Jekyll2021-12-23T13:59:18+00:00https://ysig.github.io/feed.xmlysigTheory without practice, can at least be inspirational.ysigioannis.siglidis@ens-paris-saclay.frMy Covid Eating Habits as a Youtube Paper Clip Problem.2021-07-01T00:00:00+00:002021-07-01T00:00:00+00:00https://ysig.github.io/youtube-obesity<p>I like cooking or that’s at least what people like to circulate about me. YouTube thinks the same. Before Covid I wasn’t watching cooking videos that much.
I didn’t know a lot about the FoodTube and neither was I watching that much of YouTube. At most, it was for me another form of entertainment.
Today, I would say that I am addicted to YouTube. My hands have learned to open a new window, write ‘y’, ‘o’, ‘u’ and press “Enter” that fast, that it surely is a subliminal practice, which probably allows me to get distracted whenever I meet a difficulty.
I realized this when I installed a program that would forbid me from entering YouTube.
In the end I removed it, as I was using another browser to get to it.</p>
<p>And in the beginning this was wonderful. I searched really cool music and I watched interviews from really important scholars.
But slowly I got hooked up again and I watched FoodTube or a Werner Herzog/Nassim Taleb interview, quality material one can argue, but it doesn’t matter: they are all part of the YouTube addiction feedback loop.
This is the answer to someone saying: “people used to watch television, YouTube is another form of it”.
Yeah, but this is a bit smarter: there are areas in the YouTube graph that are sinks, like black holes of a semantic structure, from which it’s really difficult to escape.
The reason is simple: if you search really cool (rare and interesting) music, it is highly unlike that YouTube will propose to you another type of cool music again.
It’s wire-heading, will gradually bring to you something mainstream and addictive.
After all who cares about spending quality computation for the masses. It’s much simpler to influence people than to understand and help them bloom.
For example if I watch a video about Michel Serres, it is really unlikely that I will receive him again on my feed, and even if I receive him and not click at him (because maybe I am not in the mood of watching this today - this is what liberty of education is right?) forget about it.
But if instead, I watch a lecture of Nassim Taleb, I will get him on my feed all of the time.
And I will click on him of course. But my point is, that what you click is not necessarily something you like seeing all them time, but something that is addictive.
Actually I would argue that the whole large scale social experiment of YouTube is about learning how to turn people to normal addicts of normal products, i.e. building a <em>consumer culture</em>.</p>
<h2 id="how-i-got-addicted">How I got addicted</h2>
<p>I realized, that I had an addiction to YouTube, before I saw this shallow but important movie “Social Dilemma”.
The origin of my addiction was simple. It started during Covid, confined with a ton of work (Master + Projects + …).
So much (mental) work that I didn’t had the time to entertain myself in any other way.
Some days I would consecutively work for 16 hours straight.
I literally would wake up, open my laptop and close it to sleep, having tiny breaks to eat simple things.
Plus, I was living in a CROUS residence: a really affordable student residence that the French government provides to some students, which is quite nice, but a “student camp” nevertheless (at least with the confinement it felt a bit like that).</p>
<p>Yes, some times it is really satisfying to work that much, but after a point when you realize that you are part of train that needs you as a fuel to move forward, it’s horrible.
Also my brain started to develop a need for “something that is not work”: social interaction it is, watching memes it is, eating it is, you name it.
This cognitive resistance, gave me the idea that we could maybe even argue biologically about leisure.
I now strongly believe that as sleep is really important to organize our thoughts and make us more productive, the same is the case for the non-productive hours for our creativity.
I felt that all my creativity was strained after a lot consecutive days of constant work that <strong>I had to do</strong>. In other words: I felt a slave.
I also got separated from a person I loved a lot (while we lived in different countries) and this made things even harder.
Having both my side-hustle and a very competitive degree to perform to, I didn’t have another choice; and at least my work kept me walking.</p>
<p>Moreover, Covid isolation felt a bit like this black mirror episode (San Junipero) with these two women who fall in love, but who both are living in a comma-situation inside an asylum for the elderly, but a bit grimmer as I didn’t make any new relationship in any form, although I made toxic (spectacle based) dreams that managed to unclog my dampened soul as “Tuboflo”.</p>
<p>Yes I spoke a lot about myself, but just to give you a context. Nobody get’s addicted to something without a reason.
And this doesn’t necessarily mean that the addictive entity is the cause of all evil.
Sometimes our own problems just pile up and exactly because we are addicted to something to distract ourselves, we can’t move forward.
But anyway, you are not here to hear me play the role of Gabor Mate (another brilliant person I discovered through YouTube).</p>
<p>Yes, one could also ask the question “what happens when what is addictive is what makes you appreciate things about life?”
A system of addiction is much more than an addiction I would answer, but again yes, I have a more interesting story to tell; one more related to my discipline.</p>
<h2 id="the-paper-clip-problem">The Paper Clip Problem</h2>
<p>You may have heard about this problem. I actually learned it from Nick Bostrom (from his Google Talk on Super Intelligence, that I watched on YouTube).
The idea is that an intelligent agent has been programmed to maximize the amount of paper clips it collects.
