Tag Archives: machine learning

Things July 2017: Archive Adventures – part 1

In November, Things will be 10 years old. Since the beginning, I’ve collected an archive of interesting things, and at various intervals then created Things out of that archive. I tend to collect slightly more things than I publish, so the archive has grown. Rather than cull it, I’m going to just put them all out in two big catch-up editions. In this edition: Stories, Technology, Imagery, and all the Puzzles. This is going to be intense!

This Just In

Before I get to the archive, a couple of recent things.

Damien Henry’s video for Steve Reich’s “Music for 18 Musicians” makes excellent use of machine learning for art. Using video shot from a moving train, an algorithm learns how to predict what the next frame will look like. Every 20s into the video the amount of learning used increases. The result is fascinating, and perfectly complements the music. You can skip through the video to get an idea of the effect, but it’s best played in full!

There is an awkward vein of humour in which comedians interview (often) unsuspecting subject-matter experts in a non-serious manner (Philomena Cunk interviewing Brian Cox is one I don’t mind so much). Superficially, Werner Herzog sometimes takes a similar approach in his documentaries (noted in Things 118 in 2012, “Please describe an encounter with a squirrel”), but there is a hidden earnestness behind his questions. If you’ve never seen any of his documentaries, this excellent short clip on penguins shows you what you’re missing.

Stories

I think via Richard L, here’s an interesting piece on Plots not involving conflict.

In some stories/fables/myths the inciting incident or key point of drama is a character attempting to do something different to normal. In old stories, the character often fails and is punished, the moral being “know your place” (for example, the crow that tries to sing and in so doing drops a piece of cheese). In more modern stories, the character often succeeds and is rewarded. Is that generalisation true though? And what does it mean? I was going to collect examples of each and try to see a pattern, but never ended up doing that, so I’ll just leave it here as an unfinished thought.

The Cosmology of Serialised Television is an essay by David Auerbach which categorises long-running TV fiction by cosmological universe types: Steady State, Expansionary, Big Crunch. I didn’t find it particularly useful in terms of identifying solutions to the intrinsic storytelling problems of the medium, or even uncovering hidden gems (just the expected “Everything is terrible except for The Wire and Babylon 5“).

Still, it’s a lot of fun to read and nod along to, with some great terse summaries along the way, for example:

“So comics evolved by directing creative effort away from any moments of quality and toward large-scale creative bankruptcy”

See Maris Wicks’ 16-panel summary of Kitty Pride’s relationship with Colossus (penultimate comic in this post) for a great example of this.

Choose Your Own Adventure

One of my favourite kind of stories is any subversion of the “Choose your own adventure” format. Previously on Things I’ve linked to Luke Surl’s perfectly terse ‘Free Will‘ page. As an implied CYOA, this previously-linked Beaver and Steve comic also remains a classic.

The “Prog 2012″ issue of 2000AD featured a story called “Choose your own Xmas”, and subverts the format while also shattering the 4th wall. It’s completely brilliant but isn’t online so you’ll just have to trust me and buy or borrow it.

Viviane Schwarz also subverted the format beautifully to convey what it’s like to suffer from anxiety.

Working within the extraordinarily tight constraints of having a fixed panel and art layout, Ryan North still found a way to make a CYOA version of Dinosaur Comics.

Save The Date (recommended to me by Tarim and Richard L, available for free on Windows, Mac or Linux) leverages the Ren’Py ‘visual novel’ engine in a rather clever way, and if you’re interested in games and storytelling, have 30-60 minutes to spare, and can tolerate an apparently high rate of failure, you should give it a go. It could perhaps have delivered its message more elegantly at the “end”.

Also via Richard L, Trapped In Time (pdf) is a nice twist on the traditional choose-your-own-adventure format that leverages the format in a few interesting ways I’ve not seen before, and actually works very well as a pdf!

