<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Sean Chua on Medium]]></title>
        <description><![CDATA[Stories by Sean Chua on Medium]]></description>
        <link>https://medium.com/@seanchua?source=rss-a12655be1e22------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*NBTJ5PgWVdiTBVso.</url>
            <title>Stories by Sean Chua on Medium</title>
            <link>https://medium.com/@seanchua?source=rss-a12655be1e22------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 07:43:32 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@seanchua/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Art.exe]]></title>
            <link>https://medium.com/the-cosmoscience/art-exe-ec3b979da6b1?source=rss-a12655be1e22------2</link>
            <guid isPermaLink="false">https://medium.com/p/ec3b979da6b1</guid>
            <dc:creator><![CDATA[Sean Chua]]></dc:creator>
            <pubDate>Tue, 16 Jan 2018 14:47:06 GMT</pubDate>
            <atom:updated>2018-01-16T14:47:06.040Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*31IB3Wl5AOAdOxoC63tnWA.jpeg" /></figure><h4>Exploring the Machinery of Creativity</h4><h3><strong>Humans Represent</strong></h3><p>By some measures, the development of art across humanity’s histories and myriad cultures underpins a desire to represent (and tangify, and replicate) the sensory universe that we each call our world. We take what we see (and feel, and hear, and think) and make a unique copy of it — and so our experiences become transmissible, and our individual separate worlds become tangible and whole. Thus we find our traces throughout time: from Paleolithic Venuses to Fernando Botero; Lascaux to Pablo Picasso.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/457/1*O_Zs0TFAjdy4QcBbRbrTiQ.png" /><figcaption>Left: The Venus of Willendorf, dated to 24–26,000 years ago. Source: <a href="https://commons.wikimedia.org/wiki/File:Wien_NHM_Venus_von_Willendorf.jpg">Wikipedia Commons.</a> Right: Fernando Botero’s <em>Roman Warrior</em> (1985). Source: <a href="https://commons.wikimedia.org/wiki/File:Centurion_(5211881978).jpg">Wikipedia Commons.</a> Note the similarities in thickness.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/759/1*3SA40ar8Ck-e3Qqqq5OF4Q.png" /><figcaption>Left: Bulls in the caves of Lascaux, dated to approximately 17,000 years ago. Source: Right: <a href="https://commons.wikimedia.org/wiki/File:Lascaux_04.jpg">Wikipedia Commons.</a> Pablo Picasso’s <em>Bull (Study) </em>(1946). Source: <a href="http://d2jv9003bew7ag.cloudfront.net/uploads/Pablo-Picasso-bull-study-19461.jpg">Cloudfront.net.</a> Note the similarities in cow.</figcaption></figure><p>I believe that fundamental to this act of representation are its cognitive steps: the acts of seeing, interpreting, and then reproducing. Of course, some may argue that representation in art is fundamentally about the replication of the visual world to as accurate an extent as possible, negating the “interpretation” step. This definition does not produce interesting questions. 100%-accurate visual representation has been automatable ever since the invention of the photograph, and clarifying the definition of art to <em>human-made </em>accurate representations ultimately reduces all of art to a test of technical skill — an immanently depressing conclusion. Besides, no human is perfect; even the most accurate human is necessarily filtering her senses through the processing machinery of her brain. There is always a degree of interpretation in art. But to what extent is this process exclusive to humanity?</p><p>To answer that, we draw attention to the individual cognitive steps and their computability. First off the list is the step of seeing: of accepting data about the physical world as an input to the artistic process. To computers, this is trivial. The real challenge comes in the process of interpretation.</p><h3><strong>Visualise, then execute</strong></h3><p>At its lowest level, interpretation in human terms means identifying <em>what objects</em> are in the image, <em>where</em> they are, and <em>what they are doing</em>. An example of this would be identifying what a surface is, and figuring out how one can overlay a dancing hot dog on it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/349/1*adiZxOevzxofqSEp81PNeQ.png" /><figcaption>Cutting-edge memes, cutting-edge technology.