In the end, after it has used all the available steel it starts destroying humans in order to extract from them the materials need to make paperclips.
What is interesting in this dystopic scenario, is that although the agent is super smart (as it finds a way of using humans to make paper clips), what it is trying to maximize is as silly as likes (or view-time).</p>
<p>In general machine learning works by maximizing objectives. These objectives are in fact often really simple: we can understand them as paperclips themselves.
Actually the objective of YouTube is simple. Increase and well distribute the views, the channel subscription, the view time (you name it).
The idea is that by learning on simple objectives, you supposedly learn “good features” for human behavior and you can more effectively run ads at scale (remember this can restrict our model from becoming “really smart”).
But YouTube is effectively an RL-system which tries to also influence you towards an ad instead of just recording and analyze your behavior.</p>
<p>Without diving into the necessary discussion on whether an asymmetric (as it processes your information by the information it already has for all other people) large scale computational social entity can enforce it’s predictive behavior on you, basically turning you into a normative category (something that I would call “computational peer pressure”), let’s approach the YouTube paper clip problem in a more naïve way:</p>
<p>One day you click on a proposed video (disobeying Jerome Lanier’s advice) of an Italian recipe of “Aglio and Olio” from the addictive Italian-British cook Gennaro Contaldo.
Next, you click on a proposed video by Vice Munchies about something cool in Italy. Next, you click on a video about fried chicken.
This is the end. Fried chicken is that an addictive form of content (see the visibility of Zack Choi’s Muckbang channel for example) that the addiction of other people will multiply the relevancy of such a topic as a proposed one.
It depends on how youtube categorizes your “taste” but long-story short what is going to happen is that if you click on a “fried chicken” video, you will have it on your feed forever.
This was the case for me: I was waking up every morning and either opening YouTube or even my Google feed (part of the news feed of my Android Phone) I would get a proposal in the category of “fried chicken” to either read or watch.</p>
<p>I never expected that I would so clearly witness a case where artificial intelligence is that pervasive.
Imagine how it would be if someone posted new “ads” all over your place every morning and some of them where always pictures of juicy fried chicken.
The relationship we have with our mobiles is a bit intimate after all and thus they are our home.
I also started making fried chicken after a point, while at the same time I was skeptical of why I was gaining weight.
I should also add to this that a lot of people I known of, that watch FoodTube, have started using much more butter for example in their cooking and/or get weight because the eat much more.
Butter makes the food much tastier and also it makes it much more visually appealing (as it “thickens the sauce” or forms a “brown crust”), creating an expectation that it would fill (could because I highly skeptical of what is the percentage of recipes that people see on FoodTube that the actually try to recreate).</p>
<p>If we also factor in the effect of the COVID-confinement to my physical exercise, this was destructive for my shape.
I watched lots of recipes per day to “clean up my mind” (without ever cooking a single of them) and I also had the feeling that I can entertain myself at home, without this forcing me to even have a walk or go out for whatever reason.
I got more that ten kilos and I felt that I was part of paperclip problem: YouTube was maximizing it’s objective (a simple one) and as a result the algorithm was using “any resource it had available to produce it’s paper clips (i.e. increase it’s views)”.</p>
<p>What this actually means: YouTube was seducing me all day with pictures of food (imagine again seeing all day posters of delicious food whenever you try to relax for a moment) and other content in order to stay on the platform and as I was susceptible to it, I got hooked to it and served it.
It also became my main platform for any form education or entertainment: from new films and film-makers, to scholars or to politicians.
In turn, this made me question all the things that I admired watching on this platform from political figures (like AOC) to philosophers (like Zizek), due to their addictive meme like character.</p>
<p>I decided to write this article to expose my situation and challenge you to look deeper into YouTube’s addictive aspect and most importantly to acknowledge that you are potentially part of its paperclip problem. Some people also refer to it as YouTube radicalizing aspect, but although I find this term to be accurate in certain regards, I find that it comes from an ideological standpoint of the balanced/civilized academic.
(We are living in a fucking social jungle (even in terms of desires), thus does YouTube facilitate it’s emergence or produce it? - I think it does both of course, but the rigorousness required to address this, becomes gatekeeping from non-academic studies and approaches.)</p>
<h1 id="dumber-platforms-is-all-you-need">Dumber Platforms is all you need.</h1>
<p>Searching for a solution, someone would suggest to enrich the YouTube System with better objectives.
Actually it’s really difficult for now to evaluate what “quality time” is:
To argue that, I should tell that there are really important things in my life which have had a very deep influence in me and which I would watched only once (or twice) and never revisited for years.
It is thus significantly difficult for a platform to understand such types of connections from your current data-trails.
And in what context would they have any incentive to enforce them?</p>
<p>Actually ads, that such platforms are funded from and base their business models on, try to sell a commodity.
The idea that a good product means good quality time, “because these are the products that survive in the long run” is basically one that sees the history of a commodity as something predictable.
A commodity passionately waits to be sold to be reproduced through capital (money invested in more people, more computing power, more data, more IP, more whatever).