And I should mention this just in – browser-based interactive time travel fiction, One Night in Skegness, by Matheson Marcault (Holly Gramazio and Sophie Sampson). More time travel, but in a more relaxed way.

Technology

Perhaps surprisingly, a link about technology that sat in my backlog for 5 years has become more interesting with age. John B recommended the article, in which Alexis Madrigal laments that (in 2012), internet startups are just retreading the same ground and no longer promise to revolutionise our lives. John also pointed out the comment below the article by Urgelt (comment link doesn’t seem to work, wait for comments to load, which they often don’t, then find ‘Urgelt’ in the page). Urgelt more precisely categorises startups into those that grow the economy vs those that just take market share from existing businesses, and the issue is that most startups are falling into the latter category.

Both make a very interesting read 5 years further on, leading one to ask: has anything changed since then? Two trends jump to mind.

The Gig-economy-style startups (Uber, Deliveroo, AirBnB) suggest major changes, but might not actually scale. Uber charges around half of the true cost, subsidising the rest in a bid to achieve market dominance when driverless cars arrive; Deliveroo similarly rides an unsustainable cost/charge balance for much the same reason; only AirBnB arguably doesn’t fit into this category as the market sets the price.

Crowdfunding seems to be a more impressive development. In 2012, Kickstarter was just turning the corner, and Patreon is now in the ascendancy. From artists/game-makers that I follow, these services genuinely seem to be creating viable revenue-streams that were not previously available, to the benefit of culture in general. For example, Captain Disillusion (referenced in Things March 2015 and Things 17) was never mass-market enough for ad-revenue to be viable, but now raises sufficient money from Patreon to work on his videos pretty much full time.

In more “modern life is terrible” news, here’s an article from Cracked in late 2013 that is really just an enjoyably angry and sarcastic rant about clickbait content-farming. I remember at the time thinking that, if nothing else, the clickbait headline style would have to change as humans will fall for anything once, or maybe a hundred times, but eventually will develop a sort of semiotic herd-immunity to these well-dressed empty promises. Four years on… have things changed? Well, if nothing else, Facebook is at least attempting to deprioritise these sorts of headlines; more specifically, headlines that withhold key information, and headlines that senselessly exaggerate the content.

Imagery and Comics

Wondermark on the thought-experiment of money having a continuity rather than just being an abstract quantity of value. Wet owlsInopportunely placed stickers. Bikes recreated (digitally) from drawings. The Door to Hell: “They set the hole on fire, expecting it to burn itself out of fuel in a few days. Now, some 42 years later, it is still burning”.

Puzzles/Questions

Collecting all unused puzzles here is probably too intense, but I quite like that about it, so I’m doing it anyway.

1) 2D News

Sci/tech news is often quite one-dimensional, revealing a single scientific discovery or technological advancement at a time. As a thought-experiment, try combining two or more such stories from the past year into something amazing. For example, advances in drone technology + advances in ‘invisibility cloaks’ = army of invisible drones. Finding loads of other planets + anything = awesome.

2) Put Put boat

A toy Put Put boat has an amazingly simple heat engine, which you may recognise from the film Ponyo. A candle heats a small boiler; some water in the boiler vaporises but cannot escape, generating pressure that pushes the remaining water out a pipe. The momentum of the water keeps it pushing out, leaving the inside of the boiler with low air pressure. As a result, water then rushes back in, and the process repeats. Water thus repeatedly enters and exits a pipe pointed out the back of the boat, which then travels forward in a halting manner. The puzzle is this: why does the boat actually move forward, instead of just moving back and forth, given water is just going in and out?

3) Bernoulli vs the Train Window

In a similar vein to the Put Put boat, we have the Bernoulli train window problem. Bernoulli’s principle roughly states that faster moving air acts as if it has lower pressure. The classic demonstration is to hold a piece of A4 paper by one end in front of your chin so it droops downwards away from you; by blowing over the top, the pressure is reduced, and the piece of paper rises up due to the higher pressure underneath it.