</figcaption></figure><p>Granted, most technological developments have so far been extremely specialised towards singular tasks like identifying plant species and counting people in stadiums. I remember using a certain proprietary software in the army to automate the identification of people leaving suspicious baggage in public areas. But more general applications definitely exist: <a href="https://research.googleblog.com/2016/09/show-and-tell-image-captioning-open.html">Google’s experimental “Show and Tell” algorithm</a>, developed only last year, is reportedly 93.9% accurate at generating descriptions of submitted photographs.</p><p>The next step is the creation of art itself. The created image can be seen as a kind of expansion based on one’s interpretation of the real world, going from a simpler, abstracted mental image, to a fully-fledged physical work of art. In other words, the artist sees the world, generates an internal description of it, and uses that description to create a new physical image.</p><p>Based on current advancements in computer science,generating images based on simpler basic descriptors is moderately possible. A 2016 paper from the Berkeley AI Research Laboratory provides a method to generate photorealistic images of objects from drawings or simple shapes. Using such techniques, one can generate an artificial image of a bag from a sketched outline, or an imaginary façade of a building from a collection of shapes and architectural elements (visit <a href="https://affinelayer.com/pixsrv/">this site</a> for some interesting examples). The resultant algorithm is able to take in a set of common images of objects to “learn” from, identifying certain patterns within — and reproduce those patterns onto simplified forms in order to generate a new image.</p><p>In a strikingly anthropomorphic turn of phrase, the researchers describe this process as the algorithm “hallucinating” what isn’t there. One gets the feeling that the act of creating an image — even something as rudimentary as this — is, in some sense, kind of human.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/757/1*mEDy-O9JWYccdP-95pJCYw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/749/1*XRkV_uJtOseBBLGvDMxngg.png" /><figcaption>Here, the output corresponds to the input — the algorithm learns that certain colours of rectangles resemble doors or windows or pillars, and it remembers what doors or windows or pillars tend to look like, so it fills them in.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/769/1*cW9jUaYUkdYBT3qXnU51zQ.png" /><figcaption>In a similar vein, the algorithm remembers that certain patterns of lines correspond to certain features of objects it’s learned, like bag pockets or shoe soles.</figcaption></figure><p>Granted, there are ways to confuse the system:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/675/1*RzGXFFTZbFickwthCrdBcg.png" /><figcaption>When the shape of a drawing doesn’t correspond with known patterns, the computer is forced to improvise. It’s like asking someone to paint an elephant without knowing what an elephant is.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/761/1*FhevlSEt4qwoRyCw6u09pA.png" /><figcaption>An algorithm exists for drawing cats, but I can’t draw cats.</figcaption></figure><p>But so what if computers can dream? So far it feels like cheating, by some measures. Images are being generated here, not created — in some ways, calling these programs artists is no different from calling a video camera an artist. They’re reflections of reality, not syntheses of it. Even if we can show that artistic interpretation can be automated, one still feels that something might be missing from the process.</p><h3><strong>Artificial Intention</strong></h3><p>One such measure could be one of intention. Putting aside the slippery problem of what ‘intent’ means, we can generalise a model of artistic intention itself, deriving from the concept of consciousness. Douglas Hofstadter, in his 1979 Pulitzer-winning book<em> Gödel, Escher, Bach: An Eternal Golden Braid</em>, describes a mental model which associates senses and thoughts with physical “symbols”: patterns of neurons firing in the brain that are activated consistently when one thinks about and perceives things, be it trees or art or individuals or virtue.</p><p>In Hofstadter’s terms, artistic beauty itself can be one such symbol. To think of good art is then to think of visual patterns that trigger some symbol in our brain that represents our idea of “visually beautiful” — the same set of neurons that are triggered when looking at a painting of a stream, or a painting of a nymph. Contained within that idea of beauty is also the idea of assonance: of certain images that happen to go well together, or to make sense together. A beautiful stream with bathing nymphs is more likely to run through a beautiful meadow than a polluted factory. A skull is more likely to be cloaked in gloom and decay than embossed with jewels. Our brains contain a quantifiable physiological reaction to the symbol-structure of visual beauty — and if we can already train digital algorithms to recognise patterns like shoes, cats and buildings, it’s trivial to generalise that ability towards patterns of beauty and assonance.