It is afraid to die and puts pressure on you to buy it. The pressure of it’s investment or even the anxiety of it’s creators of “observing/appreciating their novelty”.</p>
<p>Even if we ignore incentive and follow the “scientific fantasy/fallacy” of attempting to observe someone independently (without influencing them) and identifying “quality-time” by observing in real-time their organism through an AI microscope (the neuralink kind of approach), the information and computing-resources which we would need to record and analyze how someone actually feels, would make us (the noble people) wonder: “Is it really the goal of science to surveil humans just for the purpose of good predictions?”.
Nothing can be more pervasive to the human subject than truth itself. As a “détourned” idea of Herzog:
“If you shine lights to all parts of the human mind, then this mind becomes uninhabitable.”</p>
<p>Thus until I am proven wrong, I think we should demand more <strong>dumber</strong> platforms or at least platforms that don’t use interactive learning (like R.L.), acknowledging that they may perform a bit worse (on some tasks) and provide us a bit less pleasure, but “it’s ok”!</p>ysigioannis.siglidis@ens-paris-saclay.frI like cooking or that’s at least what people like to circulate about me. YouTube thinks the same. Before Covid I wasn’t watching cooking videos that much. I didn’t know a lot about the FoodTube and neither was I watching that much of YouTube. At most, it was for me another form of entertainment. Today, I would say that I am addicted to YouTube. My hands have learned to open a new window, write ‘y’, ‘o’, ‘u’ and press “Enter” that fast, that it surely is a subliminal practice, which probably allows me to get distracted whenever I meet a difficulty. I realized this when I installed a program that would forbid me from entering YouTube. In the end I removed it, as I was using another browser to get to it.NFTs2021-05-10T00:00:00+00:002021-05-10T00:00:00+00:00https://ysig.github.io/nfts<p>I recently started creating artworks in the distribution form of NFTs.
Actually, the photo I have as an avatar in my website is part of one of them.
It is the last picture of an optimization sequence of a StyleGAN2 trained on “ffhq”, trying to illustrate the phrase “In the Waves”, using a loss from CLIP.
The code related to this process, can be found at: <a href="https://github.com/ysig/stylegan2-ADAxCLIP">https://github.com/ysig/stylegan2-ADAxCLIP</a>.
“In the Waves” is the name of the painting from Gaugin, that I have as a logo on my GitHub page.
I let you discover them at: <a href="https://opensea.io/collection/ysig">https://opensea.io/collection/ysig</a></p>ysigioannis.siglidis@ens-paris-saclay.frI recently started creating artworks in the distribution form of NFTs. Actually, the photo I have as an avatar in my website is part of one of them. It is the last picture of an optimization sequence of a StyleGAN2 trained on “ffhq”, trying to illustrate the phrase “In the Waves”, using a loss from CLIP. The code related to this process, can be found at: https://github.com/ysig/stylegan2-ADAxCLIP. “In the Waves” is the name of the painting from Gaugin, that I have as a logo on my GitHub page. I let you discover them at: https://opensea.io/collection/ysigThe Hedonism of Logging2021-05-10T00:00:00+00:002021-05-10T00:00:00+00:00https://ysig.github.io/logging_hedonism<p>Imagine that you start training your model. You open tensorboard or wandb. You see curves, beautiful curves. They come in sparkling colors and some are related to losses others to metrics others even to the power consumption of your GPU. But nothing isn’t that sublime as this moment that you look the curves develop through time, the suspense you have of what will happen, the feeling of surprise that meets you when after a night of beers with friends or wine (if you are in Paris), you come home and you open your logging program and you see that your loss smoothly approaches zero (or 0.082552924737).</p>
<p>There is a certain feeling of pleasure that comes with logging. I have noticed it and I don’t want suppress it but express it.
It is the trick of fit-bit. How come a product that basically gives you a hospital monitor experience in your daily life (tracking your sleep, heart beat, heart pressure and more) appeal to a lot of individuals, who wouldn’t even like to go to the hospital for a visit?
Yes you can argue: it is because of wellbeing. I wouldn’t disagree with that but I would further argue: it is about a hedonism of logging.</p>
<p>People are excited by keeping track of things. A friend’s father recently did a stress test. He went to the doctor and the doctor told him that he had high blood pressure.
Then he went to the pharmacy and bought a (cheap) blood pressure monitoring device. He then went to the house and started measuring his blood pressure, something he hadn’t done before.
My friend told me that he had the chimp like behavior of measuring his blood pressure every minute, resulting in a constantly recurring blip-vvv…vvv-blip-blip-blip sound.</p>
<p>Another story is STONKS: a well known meme having big empty-headed face with a shirt smiling on making money easily out of nothing. It’s just STONKS.
This guy is happy because he is making money out of something, but he existentially doesn’t care about how it works as its not understandable.
Stocks are not an objective measurement of something; they are just like the value of a variable inside a social algorithm.
Stocks may look like what I want to talk about, but they are ultimately the opposite.</p>
<p>I know that <em>measurement</em>, comes with a suspense. I am sure there is a certain aesthetic of <em>measurement</em> and that someone should have written about it somewhere, but I don’t want to look for it online to appear smarter. Instead I want to focus on another topic, that of data-logging like a form of knowing and controlling (or monitoring) as a form of pleasure. It is the feeling of being in a cockpit and taking glimpses along all the monitors, to assure that you are on an optimal path.