A similar effect could be seen in an old-fashioned push-to-close narrow train window. If such a window was open and the train entered a tunnel, the window would slam shut. Or, would a shut one blow open? Depending on whether you take the perspective of the tunnel or the train, the faster moving air is on one side of the window or it is on the other. So which way does the window go?

4) Catbird seat

The Catbird seat is one of those puzzles I like because you can solve it with drawings and trigonometry, and then you can solve it better with simpler drawings and simpler trigonometry, and then you wonder if you can solve it in some kind of purely intuitive manner.

5) Shape of a Harp

When plucked, a longer taut string makes a deeper note. A harp has progressively longer strings to cover a range of notes. However, even though the interval between each note is the same, the length of the strings does not change linearly, or even following a simple curve, but rather an S-curve. Why is that? I thought this question might have a nice intuitive solution that could be reached by reason rather than by physics, but the answer is a bit more disatisfying, so this question remained in my backlog unasked. If you don’t want to figure it out (and my personal opinion is it’s not that interesting to do so), you can read about it here, although you’ll need the internet archive page for the harp citation.


Not a harp, but the principle is the same

- Transmission finally ends

Things April 2017: Multiplayer Mobile games, Weight loss and Physics, Paranoia and Tolerance

This issue of Things was initially drafted in January 2016 and for various reasons is just coming out now. Let’s see how out of date it is…

Mobile Game of the Moment: Dreii

Mobile gaming is very exciting, because touch-interface location-aware always-online devices open up an amazing new landscape of interactive possibilities. Mobile gaming is also very depressing because these possibilities are rarely harnessed in an interesting way, and even when they are it rarely leads to commercial success.

This is exactly why I’m recommending you get Dreii right now (available on Android, iOS and also on Steam).

It’s an elegant physics-based puzzler in which you try to stack objects under various challenging constraints. The really brilliant bit is that you get to co-operate with anyone else that happens to be playing the same level (on any device) at the same time as you. There’s also a rather lovely map visualisation in which you can see what levels others are playing to better seek them out – or go back and help another player make it through to the one you’re stuck on!

The helpfulness of others is massively varying, which is part of the charm. I recommend it right now as the recent launch ensures a goodly surge of currently active players.

Update: well, that was written about a year ago, so I can’t really recommend it as much because few are now playing it. You can do local co-op with someone else playing at the same time though, so it’s still worth a go if you can arrange for that!

Or go and play the other brilliant game I got into in the intervening time: Snakebird (iOS, Android, Steam), an incredibly simple yet extremely difficult puzzle game.

Or go and play Crash of Cars (iOS, Android), which is real-time arcade-style car combat and is being played by a few million people at the time of writing.

Humans are More Complicated than Physics, part 94: Weight vs Diet

If you want to lose weight, don’t eat. This is not medicine, it’s thermodynamics. If you take in more than you use, you store it.
– Michael Bloomberg

It seems most people agree with Bloomberg’s appealing logic. It turns out, as it often does, that humans are more complicated than simple physics would suggest.

Vox has a great look at the energy-in energy-out model here. The most striking conclusion of their wide-ranging review of research is that exercise is great for health, but not for weight loss. For weight loss, diet is a much more important component. So Bloomberg is sounding roughly right, but the mystery then deepens when we consider a study that found people on the same diet and exercise regimes put on more weight now than they did in the 80’s.  Note that although the Atlantic’s headline is “Why It Was Easier to Be Skinny in the 1980s”, the article doesn’t answer the question, because we don’t actually know. We just have some theories.

Finally, on Aeon, David Berreby dives a lot deeper into this mystery. Most striking for me was the finding that over the last two decades, animals as well as humans have gained weight – including lab mice, which have gained an average of 11% per decade despite having rigorously consistent diets. If you want to read about some fascinating theories on why this might be happening (light? BPA? A virus?), go check it out.

