AI Delivered: the Abject and Redemption (Part 1)

Editor's note:

This text is a keynote address delivered by ZHANG Ga on May 20, 2021 at aai International Conference on art (ai) organized by Tongji University . It serves as a curatorial annotation to the exhibition AI Delivered. 

First a quick note: I received this conference invitation while I was in the middle of preparing for a forthcoming exhibition at Chronus Art Center, and I appreciate the opportunity to speak on the subject. Titled AI Delivered: the Abject and Redemption, the two-part exhibition, which is also the title of this talk, attempts to raise questions on the epistemological limits of Artificial Intelligence. The exhibition implicitly aims to problematize unfettered optimism about AI and its instrumental utility propagated by corporate interests and geopolitical powers and, at the same time, to point out how AI is envisioned by artists to explore a cosmopolitically conscious ecology and the posthuman prospects of symbiosis and of collective commons. In conceiving the exhibition, I was motivated to bring to public awareness other artistic practices that work outside of the screen-based, image-centric paradigm which often reduces AI to a pre-scripted visual trope on the one hand, while aligning with current AI’s business prospects on the other. Joanna Zylinska’s characterization of AI-driven art in her 2020 book AI Art: Machine Vision and Warped Dreams well captures this prevailing consumption of “AI Art” where she argues that “Much of what passes for AI-driven art, especially of the industry-sponsored variety, remains quite superficial, even if visually captivating. The projects that gain most public attention are those that embrace AI rather instrumentally, with aesthetics reduced to things looking ‘beautiful’, i.e. symmetrical, mesmerizing, garish, and, first of all, similar to what already exists.” ¹ She continues rather unsparingly:

Nevertheless, I would like to explore a defense in the name of AI. Artificial Intelligence often seems to be a surrogate or scapegoat for human hubris and embarrassment.

We tend to use the human as the yardstick to measure things. Heidegger once bluntly said that was the epiphany of modernity: “[t]hat period we call modern…is defined by the fact that man becomes the center and measure of beings. Man is what lies at the bottom of all beings, that is, in modern terms, at the bottom of all objectification and representability.”³ Media theorist Eugene Thacker further debunks the myth of life and its seemingly indisputable authenticity of human incarnation, “Life is projected from subject to object, self to world, and human to nonhuman.” ⁴

We inflict a prognosis of human intelligence onto machines, and therefore render Artificial Intelligence a vision both sublimely grandiose and abjectly undeliverable. We are aware that science has yet to solve the riddleof human intelligence, nor has understood the inner workings of the human brain. As professor of AI and Cognitive Science Brian Cantwell Smith contests: “For one thing, it would be premature to assume that what matters for our brain’s epistemic power is our general neural configuration. That is all that current[AI] architectures mimic.” ⁵ Philosopher John Searle has argued that “any attempt literally to create intentionality artificially (strong AI) could not succeed just by designing programs but would have to duplicate the causal powers of the human brain.” ⁶

Yet, perhaps the forerunners of modern computing machines might have had something else in mind when they thought about machine intelligence in the image of a thinking machine. When answering the question “Can Machines Think?” Alan Turing in his 1950 essay “Computing Machinery and Intelligence” proposed his infamous Imitation Game (aka TheTuring Test) as a counterargument to his own self-imposed question, writing “The original question, ‘Can machines think?’ I believe to be too meaningless to deserve discussion.” Turing said instead “that in about fifty years' time it will be possible, to program computers, with a storage capacity of about 109 (10 to the 9th power), to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning.” ⁷ The computer sure will have enough computing power to assume the task of masquerading as having intelligence. But maybe an intelligence of its own as Daniel Dennett speculated in his text “Can Machines Think”. He wrote: “Turing was not coming to the view (although it is easy to think how one might think he is) that to think is just like to think like a human being, -any more than he was committing to himself to the view that for a man to think, he must think exactly like a man. Men and women, and computers, may all have different ways of thinking. But surely, he thought, if one can think in one’s own peculiar style well enough to imitate a thinking man or woman, one can think well, indeed.” ⁸

Computing Machinery and Intelligence by Alan Turing, published on Mind, vol. 59, No. 236 in 1950.