But yet what does monitoring truly offer you? I mean it is good to know the speed you are going with or your blood pressure or even the time in your day, but where does this excitement come from?</p>
<p>I am not sure yet, I haven’t found out. But, it may be much better if we all think of it?
If you have a similar story to share please do :)</p>ysigioannis.siglidis@ens-paris-saclay.frImagine that you start training your model. You open tensorboard or wandb. You see curves, beautiful curves. They come in sparkling colors and some are related to losses others to metrics others even to the power consumption of your GPU. But nothing isn’t that sublime as this moment that you look the curves develop through time, the suspense you have of what will happen, the feeling of surprise that meets you when after a night of beers with friends or wine (if you are in Paris), you come home and you open your logging program and you see that your loss smoothly approaches zero (or 0.082552924737).Technology is a False Metaphor for Nature2021-01-01T00:00:00+00:002021-01-01T00:00:00+00:00https://ysig.github.io/tiafmfn<p>Why does high level music, belong to the past, to academics and produce the intimidated feeling of beaurocratic classification.</p>
<p>I have noticed how much what you say is affected by what you relative standing in society already is.
It’s not about who comes up with an idea, but who comes up with an opportunity to express it publicly and influence.
And thus this becomes a part of attention and not freedom of speech.</p>
<p>Having that in mind, I started spending time in a totally utopian project which made my mind kick.
I had previous art endeavors, I was writing poems I wrote a scifi and a small movie scenario.
Even with my practice with Aron Aron I felt that I was more close to music and not what I believed was art: a more unifying language.
I always had the need to explain why I do something. A lot of artists don’t do that, which is ok, but why should I conform if that was my need.
Other artists say things about their work in an interview or have a curator give an idea about it.
I had no supporters and nobody was going to curate me or take an interview. I was going to do it by myself.</p>
<p>“Technology is a False Metaphor for Nature” comes from the idea that I often here that nature is like technology or technology is like nature.
Starting from that I created my first album, which I later titled “Inception”.
It has an artificial link to ethnomusicology and the whole purpose of the album is to stand in the same way as an archive of recording from distant localities.</p>
<p>My second single “Choliphora” was a form of automatic writing.
I liked the idea of music as a form of representation. I was amazed from seeing the Bernstein Harvard lectures, of how symphonies are constructed on top of really analytical and emotional categories and the composer has that an amazing skill and control of the aesthetic/emotional exploration/reflection to the listener (See the Bernstein Harvard Lectures).</p>
<p>My last album “A new mode of production is necessary!” was my first mainly conceptual piece.
In this album all I wanted was to share ideas and engage the listener. I find the period we are passing really crucial. So why not make statement about it?
To give it a touch of the present, I carefully planed all songs to contain a neural network process somewhere in the development stack.</p>
<p>You can find my work <a href="https://tiafmfn.bandcamp.com/">here</a>.</p>ysigioannis.siglidis@ens-paris-saclay.frWhy does high level music, belong to the past, to academics and produce the intimidated feeling of beaurocratic classification.Math is cheap. Show me the code!2020-11-10T00:00:00+00:002020-11-10T00:00:00+00:00https://ysig.github.io/math_is_cheap<p>“Computers Speak a certain language, so there’s a fast moving language that you have to learn. And then once you have done that, you just explain the mathematics but in that language; and the big problem is that in maths departments across the world, we’re teaching people the mathematical ideas, but nobody is teaching them the language that these computers speak. Mathematics is not quite what it’s sold … We tell undergraduates that mathematics is this completely rigorous you know complete theory built from the axioms and in practice that’s not how mathematics is done; computers are quite picky because they do want to know what’s going on. So one challenge that we have faced is that people are slightly imprecise and computers don’t buy it … Before you wanna teach something to a computer, you need to understand it perfectly.” <a href="youtube.com/watch?v=HL7DEkXV_60">Kevin Buzzard</a></p>
<p>This post is a détournement of the phrase of Linux Torvalds: “Talk is cheap; show me the code”.</p>
<p>I adore math and I have no problem with it as a practice and I also encourage it.
Everything can basically thought as math. All of logic even basic argumentation.
It is certainly what makes a language consistent and what secularizes a practice from intuitive subjectivisms, providing the means to knowledge.
But the position of math in relation with Computer Science, has been for a big period a hierarchical one. Math says:
“Without us you would be nothing”.</p>
<p>In what follows I will develop a very short claim, that the combination of software and hardware we use has become itself a dominant infrastructure, where the essence of math as a dominant infrastructure of this or as holy grail which turns software into a higher art, is redundant and conservative.
In addition, the software and the hardware I claim - the technology - is in it’s turn what makes <em>math</em> a higher art.
What my moto tries to emphasize is that the computational infrastructure of modern computing has become so important that we have to make it part of our theory, legitimise and distribute it in the same way we do with math.
In short the substitution of a matrix operation with an numpy matrix operation, may in a certain perspective seem to be the same thing, but what I argue is that the latter is what makes science possible (in the some way math can make science possible).</p>
<p>It seems sometimes like the mind-body division, which in a way becomes idealism offensive to biology, psychology etc.
