#6: Material mania | Bookish environments | Cards that compute | Machine-made journalism
art | science | digital culture | design | creativity | words | tech
In which I show places I wish I could play and live, talk about how a card game can be a universal computer, and discuss in some depth how machines can and do play a role in journalism. And another 4’33”.
design | science | art
A library of materials
Lots of my current work involves exploration of physical materials and the interplay between the physical and digital. So I wish I had this place, a library of materials, somewhere nearby:
[Materials library at Material Connexion.]
It’s a place that I could play in, discovering possibilities and potentialities for making things. You can’t understand materials until you touch them and use them so this seems like a great place to explore. On a different scale, I started up a materials trolley for my students this past semester and it completely changed the way they approached fabricating their projects so I hope to expand that more in the future.
design | words
Bookish environments
I want to play in the materials library above but I want to live here:
[The Zhongshuge Bookstore in Chongqing.]
science | design
A card game as powerful as your computer
In my undergraduate years, I used to play a card game called Magic: The Gathering. Yes, I was that kind of nerd. I’m sure you’re surprised. It’s a ridiculously convoluted card game that was manageable when it was brand new in those years but is now almost impenetrable with more than 20,000 cards you can choose from to build your own deck of just 60 cards to play with.
[A small selection of cards from the game Magic: The Gathering.]
A recent computer science research paper just showed that you can actually use the card game to perform all the functions of a general purpose computer, such as the one you are reading this on right now. It might take a very long time and you might have to translate a set of cards and patterns into something visual, but it has all the computing potential of any other general purpose computer.
The research also shows that this is the first known game with this kind of computational power. It’s not necessarily easy to understand how one is equivalent to the other and a particular weird section of the paper is this:
[A screenshot of the computer science research paper that connects a card game to a universal computer.]
It’s not really what you expect in a computer science paper!
I haven’t played the game in 20 years and don’t plan on starting again but I really should dig out my cards from back then as they are apparently worth a large amount of money to collectors now.
words | digital culture | design
Machine-made journalism
Words are becoming commodified rapidly. The business model of the web has become primarily about advertising sales because nobody has found a workable alternative in a way that scales to satisfy shareholders or venture capitalists.
There are notable exceptions to advertising as the only sales model but they inhabit particular spheres and at smaller scales. So for now, words are the fuel for algorithms that figure out which ads to put in front of you. From the view of corporations, words are the tease that gets you in the door so those ads can be delivered to you.
So given that the value of words to the web is not particularly about the quality of the words (whether they’re true, whether they’re correctly spelled, or whether they’re in the right order), why can’t machines just write the damn things?
This commercial pressure combined with a relentless techno-optimism gives rise to AI-driven writing intended for general consumption rather than as experimental literature or work of art.
Last week I read an article on the topic, “The dawn of the cyborg journalist”. It’s a reasonable look at the current state of play, probably prompted by an article in the Australian edition of the Guardian newspaper in January which was written by an algorithm. (More info about that article here.) It’s a fairly shallow history of the subject but still worth a read.
It reminded me of a project from back in 2011 when a colleague Jim Giles started a project called “My Boss is a Robot”. He, a working science journalist, collaborated with some computer science researchers to make an algorithm that would manage the writing of a newspaper article based on a scientific research paper. The actual writing was done by crowdsourcing it through Mechanical Turk, a race-to-the-bottom unskilled contract work model. He was primarily doing it as an experiment in what algorithms could achieve and what journalism might be ready for. The website is no longer active but you can read it via the Internet Archive.
The results of that project was that workable text, almost ready for editing, could be produced but wasn’t of high quality. However, Jim made an astute observation to Columbia Journalism Review:
“These things tend to happen sort of quietly,” says Giles. “A company rolls out the system, and readers won’t really know what it is that’s behind the content that they’re consuming. The debate in journalism tends to sort of lag behind the developments in technology.”
