, then if you don’t see the green bar one day it most-likely means that your connection is being intercepted in the kind of way I described earlier this year, and everything you see or send including passwords and credit card numbers could be at risk. This could be malicious software (or nonmalicious software: some antivirus software breaks EV certificates!) or it could be your friendly local network admin’s middlebox (you trust your IT team, right?), but either way: at least you have a chance of noticing, right?
Browsers requiring that the EV certificate be signed by a one of a trusted list of CAs and not allowing that list to be manipulated (short of recompiling the browser from scratch) is a great feature that – were it properly publicised and supported by good user interface design, which it isn’t – would go a long way to protecting web users from unwanted surveillance by network administrators working for their employers, Internet service providers, and governments. Great! Except Internet Explorer went and fucked it up. As Gibson reported, not only does Internet Explorer ignore the rule of not allowing administrators to override the contents of the trusted list but Microsoft even provides a tool to help them do it!
I decided to replicate Gibson’s experiment to confirm his results with today’s browsers: I was also interested to see whether Edge had resolved this problem in Internet Explorer. My full code and configuration can be found here. As is doubtless clear from the title of this post and the screenshot above, Edge failed the test: it exhibits exactly the same troubling behaviour as Internet Explorer.
I shan’t for a moment pretend that our current certification model isn’t without it’s problems – it’s deeply flawed; more on that in a future post – but that doesn’t give anybody an excuse to get away with making it worse. When it became apparent that Internet Explorer was affected by the “feature” described above, we all collectively rolled our eyes because we didn’t expect better of everybody’s least-favourite web browser. But for Edge to inherit this deliberate-fault, despite every other browser (even those that share its certificate store) going in the opposite direction, is just insulting.
For the past 9 months I have been presenting versions of this talk to AI researchers, investors, politicians and policy makers. I felt it was time to share these ideas with a wider audience. Thanks to the Ditchley conference on Machine Learning in 2017 for giving me a fantastic platform to get early…
Summary: The central prediction I want to make and defend in this post is that continued rapid progress in machine learning will drive the emergence of a new kind of geopolitics; I have been calling it AI Nationalism. Machine learning is an omni-use technology that will come to touch all sectors and parts of society. The transformation of both the economy and the military by machine learning will create instability at the national and international level forcing governments to act. AI policy will become the single most important area of government policy. An accelerated arms race will emerge between key countries and we will see increased protectionist state action to support national champions, block takeovers by foreign firms and attract talent. I use the example of Google, DeepMind and the UK as a specific example of this issue. This arms race will potentially speed up the pace of AI development and shorten the timescale for getting to AGI. Although there will be many common aspects to this techno-nationalist agenda, there will also be important state specific policies. There is a difference between predicting that something will happen and believing this is a good thing. Nationalism is a dangerous path, particular when the international order and international norms will be in flux as a result and in the concluding section I discuss how a period of AI Nationalism might transition to one of global cooperation where AI is treated as a global public good.
Excellent inspiring and occasionally scary look at the impact that the quest for general-purpose artificial intelligence has on the international stage. Will we enter an age of “AI Nationalism”? If so, how will we find out way to the other side? Excellent longread.
Those who know me well know that I’m a bit of a data nerd. Even when I don’t yet know what I’m going to do with some data yet, it feels sensible to start collecting it in a nice machine-readable format from the word go. Because you never know, right? That’s how I’m able to tell you how much gas and electricity our house used on average on any day in the last two and a half years (and how much off that was offset by our solar panels).
So it should perhaps come as no huge surprise that for the last six months I’ve been recording the identity of every piece of music played by my favourite local radio station, Jack FM (don’t worry: I didn’t do this by hand – I wrote a program to do it). At the time, I wasn’t sure whether there was any point to the exercise… in fact, I’m still not sure. But hey: I’ve got a log of the last 45,000 songs that the radio station played: I might as well do something with it. The Discogs API proved invaluable in automating the discovery of metadata relating to each song, such as the year of its release (I wasn’t going to do that by hand either!), and that gave me enough data to, for example, do this (click on any image to see a bigger version):
I almost expected a bigger variance by hour-of-day, but I guess that Jack isn’t in the habit of pandering to its demographics too heavily. I spotted the post-midnight point at which you get almost a plurality of music from 1990 or later, though: perhaps that’s when the young ‘uns who can still stay up that late are mostly listening to the radio? What about by day-of-week, then:
The chunks of “bonus 80s” shouldn’t be surprising, I suppose, given that the radio station advertises that that’s exactly what it does at those times. But still: it’s reassuring to know that when a radio station claims to play 80s music, you don’t just have to take their word for it (so long as their listeners include somebody as geeky as me).
