[-empyre-] A open call to our subscribers: Week 1: New Year/New Tools and Technologies
B. Bogart
ben at ekran.org
Sat Feb 7 12:51:55 AEDT 2015
Hello Murat,
I agree with you that computer vision (and all computer programs) are
value-laden. I'm not trying to make an argument that they are unbiased,
as I believe the opposite is true. I think the use of statistics and
'objective' measures makes computer vision *appear* to be objective, but
as it is embedded in a particular use context (find people, find
expressions, find genders, etc.) that gives it a point of view. Each
use-case could be considered a different point of view, and thus
considering computer vision as a single thing (as implied in Grosser's
work) is missing this aspect. There is always the ongoing problem that
technologies are composed of other technologies.
The power relations and values around not only the deployment (but also
development) of computer vision certainly impacts the social function of
these tools. From my personal experience few researchers in this area
reflect on these issues of bias and imbued values in these technologies
which are often treated as neutral.
In Dreaming Machine #3, there is always this tension of surveillance.
Unlike in most uses of computer vision (for interactivity for example),
the artwork's point of view (in the sense of use-case above) has no
particular interest in people or faces. That being said, the underlying
algorithms (like the segmentation) is still based on the assumption that
things can be broken up into segments, and that discontinuity in colour
implies a difference in object. This is very different than human
perception where we project structural boundaries based on movement and
interactions with objects. Objects to the system are just blobs of
pixels, whereas objects to a human are concepts that are associated with
blobs of colour.
Even the point of view (in the literal sense) of the camera itself
changes the reading of the images produced. I attempt to give the camera
a human eye-level view where those being viewed can also see what the
system's interpretation of the camera image is, but this is not always
possible. Due to limitations of sites, the camera most often sees a wide
city-scape and individual people are actually too small to be seen as
independent of their context. Some examples:
http://www.ekran.org/ben/wp/2014/isea-2014-in-dubai/
http://www.ekran.org/ben/wp/2015/dreaming-machine-3-test-with-tower-view/
Blade Runner and Metropolis are both planned for "Watching and Dreaming".
On 15-02-06 07:57 AM, Murat Nemet-Nejat wrote:
> ----------empyre- soft-skinned space----------------------
>
>
>
> "Many computer vision systems are oriented to recognizing faces or
> numbers, or
> even genders and expressions. They are intentionally developed for these
> specific purposes that cannot be excised from the contexts of
> surveillance and tracking, they tell us specific information that
> appears to be objective and unbiased: A woman in white stands in front
> of two men and a green background.... In Grosser's work, the algorithms
> used appear to be tracking the
> movement of objects or perhaps whole scenes, tracing lines over a a
> blank canvas. There is a tension in provoking the viewer to consider a
> monolithic sense of Computer Vision (developed and funded largely for
> the purpose of tracking and surveillance)"
>
> Ben, as far as I am able to follow it, your analysis is creating a
> dicathomy which is not there: "A woman in white stands in front of two
> men and a green background". Colors may have an "objective" reality that
> can be measured in wave lengths, etc.; but colors also have a potent
> emotional, i.e. cultural, context. To the extent that the programing of
> the algorithms is done by humans, how can one say that its observations
> are not touched by values? That brings me to the issue of surveillance.
> Isn't surveillance a totally cultural/political act? What is the "use"
> of surveillance if it ignores the emotional condition and cultural
> relations of the "subject"? One surveils subjects (not objects). In
> other words, implicitly it surveils subjectivities.
>
> I would rather claim that computers (including artificial intelligence)
> constitute a radical, transforming extension of subjectivity, from the
> "individual" to a wider field--as much a march towards a solipsist humanity.
>
> What about the film "Blade Runner"?
>
> Ciao,
> Murat
>
>
> On Wed, Feb 4, 2015 at 6:30 PM, B. Bogart <ben at ekran.org
> <mailto:ben at ekran.org>> wrote:
>
> ----------empyre- soft-skinned space----------------------
> Hello Empyrians,
>
> I'll break this comment into three sections, each inspired by a
> different aspect of Renate's opening message.
>
> *Cycles*
>
> It's easy to think of progress as a ray projecting to an ever better
> future. I find it interesting that so many 'innovations' happen in
> cycles, even in a world apparently based on incremental improvement. The
> most salient for me, at this point, is the new trend of VR. This stands
> out because of the hype that surrounded it in the mid 90s when I was and
> centrally concerned with "new media" and computer graphics. I
> experienced a VR system at SIGGRAPH in 1995, and did not try it again
> until a recent Oculus Rift demo. The nearly 20 year gap gave lead me to
> some quite high expectations. I was left being totally underwhelmed. I'm
> not convinced that this new VR will be any more successful than it was
> in the 90s. More details on my experience here:
> https://plus.google.com/104335933112525967488/posts/DxeZLjBFu9N
>
> Even computer architecture has slid back and forth between single and
> multi processor systems over the years. I think of tradition as a way of
> thinking about the world that emphasizes what has stayed the same.