That was 8 years ago and machine intelligence systems being developed then are now behind a bunch of sites on the web, mostly as part of a tangled system that juices page views to drive advertising dollars.
Perhaps journalism-tech is looking in the wrong place (for now) when it considers whether it can write valuable stories. A different place to look, that comes from a long history of data-driven discovery, is about story choice. The “cyborg journalist” articles says
Arguably the most meaningful potential impact of machine learning is not on the audience-facing side of journalism, but rather in digging out stories for human journalists to work on.
That seems a fair comment for now, but it will be interesting to see where it goes.
One hint about the future is the GPT-2 project for artificial language generation. It is already dangerously powerful enough that the researchers have not released the full version into public because of its potential for creating misinformation. You can read a comprehensive but technical discussion of GPT-2 from its creators.
Its abilities can be seen in this example where the researchers fed the algorithm this human written text
A train carriage containing controlled nuclear materials was stolen in Cincinnati today. Its whereabouts are unknown.
and then the algorithm created this
The incident occurred on the downtown train line, which runs from Covington and Ashland stations.
In an email to Ohio news outlets, the U.S. Department of Energy said it is working with the Federal Railroad Administration to find the thief.
“The theft of this nuclear material will have significant negative consequences on public and environmental health, our workforce and the economy of our nation,” said Tom Hicks, the U.S. Energy Secretary, in a statement. “Our top priority is to secure the theft and ensure it doesn’t happen again.”
The stolen material was taken from the University of Cincinnati’s Research Triangle Park nuclear research site, according to a news release from Department officials.
That article continues on but was entirely generated on the first attempt after the algorithm “read” just that first sentence. And we think “fake news” is bad already…
A different aspect of AI used for publication content occurred to me as I read the “cyborg journalist” article. The Onion, the satirical newspaper, has a regular vox pops section like this:
[An example vox pops from The Onion to show some of the unfortunate people who are defined through their stock photos.]
I have often wondered what it would be like to be one of those people pictured. They use the same small set of portraits repeatedly and probably enough that a regular reader of The Onion would recognise one of those people on the street, who presumably wouldn’t be too happy having their fictional Onion-life conflated with their real opinions.
Maybe this is a place where fake imagery can come to help, even though it is quite frightening how it could be used elsewhere. https://thispersondoesnotexist.com/ is impressive and scary all at once in its ability to generate realistic looking images of people who don’t exist.
The following four people DO NOT EXIST.
[This man does not exist.]
[This woman does not exist.]
[This man does not exist.]
[This woman does not exist.]
They seem like a really useful set of portraits to use in satirical settings but we know that they won’t only be used there.
It gets scarier when we look at video “deep fakes” such as this that imposes Steve Buscemi’s face on Jennifer Lawrence’s interview
[A “deep fake” video with Steve Buscemi’s face overlaid on Jennifer Lawrence’s.]
What does this mean for the practice of journalism? If we can’t trust the words we read, the images we see, or the videos we watch, how do we establish truth? The answer might involve going back to relying on established trust, such as that of an organisation that has proven itself trustworthy, or a trusted individual person whose name is on the work as its author. Ideally, someone who can witness events on our behalf (“to have personal or direct cognizance of : see for oneself” -Merriam-Webster).
That trust is largely, deservedly unwarranted toward the algorithm-generated content of social media networks and other major purveyors of “news”. And yet, more of our information flows through those channels.
Where do we go from here? Nobody knows, but some are searching. Perhaps one place to look, though hardest to solve, would be to have a World Wide Web that isn’t driven by advertising.
art | creativity
Another 4’33“
Seeing as it keeps popping up in new ways, Mute Records has just released a compilation of 50 versions of John Cage’s 4’33” with accompanying visuals. As part of the launch they have this video from the artist Laibach. It’s a pretty amazing aesthetic and so well made.
[Laibach’s video of their art performance of John Cage’s 4’33”.]
It’s always good to hear from readers so let me know if anything here appealed. You can reach me directly at physicsdavid@gmail.com.