It feels to me like every time I tune in they’re playing an INXS song. That can’t be a coincidence, right? Let’s find out:
Yup, there’s a heavy bias towards Guns ‘n’ Roses, Michael Jackson, Prince, Oasis, Bryan Adams, Madonna, INXS, Bon Jovi, Queen, and U2 (who collectively are responsible for over a tenth of all music played on Jack FM), and – to a lesser extent – towards Robert Palmer, Meatloaf, Blondie, Green Day, Texas, Whitesnake, the Pet Shop Boys, Billy Idol, Madness, Rainbow, Elton John, Bruce Springsteen, Aerosmith, Fleetwood Mac, Phil Collins, ZZ Top, AC/DC, Duran Duran, the Police, Simple Minds, Blur, David Bowie, Def Leppard, and REM: taken together, one in every four songs played on Jack FM is by one of these 34 artists.
I was interested to see that the “top 20 songs” played on Jack FM these last six months include several songs by artists who otherwise aren’t represented at all on the station. The most-played song is Alice Cooper’s Poison, but I’ve never recorded them playing any other Alice Cooper songs (boo!). The fifth-most-played song is Fight For Your Right, by the Beastie Boys, but that’s the only Beastie Boys song I’ve caught them playing. And the seventh-most-played – Roachford’s Cuddly Toy – is similarly the only Roachford song they ever put on.
Next I tried a Markov chain analysis. Markov chains are a mathematical tool that examines a sequence (in this case, a sequence of songs) and builds a map of “chains” of sequential songs, recording the frequency with which they follow one another – here’s a great explanation and playground. The same technique is used by “predictive text” features on your smartphone: it knows what word to suggest you type next based on the patterns of words you most-often type in sequence. And running some Markov chain analysis helped me find some really… interesting patterns in the playlists. For example, look at the similarities between what was played early in the afternoon of Wednesday 19 October and what was played 12 hours later, early in the morning of Thursday 20 October:
|19 October 2016||20 October 2016|
|12:06:33||Kool & The Gang – Fresh||Kool & The Gang – Fresh||00:13:56|
|12:10:35||Bruce Springsteen – Dancing In The Dark||Bruce Springsteen – Dancing In The Dark||00:17:57|
|12:14:36||Maxi Priest – Close To You||Maxi Priest – Close To You||00:21:59|
|12:22:38||Van Halen – Why Can’t This Be Love||Van Halen – Why Can’t This Be Love||00:25:00|
|12:25:39||Beats International / Lindy – Dub Be Good To Me||Beats International / Lindy – Dub Be Good To Me||00:29:01|
|12:29:40||Kasabian – Fire||Kasabian – Fire||00:33:02|
|12:33:42||Talk Talk – It’s My Life||Talk Talk – It’s My Life||00:38:04|
|12:41:44||Lenny Kravitz – Are You Gonna Go My Way||Lenny Kravitz – Are You Gonna Go My Way||00:42:05|
|12:45:45||Shalamar – I Can Make You Feel Good||Shalamar – I Can Make You Feel Good||00:45:06|
|12:49:47||4 Non Blondes – What’s Up||4 Non Blondes – What’s Up||00:50:07|
|12:55:49||Madness – Baggy Trousers||Madness – Baggy Trousers||00:54:09|
|Eagle Eye Cherry – Save Tonight||00:56:09|
|Feeling – Love It When You Call||01:04:12|
|13:02:51||Fine Young Cannibals – Good Thing||Fine Young Cannibals – Good Thing||01:10:14|
|13:06:54||Blur – There’s No Other Way||Blur – There’s No Other Way||01:14:15|
|13:09:55||Pet Shop Boys – It’s A Sin||Pet Shop Boys – It’s A Sin||01:17:16|
|13:14:56||Zutons – Valerie||Zutons – Valerie||01:22:18|
|13:22:59||Cure – The Love Cats||Cure – The Love Cats||01:26:19|
|13:27:01||Bryan Adams / Mel C – When You’re Gone||Bryan Adams / Mel C – When You’re Gone||01:30:20|
|13:30:02||Depeche Mode – Personal Jesus||Depeche Mode – Personal Jesus||01:33:21|
|13:34:03||Queen – Another One Bites The Dust||Queen – Another One Bites The Dust||01:38:22|
|13:42:06||Shania Twain – That Don’t Impress Me Much||Shania Twain – That Don’t Impress Me Much||01:42:23|
|13:45:07||ZZ Top – Gimme All Your Lovin’||ZZ Top – Gimme All Your Lovin’||01:46:25|
|13:49:09||Abba – Mamma Mia||Abba – Mamma Mia||01:50:26|
|13:53:10||Survivor – Eye Of The Tiger||Survivor – Eye Of The Tiger||01:53:27|
|Scouting For Girls – Elvis Aint Dead||01:57:28|
|Verve – Lucky Man||02:00:29|
|Fleetwood Mac – Say You Love Me||02:05:30|
|14:03:13||Kiss – Crazy Crazy Nights||Kiss – Crazy Crazy Nights||02:10:31|
|14:07:15||Lightning Seeds – Sense||Lightning Seeds – Sense||02:14:33|
|14:11:16||Pretenders – Brass In Pocket||Pretenders – Brass In Pocket||02:18:34|
|14:14:17||Elvis Presley / JXL – A Little Less Conversation||Elvis Presley / JXL – A Little Less Conversation||02:21:35|
|14:22:19||U2 – Angel Of Harlem||U2 – Angel Of Harlem||02:24:36|
|14:25:20||Trammps – Disco Inferno||Trammps – Disco Inferno||02:28:37|
|14:29:22||Cast – Guiding Star||Cast – Guiding Star||02:31:38|
|14:33:23||New Order – Blue Monday||New Order – Blue Monday||02:36:39|