> Novelty is a way of thinking about the world that emphasizes what has
> changed. Realism is somewhere in between, where we accept that at some
> level of description nothing changes, but at other levels of description
> nothing stays the same. More and more, I find myself not seeing through
> the lens of novelty, but through the lens of tradition.
>
> *Networks*
>
> Networked and telematic work was very popular when I started my B.F.A.
> in 1999. I can't articulate exactly why, but it held little interest to
> me. I even worked in a research lab at Ryerson University where we made
> almost entirely telematic events and performances as part of the MARCEL
> network (http://www.mmmarcel.org/). While I did make art in that
> collaborative paid context, my individual work was quite a bit less
> about the network. Rather, less about interactivity across networks. In
> Aporia (http://www.ekran.org/ben/wp/2007/aporia-2001/), I considered the
> network as a sea of signifiers that could never be fully grounded.
>
> I think it's this sense of grounding that made telematic art less
> attractive to me. Perhaps it's the ephemeral quality of 'new media'
> itself that drove my interest in here and now. Through my graduate
> studies, I have let go of interactivity entirely and am focused on
> generative processes in the Context Machines series
> (http://www.ekran.org/ben/wp/wp-content/uploads/2013/09/Context-Machines_A-Series-of-Situated-and-Self-Organizing-Artworks.pdf).
> Generative processes are used in the context of grounding and artworks
> are considered site-specific and augmented by sensors in the
> environment. The latest incarnations of Context Machines are the
> Dreaming Machines whose processes are inspired by biopsychological
> conceptions of perception, mental imagery, mind wandering and dreaming.
> In essence, these works are trying to ground themselves by learning
> about the world (using machine learning methods) and generating
> simulations of it. I have been asked about networked versions of these
> installations that are served images from flikr or the like, rather than
> images from a single place and a particular time. There is something in
> the lack of constraint or bounding that makes such proposals quite
> unpalatable to me.
>
> Context machines start with a point of view. They see the world through
> a single and unified lens. They are context dependent and inherently
> subjective. There is no sense of Truth or objectivity, they are in a
> constant process of trying to understand the world, without ever being
> able to perfectly reproduce it. They are single nodes and have no
> culture or communication. They reproduce not what they see in the world,
> but what they understand of it, which can be very little.
>
> A new project titled Watching and Dreaming
> (http://www.ekran.org/ben/wp/2014/watching-and-dreaming-2001-a-space-odyssey-2014/)
> moves away from this sense of place in space and time where the work is
> subjected to a cultural world, rather than a physical one. This is the
> world represented in cinema. It's a highly constructed discontinuous
> world meant to be interpreted by a human viewer. By the sheer fact that
> a film has a fixed length, it is more bounded than a live context, but
> it's also a challenging world to learn from. The learning mechanisms of
> Watching and Dreaming exist in the context of Artificial Intelligence,
> and the choice of film (currently only 2001: A Space Odyssey) is focused
> on cultural representations of artificially intelligent systems, in this
> case the homicidal HAL9000. My 'intelligent' machine watches a film
> about a fictitious intelligent machine, all the while trying to make
> sense of what it means while never being able.
>
> *Tools*
>
> On the surface, Watching and Dreaming seems a lot like Grosser's
> "Computers Watching Movies"
> (http://bengrosser.com/projects/computers-watching-movies/), where
> computer vision methods are used to interpret cinematic images. Grosser
> frames the work as an attempt to provoke viewers into asking "how
> computer vision differs from their own human vision, and what that
> difference reveals about our culturally-developed ways of looking." This
> seems to imply that computer vision is objective and unified. In truth,
> computer vision is really just the application of machine learning to
> visual images, and thus has much diversity in how it sees and why. Many
> computer vision systems are oriented to recognizing faces or numbers, or
> even genders and expressions. They are intentionally developed for these
> specific purposes that cannot be excised from the contexts of
> surveillance and tracking, they tell us specific information that
> appears to be objective and unbiased: A woman in white stands in front
> of two men and a green background.
>
> In Grosser's work, the algorithms used appear to be tracking the
> movement of objects or perhaps whole scenes, tracing lines over a a
> blank canvas. There is a tension in provoking the viewer to consider a
> monolithic sense of Computer Vision (developed and funded largely for
> the purpose of tracking and surveillance) in comparison with human
> vision (oriented towards making sense of the world to facilitate
> survival).
>
> In Watching and Dreaming, the specific methods used include segmentation
> (breaking images into pieces), clustering (recognizing that a piece in
> the previous image is the same as 'object' as the current image) and
> prediction (learning the sequence in which these 'objects' appear).
> These methods serve to help the machine generate images that are
> grounded in reality, and yet are not a perfect reflection, nor a random
> composition. The "task" is to make sense of the world by recognizing and
> predicting perceptual 'objects'. These methods are all on-line and
> unsupervised, i.e. they don't make use of previously recorded
> information nor depend on a 'right answer' provided by an expert to help
> them learn.