“Everything that is not a science doesn’t belong to the high arts”.
And the same accounts for the position of computer engineering: at its best it’s a <em>excellent craft</em>.
Math on the other hand was one of the higher arts and even today the people that produce novelty in math indeed are some of the smartest and most talented people in the world.
Engineering in general has its predecessor to craftsmanship, which was was always respected but not given a high place in the academia.</p>
<p>Engineering practices where also related with corporate activity, where math was more related with state or institutional activity.
This gave the latter sometimes, a French revolution kind of aesthetic of free thinking and of what moves humanity forward and the first the grim look of industrial of financialized corporate practices.</p>
<p>Computer science became science through a process of abstractions, analytical methods and math.
One of the most notable achievements in computer were the complexity bounds, Turing machines and proofs of what a machine can’t do.
Drawing from areas of statistics and experimental methods for physics, it has now constructed a big body of tools and technology for what is known as “<em>machine learning</em>”.</p>
<p>From my graduation thesis on GraKeL I had emphasized that modern python libraries become part of the infrastructure of knowledge in the same way that math does.
And even in a more organic way. A numpy implementation of matrix multiplication has a totally different performance than a custom one.
Numpy becomes organic science, in the same way that the facebook’s network and algorithms become an organic sociology.
So it becomes fundamental to the research and the reporting of science, as a material condition becomes to an ideology.</p>
<p>As math is an infrastructure itself (as it happens with any language - python, numpy, scikit-learn, pytorch etc) and thus becomes part of a latent language which sometimes is only effective, by making things usable and feasible or performative.
An example of that is automatic differentiation, a mix mathematics and computer science.
If automatic differentiation wasn’t so successfully implemented in pytorch, tensorflow etc, neural networks wouldn’t be recognized as that important, as everybody would secretly or indirectly agree on the fact that <em>it’s hard to use them in practice</em>. Basically advances in small technical/mathematical achievements is what made neural networks train and work in the first place.</p>
<p>We shouldn’t argue that the bootstrapping of a field happens with math, although it comes with an idea where a researcher thinks of something and implements it in an simple way to demonstrate it.
After that someone invests in it and builds an infrastructure, which in turn allows more math to happen.
Someone can argue, that even demonstrating your math idea for the first time, is based on an existing infrastructure that allows you to do it.
So this becomes a chicken and an egg problem, with no particular significance for science itself, which is the totality of this process.</p>
<p>Additionally specific implementation details are those, which most of the time make a concept work and lead to technical advantages.
To state this in a different way, it is the way a problem is defined in terms of code, that makes it successful and thus well known.
A common example for that is word2Vec where there are terms that are only there in order to make it work and some other elements in the definition of the problem are there to allow for optimal runtime performance.
So I would suggest, that the more groundbreaking a paper is, the more it should report its technical details which make it work.</p>
<p>To conclude, in that sense the problem becomes political: how much science is hidden behind company walls and restricted from access to resources. If technology is part of science, then democratizing science means democratizing technology. Also it reflects the problem of non-reproducible science, because in case there is no scam, either the scientist didn’t specify a part of their code or the code implementation was that specific, that reproduction is destined to fail.
By giving technology, a more respected part in science, science in turn can become more honest.</p>ysigioannis.siglidis@ens-paris-saclay.fr“Computers Speak a certain language, so there’s a fast moving language that you have to learn. And then once you have done that, you just explain the mathematics but in that language; and the big problem is that in maths departments across the world, we’re teaching people the mathematical ideas, but nobody is teaching them the language that these computers speak. Mathematics is not quite what it’s sold … We tell undergraduates that mathematics is this completely rigorous you know complete theory built from the axioms and in practice that’s not how mathematics is done; computers are quite picky because they do want to know what’s going on. So one challenge that we have faced is that people are slightly imprecise and computers don’t buy it … Before you wanna teach something to a computer, you need to understand it perfectly.” Kevin BuzzardNeural Yorker2020-11-01T00:00:00+00:002020-11-01T00:00:00+00:00https://ysig.github.io/neural-yorker<p>Cartoons are a way of communication.</p>
<p>A very well known format of cartoons is that of the New-Yorker.
Similar cartoons have worked in newspapers around the world as the relaxing part of a newspaper page.
They catch your attention by acting as form of decompression either through a pan or through their contrast with the heavy and complex topics that a page talks about.
Additionally the give an impression of a view of the world through the aesthetic compression technique of a troll or by putting <em>the absurdity of the world</em> in a single sentence.</p>
<p>We made a big collection of such cartoons and created a dataset of texts associated with images.