|14:41:26||Def Leppard – Let’s Get Rocked||Def Leppard – Let’s Get Rocked||02:40:41|
|14:46:28||Phil Collins – Sussudio||Phil Collins – Sussudio||02:45:42|
|14:50:30||Shawn Mullins – Lullaby||Shawn Mullins – Lullaby||02:49:43|
|14:55:31||Stars On 45 – Stars On 45||Stars On 45 – Stars On 45||02:53:45|
|16:06:35||Dead Or Alive – You Spin Me Round Like A Record||Dead Or Alive – You Spin Me Round Like A Record||03:00:47|
|16:09:36||Dire Straits – Walk Of Life||Dire Straits – Walk Of Life||03:03:48|
|16:13:37||Keane – Everybody’s Changing||Keane – Everybody’s Changing||03:07:49|
|16:17:39||Billy Idol – Rebel Yell||Billy Idol – Rebel Yell||03:10:50|
|16:25:41||Stealers Wheel – Stuck In The Middle||Stealers Wheel – Stuck In The Middle||03:14:51|
|16:28:42||Green Day – American Idiot||Green Day – American Idiot||03:18:52|
|16:33:44||A-Ha – Take On Me||A-Ha – Take On Me||03:21:53|
|16:36:45||Cranberries – Dreams||Cranberries – Dreams||03:26:54|
|Elton John – Philadelphia Freedom||03:30:56|
|Inxs – Disappear||03:36:57|
|Kim Wilde – You Keep Me Hanging On||03:40:59|
|16:44:47||Living In A Box – Living In A Box|
|16:47:48||Status Quo – Rockin’ All Over The World||Status Quo – Rockin’ All Over The World||03:45:00|
The similarities between those playlists (which include a 20-songs-in-a-row streak!) surely can’t be coincidence… but they do go some way to explaining why listening to Jack FM sometimes gives me a feeling of déjà vu (along with, perhaps, the no-talk, all-jukebox format). Looking elsewhere in the data I found dozens of other similar occurances, though none that were both such long chains and in such close proximity to one another. What does it mean?
There are several possible explanations, including:
- The exotic, e.g. they’re using Markov chains to control an auto-DJ, and so just sometimes it randomly chooses to follow a long chain that it “learned” from a real DJ.
- The silly, e.g. Jack FM somehow knew that I was monitoring them in this way and are trying to troll me.
- My favourite: these two are actually the same playlist, but with breaks interspersed differently. During the daytime, the breaks in the list are more-frequent and longer, which suggests: ad breaks! Advertisers are far more-likely to pay for spots during the mid-afternoon than they are in the middle of the night (the gap in the overnight playlist could well be a short ad or a jingle), which would explain why the two are different from one another!
But the question remains: why reuse playlists in close proximity at all? Even when the station operates autonomously, as it clearly does most of the time, it’d surely be easy enough to set up an auto-DJ using “smart random” (because truly random shuffles don’t sound random to humans) to get the same or a better effect.
Which leads to another interesting observation: Jack FM’s sister stations in Surrey and Hampshire also maintain a similar playlist most of the time… which means that they’re either synchronising their ad breaks (including their duration – I suspect this is the case) or else using filler jingles to line-up content with the beginnings and ends of songs. It’s a clever operation, clearly, but it’s not beyond black-box comprehension. More research is clearly needed. (And yes, I’m sure I could just call up and ask – they call me “Newcastle Dan” on the breakfast show – but that wouldn’t be even half as fun as the data mining is…)
Scientists investigating this week’s catastrophic lunchquake in the Dan’s Lunchbox region have released a statement today about the techtonic causes of the disaster.
“The upheaval event, which reached 5.9 on the Tupperware Scale, was probably caused by overenthusiastic cycling,” explained Dr. Pepper, Professor of Lunchtime Beverages at Tetrapak University.
“The breadospheres ‘float’ on soft, viscous eggmayolayers. Usually these are stable, but sometimes a lateral shift can result in entire breadosphere plates being displaced underneath one another.”
This is what happened earlier this week, when a breadospheric shift resulted in catastrophic sinkage in the left-side-of-lunchbox area, eggmayolayer “vents”, and an increase in the height of Apple Mountain.
No lives were lost during the disaster. However, two jammie dodgers were completely ruined.
Recent emissions in the ring of fire area is unrelated to this recent lunchquake, and are instead believed to be associated with excessive consumption of spicy food at lunchtimes.
I’m scared. Kit is researching the laws governing marriage in Hawaii, and I’m not exactly sure why.
“Hey; you can get married at 15 in Hawaii!”
Meanwhile, I’m currently coding a wiki engine. For those of you who aren’t in-the-know, a wiki is a collaborative network of web pages that anybody can edit. They’re fun, if a little anarchic.
Back to the code…