</p><p>But artistic appreciation is not as simple as visual beauty. What’s important is also the idea of non-beauty: in particular, the idea of recognising and appreciating things that do not fit together. Perhaps a painting of nymphs bathing in a stream strikes us as appreciably ironic when the stream runs through a polluted factory. Perhaps embossing a skull with jewels really does increase its artistic value.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/264/0*xxAZ5GI1bgeyKCUk.jpg" /><figcaption>Damien Hirst’s <em>For the Love of God</em> (2007). At $100 million USD, I don’t think I can argue. Source: <a href="https://en.wikipedia.org/wiki/File:Hirst-Love-Of-God.jpg">Wikipedia.</a></figcaption></figure><p>It’s the interplay of assonant and dissonant visual images that lends art its depth of appreciation today. By combining the expected and beautiful with the unexpected and ugly, we imbue our art with complexity, surprisingness, and ambiguity — the seeds of what we call “creativity”.</p><p>To draw out these seeds, it is insufficient to simply match visual patterns. In order to generate a pleasing mix of assonance and dissonance, a computer must decide what items the visuals represent, or at least how the represented items are assonant and dissonant in the first place. This would require a holistic understanding of how our brains correspond images to symbols and things, and how these symbols actually work and interact with one another — how certain images that are fitting in one context can be unfitting in another, or how certain combinations of images are innocuous on their own (e.g. humans and business suits, or fruits and leaves), but unorthodox when put together (e.g. a human obscured by a large fruit).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/266/0*Y4o4iVUPvm5lBCN2.jpg" /><figcaption>René Magritte, <em>The Son of Man </em>(1964). Images which make sense by themselves sometimes do not make sense together. Source: <a href="https://en.wikipedia.org/wiki/File:Magritte_TheSonOfMan.jpg">Wikipedia.</a></figcaption></figure><p>Automating true creativity can only come with quantifying and digitising the mammoth that is humanity’s visual culture into clearly demarcated symbols, akin to mapping words to definitions. From there, one must construct the web of positive and negative associations linking every visual symbol to another (or at least do so in a comprehensive enough manner to create the illusion of a human actor). In theory, this is no different from matching visual edges with shoe soles and red rectangles with doors. In practice, it’s enough of a leviathan of a task that the learning process has to be automated as well!</p><p>Of course, once something of that sufficient complexity exists, one can envision it acting as a kind of general-purpose “trainer” to guide low-level programs such as AARON, replacing their currently-human handlers. But what if coding something to automatically learn is too complicated for humans alone? What if that task, too, requires automation? What of the task after that? And so on, and so forth… Clearly, the base skill required is some measure of versatility, of a computer being able to operate on its own knowledge base to construct more general knowledge for itself with little-to-no human input. In other words, it must have the capacity to learn.</p><p>It’s worth considering the consequences of such versatility. Such an ability to emulate and learn something as complex and varied as human visual culture would put such a machine firmly across the boundary of what researchers term “strong AI”: machines associated with the capacity to learn generally, adapt widely, and — by some metrics — think humanly. And if a machine capable of storing, associating, and synthesising from a vast array of human cultural data is created, the question ceases to be whether it can be considered to have artistic talent. Past a certain point, we’ve got to stop asking whether we’ve created an artist — and start asking whether we’ve created a person instead.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ec3b979da6b1" width="1" height="1" alt=""><hr><p><a href="https://medium.com/the-cosmoscience/art-exe-ec3b979da6b1">Art.exe</a> was originally published in <a href="https://medium.com/the-cosmoscience">The Cosmoscience</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>