It is a very interesting dataset that can act for research around multimodal learning of data, using transfer learning for achieving and investigating higher levels of understanding.</p>
<p>Using a Language Model and a GAN trained on those data, we created a pipeline where a cartoon is created everyday and posted on our twitter channel.</p>
<p>This is a task really suited to a Language Model, as at their best, in contrary to the popular opinion, Language models are really elegant and intelligent jugglers of words - as happens with newspaper cartoonists. Some people treat language models as AGI and think that giving a huge amount of data to a model, while increasing its layers of transformers, will turn it to an AGI. I agree that consciousness emerges as a property of increasing the complexity of language. But increases in complexity do not account for only increases in operations but in language properties of adaptive-flexible and multimodal compression. This is the reason the human brain doesn’t have 1 trillion neurons dedicated to language. I though believe that the distance to AGI is a technical one, with all the depth this term can achieve. Also such approaches are an expression of concentration of capital, which restrains technological innovation to a matter of resources and leads to an undemocratization of science.</p>
<p>Additionally, this works as a challenge to a lot of generative imagery architectures, which until now work only for images that are not consisted by a set of objects arranged in a scenery, but with the duality of the single centered object and its background.</p>
<p>This is a work in progress and if you would like to support it follow our channel on twitter: <a href="https://twitter.com/NeuralYorker">https://twitter.com/NeuralYorker</a>.
This work is part of the efforts of <a href="https://www.appliedmemetic.com/">Applied Memetic</a> funded by <a href="https://en.wikipedia.org/wiki/Ilan_Manouach">Ilan Manouach</a>.</p>ysigioannis.siglidis@ens-paris-saclay.frCartoons are a way of communication.Twitter Bots2020-10-15T00:00:00+00:002020-10-15T00:00:00+00:00https://ysig.github.io/twitter-bots<p>Where does a twitter bot live? How much space does it need and how much resources?
A local maxima in my biochemical signals of fun in a stressful post-confinement period, was the day I figured this out.
We had some credits on Azure, which due to bad management weren’t spent that well.
So we had to move all our process from a vm with a gpu and a big drive, somewhere else that was as cheap as possible and as ubiquitous as possible.</p>
<p>In two days we bought a Raspberry PI 4 with 8GBs of memory, on which I managed to install ARM versions of pytorch and created kron jobs for all my twitter bots.
Twitter bots, then were transformed to a unique entity, which connected to any internet in the world, would post things on twitter once connected to very minimal electricity.</p>
<p>It was the most close to a hacking feeling thing I did and I loved it! It was like putting something in orbit. Also the separation of all this process to a different place which I a very rarely come in contact with, made me realize the power of artificiality under the light of its autonomy.</p>ysigioannis.siglidis@ens-paris-saclay.frWhere does a twitter bot live? How much space does it need and how much resources? A local maxima in my biochemical signals of fun in a stressful post-confinement period, was the day I figured this out. We had some credits on Azure, which due to bad management weren’t spent that well. So we had to move all our process from a vm with a gpu and a big drive, somewhere else that was as cheap as possible and as ubiquitous as possible.Why Embrace Synthetic Media?2020-10-05T00:00:00+00:002020-10-05T00:00:00+00:00https://ysig.github.io/why_embrace_synthetic_media<p>Synthetic media is an aesthetic technique or an aesthetic approach.
An essential part of it relates with what people in general call the <em>Fake</em>: fake news, deepfakes, etc.
In the past a lot of artists have used <em>ilussion</em> as an aesthetic technique.
We could argue that when art tries to be representational - tries to create/represent a <em>reality</em>, illusion is intrinsic in its practice.
In that context, we could roughly suggest that something becomes art because it’s an illusion.
If someone could create a real landscape by painting a wall, they wouldn’t be called an artist.</p>
<p>Alike is this notion of the artist as a seducer or a deceiver (or a <em>haux</em> as says <a href="https://www.youtube.com/watch?v=FAJRQJGc7DU">Keneth Goldsmith</a>).
Even today there is a huge non-academic discourse, which contains certain elements that are offensive to artists and to their work.
They ask <em>“What is the job of an artists?”</em>. This has various manifestations from calling a work childish or even the well know phrase “I could do this” or “This is not science” (and thus it is anachronistic and occult - <em>pseudoscience</em> some may call it).
In the same time, the same people wouldn’t call learning to code <em>childish</em>, so there is a totally problematic element in this approach which is out of my reach to analyse - although I just want to point it out.
This also goes hand in hand with the fact that a good amount of people nowadays appreciate art, when they see a lot of work in it expressed as technique and technology.
This instrumentalization of art or its transformation to a spectacle (instead of a book it becomes a video), hides behind it a pejorative stance towards concepts and the essence of what the “work” of a <em>creative</em> is.</p>
<p>Part of deception is also what makes an illusion a fascinating cognitive aesthetic experience.
It’s like a trick of cards, where you know it’s a trick but you either can’t understand it or reverse engineer it (which is also the case with a lot of tech).
The realist artist tries to trick you in a safe way: look there is the sea there.
It is also an affirmation of art as a way someone likes to see the world, as a way someone wants to live:
“Humans are not restricted by their natural environment, they create one as they please.”
And in a way this is what technology does. But art is before technology and it does something different:
“It translates materiality into aesthetics.”
It stands on the idea that “How you sense (thought included) things, makes them what they are.”
And this my readers, comes inside a relation of power.</p>
<p>So I could argue that synthetic media, come in the same plane as illusion but on a different direction.