Since that was all a bit depressing, here’s some fun data on how the UK’s diet has changed over time.

Paranoia in Politics

Quite a few months ago (er, a year ago now) Charlie Stross wrote about the Paranoid Style in politics. He cites two fascinating essays charting the relationship between paranoia / conspiracy-theories and US Republicans (one from 50 years ago and one more recent). Things like “9/11 was an inside job”, vaccine-related conspiracies, or any time you hear the phrase “Liberal elite”. Stross then adds his own interpretation of how the internet has reinforced this and given rise to (take a deep breath) an “ad-hoc movement of angry ideologues who have jabbed their fungal hyphae into the cerebral cortex of Reddit and n-chan to parasitically control the rageface collective.”

Well, a year has gone by and this has all become rather more clear. I think the only useful thing to add is that Popularism (in the sense of a political movement that believes most of the existing power structures are self-serving, corrupt, and out-of-touch) is clearly a close relative of the Paranoid Style.

Google Image Reverse Search growing in Intelligence?

If you want to search based on an image (to find where it came from, or a higher resolution version) you can use TinEye and get limited but exact results. Or you can use Google’s “Search by image” functionality to get a wide range of approximate results. I do this quite a lot while trying to track down the artists behind work I post on my daily drawing Tumblr.

I recently tried to use it to trace the origin of this piece:

I was unable to find the artist, but I was extremely impressed that Google suggested a page of ‘similar images’ which were indeed a lot of paintings of cats with a similar colour palette and often distorted proportions:

You can of course use these properties of the search algorithm to generate art.

Tolerating the Tolerable

I can Tolerate anything except the Outgroup” is an essay by Scott Alexander on filter bubbles and tolerance. It’s so interesting to me that I’ll summarise it all below, but I recommend reading it in full.

  • Alexander defines Tolerance as “respect and kindness toward members of an Outgroup”, and defines an Outgroup as a group that has “proximity plus small differences”: a group of people who live in the same neighbourhood but who are ‘slightly’ different to you. I’m capitalising these terms because the definitions aren’t sufficiently general – for example, you could be racist but still Tolerant under this definition.
  • He broadens typical US political alignments into ‘tribes’: Conservatives are Red tribe, Liberals are Blue tribe (the US political colour binary reverses the UK’s). As a side note he also identifies a libertarian-leaning Grey Tribe, which I’ve found to be a useful concept – the Grey tribe is typified by:

“…libertarian political beliefs, Dawkins-style atheism, vague annoyance that the question of gay rights even comes up, eating paleo, drinking Soylent, calling in rides on Uber, reading lots of blogs, calling American football “sportsball”, getting conspicuously upset about the War on Drugs and the NSA, and listening to filk”

  • Alexander then asserts that the liberal Blue tribe’s outgroup is in fact the Red tribe. He gives the interesting example of being roundly criticised by Blues for expressing any kind of relief at Osama bin Laden’s death, only to later see those same people openly celebrating the death of Margaret Thatcher.
  • Finally, he suggests that political articles that reference ‘Americans’ or ‘White people’, written by Blue tribe people (who are notionally included in those groups), are tacitly actually about Red tribe members, and when a Blue says “I can tolerate anything except intolerance”, they are identifying intolerance with the Red tribe and are actually saying “I can tolerate anything except the Red tribe / my Outgroup” which no longer sounds that impressive.

In an excellent coda, Alexander then realises that he’s doing the same thing: criticising a group he notionally is a member of (the Blue tribe), but actually criticising his Outgroup, as he realises he’s probably Grey.

While I think this is an interesting argument, I do think it’s important to note that being intolerant of a group of people who hold opposing political views (which can’t be identified on sight, and can change) seems far less egregious than being intolerant of people who simply look a certain way. Views are, after all, one step away from Actions, but that’s a distinction I’ll get to in the next edition of Things.

- Transmission finally ends