>
> My point here is that computer vision is no more monolithic than
> artificial intelligence. There are many different ways in which we could
> define, frame and validate both vision and intelligence. It's temping to
> think of these technologies as monolithic "tools" but they are really
> aggregations of many different conceptions and approaches linked
> together. Each of these approaches imbues its own bias and point of
> view. To a segmentation algorithm, every image is a set of objects that
> can be separated, and borders are often made in places where a human
> would not recognize them. To a clustering algorithm, the world is made
> up of discreet elements where belonging to a group is more important
> than 'minor' individual difference. The way Computers Watch Movies
> "sees" depends on a conception of foreground and background, between
> what moves and what does not.
>
> A gender detector makes the base assumption that the gender binary is
> real and objective, and thus ignores the "noise" of androgynous,
> transgendered, intersex and etc. individuals.
>
> The problem with the biases embedded in our tools is that the use of the
> tools enforce those biases
> (http://www.unwomen.org/en/news/stories/2013/10/women-should-ads).
> Interestingly, google has changed things and thus typing "women should"
> no longer produces additional suggestions.
>
> Is it a question a tools, or a question of contexts and conceptions?
> What ways of thinking about the world are useful? How do they betray our
> implicit biases and prejudice? How can we critically reflect on them
> when they are so readily internalized into our minds and cultures?
>
> Ben Bogart
> Hons. B.F.A., M.Sc, Ph.D.
> www.ekran.org <http://www.ekran.org>
>
> On 15-02-03 05:41 PM, Renate Terese Ferro wrote:
> > ----------empyre- soft-skinned space----------------------
> > We are hoping that all of our subscribers, both active
> participants and
> > lurkers, will post a couple of times this month to share
> > your current work and what tools and or technologies you are using
> to that
> > end. Some of you may be developing new apps or platforms where
> others of
> > you are pushing the boundaries of other tools that are more
> ubiquitous.
> >
> > Just this past week Google¹s CEO Eric Schmidt spoke at the world
> economic
> > forum in Switzerland exclaiming that ³The Internet is Dead.² Schmidt¹s
> > call was simply the assertion that the internet is so ubiquitous
> because
> > of the nature of networked things. Interfaces are and will become so
> > naturally integrated into our architectural and living environments
> > Schmidt stated. The critical question remains how will the rights and
> > privacy of individuals become impacted.
> >
> >
> http://www.makeuseof.com/tag/google-internet-disappear-future-cyanogen-tech
> > -news-digest/
> >
> >
> > Ironically it was only seven years ago in 2008 that WIRED magazine
> > published an article entitled ³The Web Is Dead. Long Live the
> Internet.²
> > http://www.wired.com/2010/08/ff_webrip/all/
> >
> > A move away from the open search of the web to more enclosed and
> > prescriptive apps, music and movie services all driven by the ease and
> > impact of i-phone technology has enabled the web to become less
> important
> > asserted WIRED authors Chris Anderson and
> > Michael Wolff when they wrote: ³Sure, we¹ll always have Web pages. We
> > still have postcards and telegrams, don¹t we? But the centerOf
> interactive
> > media ‹ increasingly, the center of gravity of all media ‹is
> moving to a
> > post-HTML environment,² we promised nearly a decade and half ago.
> >
> > "The examples of the time were a bit silly ‹ a ³3-D furry-muckers VR
> > space² and ³headlines sent to a pager² ‹ but the point was altogether
> > prescient: a glimpse of the machine-to-machine future that would
> be less
> > about browsing and more about getting.²
> >
> > The internet has experienced dramatic changes since the beginning
> of its
> > inception as is the way of any of the technologies we have used in the
> > past or are currently using. Looking forward to hearing you. It
> is a new
> > year. Share you tools and technologies.
> > Renate
> >
> >
> > Renate Ferro
> > Visiting Assistant Professor of Art,Cornell University
> > Department of Art, Tjaden Hall Office: 306
> > Ithaca, NY 14853
> > Email: <rferro at cornell.edu <mailto:rferro at cornell.edu>
> <mailto:rtf9 at cornell.edu <mailto:rtf9 at cornell.edu>>>
> > URL: http://www.renateferro.net <http://www.renateferro.net/>
> > http://www.privatesecretspubliclies.net
> > <http://www.privatesecretspubliclies.net/>
> > Lab: http://www.tinkerfactory.net <http://www.tinkerfactory.net/>
> >
> > Managing Co-moderator of -empyre- soft skinned space
> > http://empyre.library.cornell.edu/
> >
> >
> >
> >
> >
> >
> >
> >
> > _______________________________________________
> > empyre forum
> > empyre at lists.artdesign.unsw.edu.au
> <mailto:empyre at lists.artdesign.unsw.edu.au>
> > http://empyre.library.cornell.edu
> >
> _______________________________________________
> empyre forum
> empyre at lists.artdesign.unsw.edu.au
> <mailto:empyre at lists.artdesign.unsw.edu.au>
> http://empyre.library.cornell.edu
>
>
>
>
> _______________________________________________
> empyre forum
> empyre at lists.artdesign.unsw.edu.au
> http://empyre.library.cornell.edu
>
More information about the empyre
mailing list