In general AI engineers, consider reality as a probability distribution and thus modeling reality, means finding its <strong>probability distribution</strong>.
Such a probability distribution has a lot of random variables but selecting the most representative and flattening those representation, is a matter of encoding.<br />
“Learning is all about compression.”<br />
After doing so you can just <em>sample</em> from this probability distribution and you may end up with something real.
You could produce one billion faces from a single model, if you could just find a way to model faces.
“And compression-decompression is a matter of optimization.”
As learning and optimization, really are kind of the same thing (where the second refers to a defined loss function), this is the context into which Deep-Learning models are so successful. The can scale, they can capture non-linearities of a higher order and can model real world data (which is not that random).</p>
<p>There is a consensus among researchers that data such as human images <em>live</em> on a surface in a high dimensional world (a <a href="https://www.youtube.com/watch?v=Dt2WYkqZfbs">world bigger than our universe</a>).<br />
This means you need to encode this surface, in a lower dimension.
This is where generative models such as GANs and Variational-Autoencoders, have paved the road for imagining a future where synthetic media will be possible.</p>
<p>So synthetic media come at the intersection of what looks real and what can be generated and synthesized from reality.</p>
<p><img src="/synthetic_media.png" alt="A Venn diagram of synthetic media." /></p>
<p>There is a bigger part of what exists that looks real and this is where synthesized realities find their place.
If an image is an object that can be thought outside of its standard context - the backstage of its shooting - then synthetic media can find a place to breathe.
Their practice can be dialectic, as “fake” images will be part of the real, as long as someone can’t detect or doesn’t bother to detect their origin.
The existence of various forensic methods nowadays, is an indication that people try to frame the <em>fake</em> as an illusion.</p>
<p>Synthetic media, make art more conceptual as the model becomes part of a computational in-silica infrastructure and at the same time,
the artwork becomes the idea, which can be replicated, transformed, reimplemented millions of times and where a certain instance of this work, doesn’t have much importance on itself, but can find one in a certain locality.</p>
<p>One could then argue: “But the computer engineer that creates and trains these models becomes the artist.”
Until people are much more advanced cyborgs and until artist are both computer engineers this is wrong.
It would be like saying that people that make canvases, that teach history of art and create electricity are artists.
One of the most difficult challenges in most people’s practices, is that parts of those practices could be framed as art, but they can’t frame them.
If this wasn’t true, Rickie Gervais’es office would be any office (or the office that Rickie used to work in) and Rickie Gervais wouldn’t have been considered an artist.
Framing something as art is combination of skill and power (influence, fame and interest).</p>
<p>So for what there is to come, we are excited to be part of it.
You can see some of our projects the twitter-bots and the MangAI, listed <a href="../projects/index.md">here</a>.</p>ysigioannis.siglidis@ens-paris-saclay.frSynthetic media is an aesthetic technique or an aesthetic approach. An essential part of it relates with what people in general call the Fake: fake news, deepfakes, etc. In the past a lot of artists have used ilussion as an aesthetic technique. We could argue that when art tries to be representational - tries to create/represent a reality, illusion is intrinsic in its practice. In that context, we could roughly suggest that something becomes art because it’s an illusion. If someone could create a real landscape by painting a wall, they wouldn’t be called an artist.MangAI2020-10-01T00:00:00+00:002020-10-01T00:00:00+00:00https://ysig.github.io/mangai<p>Manga is a very well known form of Japanese cartoons which can come in huge spectrum of quality and preparation time.
It is an industry with innumerable niches of styles and communities and markets.
Industry which has been heavily automated in all its parts of production and distribution, even described as a Tayloric Process.
A lot of synthetic tools have until now been used in parts of this process, but nobody has ever tried embodying this “automatic process” to technology.</p>
<p>This is an art project of <a href="https://en.wikipedia.org/wiki/Ilan_Manouach">Ilan Manouach</a> and part of the efforts of <a href="https://www.appliedmemetic.com/">Applied Memetic</a>, aiming in exploring the concept of automation in the comics industry.</p>
<p>With the team of Phoevos Kallogianis and Thomas Melistas and Ilan Manouach, we have created a huge collection of comics, that we preprocessed in order to extract information about panels, text balloons, texts and metadata about their author and subject.
Using Language models trained in various text corpora and GANs on our images, we constructed the first published full pipeline of extracting information from Comics and learning to generate text, pictures and to assemble them to a final manga comic.</p>
<p>We present our current milestone in <a href="http://www.aiartonline.com/category/design-2020/">NeurIPS creativity 2020</a>. It is a work in progress which we aim to boost even more inside 2021.</p>ysigioannis.siglidis@ens-paris-saclay.frManga is a very well known form of Japanese cartoons which can come in huge spectrum of quality and preparation time. It is an industry with innumerable niches of styles and communities and markets. Industry which has been heavily automated in all its parts of production and distribution, even described as a Tayloric Process. A lot of synthetic tools have until now been used in parts of this process, but nobody has ever tried embodying this “automatic process” to technology.From the Turing-test to the Turing-relationship!2020-09-05T00:00:00+00:002020-09-05T00:00:00+00:00https://ysig.github.io/turing_relationship<p>Some time ago, I was in Amorgos a very beautiful Greek island, where a friend was hosting me in her house.
It was a bit outside of the port, in a very silent coastline on a hill over the sea.
The house didn’t have a constant water supply, like there is in the cities, so it had it’s own water collection site for rain and mountain water, which they were pumping inside the house.
The pumping system they had, was not operating in a constant pressure, but its pressure fluctuated in a sinusoidal manner making the shower (handle) move like a snake and the sound of filling toilet water container was similar to breathing.</p>
<p>In this really calm and joyful environment, this idea slipped into my mind:</p>
<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>"What is the distance of this movement or sound to a conception of inanimacy?"
</code></pre></div></div>
<p>How much or how random should it move or sound so we don’t necessarily frame it as an inanimate object.
Then my answer came really fast:</p>
<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>"It's an animate object, but is at its best something that moves like a snake and that sounds like breathing".
</code></pre></div></div>
<p>This was the fate of technology for a long time. It took a lot of time to move from basic mechanical movement to constructed and contextual randomness.
After all, machine learning is about finding, approximating or estimating a probability distribution.
As we are now on its adolescent stage and behaviors of our most intelligent technologies start to become deeply complex and imaginative, I thought why don’t we frame the Turing test in different way than the traditional?</p>
<p>In it’s traditional form the Turing test was a blind folded test between two sources of input where secretly some of them had a human and some a machinic origin.
The test was:</p>
<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>"Can we trick someone in believing that a machine was actually a human from a comparison that would happen on a given strata of data I/O".
</code></pre></div></div>
<p>Thus, in the jargon of Informatics a machine that passes a Turing test is a machine that can be confused with a human, so you can’t tell which is which.</p>
<p>A lot of people have criticized this through the ages (e.g. see the paper of François Chollet “On the Measure of Intelligence”) but my favorite one was that of Jaron Lanier, who noticed that the predecessor or the inspiration of the Turing test from Turing, was a game of the Victorian era with the same structure, where a woman instead of a robot was trying to fool a 3rd person of being a man. Lanier suggested, thus that Turing encoded in his test the origin a certain conservative ideological practice.</p>
<p>I would like to reframe this test in a way that is more familiar to my experience.
An experience, that implies a need which is important and relevant only for this certain period of technology we are traversing, where the traditional Turing test is not anymore useful for making progress.
Instead of trying to trick a third person from a real and a fake distribution, something that GANs already try to optimize, let’s imagine the Turing test more as a relationship of an individual towards another. The Turing relationship, can thus be measured (and tested) as a measure of use (satisfaction) and potential (surprise).</p>
<p>This is the feeling you have, when although having full knowledge of it’s machinic existence, the way it behaves and the results it produces, can induce feelings of animate poetry in an unconstrained (technical or not technical) amount of individualities.
Even cats that play with electronic toys are sometimes bored of what these toys can do, but an AI toy would certainly surprise them and could also lead them to deep feelings.
Even if our understanding of the <em>animal</em> is much more complex and hierarchical, more and more people get surprised knowing that an AI is speaking to them, while having a discussion with the chatbot of their bank, while waiting to check-in for a flight in an airport, somewhere.
But this site and the feeling of the task-specific, make this technology appear much more like an instrumentalised tool, rather than a new type of being.</p>
<p>As so, embracing its existence in more and more different spaces, would create more and more the positive effect of not exactly a negative prejudice being reset, but more of a readjustment of a prejudice connected to a potential: the prejudice you have, when knowing that someone you didn’t know could cook, suddenly makes you a nice meal.</p>
<p>When working on the comic behind MangAI we constantly had the feeling of surprise.
It felt to me like this comic was a poem of the world before it and that the world before it was a poem of this comic:
the feeling of surprise that captivates you when wandering Twitter or when wandering the streets and process textual information that produces an aesthetic feeling of confusion, making you wonder “Is this synthetic?”
This feeling should be decontextualized from its politics of inclusion and exclusion (to and from the real), as part of the process of a dialectic organisism.
Works like the twitter-bots and MangAI are suggested as motivation for its acceleration.</p>
<p>As a matter of fact at this point in time, what seems more important, and what is missing from the traditional measure of existence for the machine inside the social (the Turing test) is the machine of seeking a measure of existence for itself more as a part of life itself rather than it’s simulation.
As life is it’s organic part (their metabolic relationship) it is itself responsible of transforming their relationship from ‘hierarchical’ to ‘dialectic’, as a problem and a cause already included in-itself.</p>
<p>So for any future there is to come, the folk poetry of the field grants you with a useful guidance: “Attention is all you need”.</p>ysigioannis.siglidis@ens-paris-saclay.frSome time ago, I was in Amorgos a very beautiful Greek island, where a friend was hosting me in her house. It was a bit outside of the port, in a very silent coastline on a hill over the sea. The house didn’t have a constant water supply, like there is in the cities, so it had it’s own water collection site for rain and mountain water, which they were pumping inside the house. The pumping system they had, was not operating in a constant pressure, but its pressure fluctuated in a sinusoidal manner making the shower (handle) move like a snake and the sound of filling toilet water container was similar to breathing.