In 1956 the Dartmouth “Summer Research Project on Artificial Intelligence” conference had already foregrounded and prescribed the sounding jargons of today’s AI industry, although the very term Artificial Intelligence could have been written otherwise. John McCarthy, one of the AI progenitors attributed the use of the term to “escaping associations with ‘cybernetics’,” because of either his cynicism or begrudging feelings toward Nobert Wiener, founder of cybernetics, according to Nils Nilsson’s account.⁹ Marvin Minsky, then a junior fellow at Harvard University proposed to design such a feedback-loop apparatus: “the machine is provided with input and output channels and an internal means of providing varied output responses to inputs in such a way that the machine maybe “trained” by a “trial and error” process to acquire one of a range of input-output functions. Such a machine, when placed in an appropriate environment and given a criterion of “success” or “failure” can be trained to establish “goal-seeking” behaviour.” ¹⁰ The celebrated inventor of information theory Claude Shannon presented a research on “Application of information theory concepts to computing machines and brain models,” specifically, “brain model to automata,” while “Originality in Machine Performance” and “The Machine with Randomness” were research topics suggested by Nathaniel Rochester, another founder of AI research. If AI was not, as some cultural critics chided, “a marketing schtick” (Florian Cramer), it certainly wasn’t so much of artificial intelligence per se, but anything along the line of digital computing technology in general, machine learning, neural networks, data mining, statistical science and cognitive science conglomerated. All of which was laid out in the early 50s except it had yet to wait for the computing speed, storage capacity and new input-output equipment to realize a vision in which “There is considerable promise that systems can be built in the relatively near future which will imitate considerable portions of the activity of the brain and the nervous system.” ¹¹

The later story became a familiar one: The highly anticipated triumph of Good Old Fashioned AI (GOFAI) in the 60s and 70s did not deliver what was promised as aptly critiqued by Smith: “GOFAI’s ontological presumptiveness, its blindness to the subtleties of registration, and its inadequate appreciation of the world’s richness are the primary reason, for its dismal record on commonsense.”¹² PhilosopherHubert L. Dreyfus’ infamous 1965 RAND Corporation report “Alchemy and Artificial Intelligence”  on the state of AI already foreshadows the demise of the promise of first wave AI. He famously writes:

Early successes in programming digital computers to exhibit simple forms of intelligent behaviour, coupled with the belief that intelligent activities differ only in their degree of complexity, have led to the conviction that the information processing underlying any cognitive performance can be formulated in a program and thus simulated on a digital computer. Attempts to simulate cognitive processes on computers have, however, run into greater difficulties than anticipated.¹³

AI was coldly greeted by the first so-called “AI Winter” in the 1980swhen government research funding plummeted, business support pulled out, and university research into AI stagnated. With the advent and the commercialization of the internet in the 90s and the thriving of social media platforms came a slow revival of AI research and the acceleration of development since the 2010sonward. The new generation is fascinatingly fueled by the not-so-unfamiliar taglines such as Neural Networks, Data Mining, Machine Learning, etc. Today we seem to feel, touch, breathe the ubiquity of Artificial Intelligence not only in work and play, but also in politics and our cultural imagination. Indeed, AI is everywhere and continues to expand. But as Smith also remarks: “That thinking, intelligence, and information processing is the fundamental ideas behind form allogic and computing [in the first wave AI], that they are not just things that humans do, but also things we can build automatic machines to do also underlies the second wave AI.”¹⁴ (with author’s modification)

The most recent eye-popping AIheadline comes from TNW in January of 2021, the technology insider news outlet. It reads as follows: “A trio of researchers from the Google Brain team recently unveiled the next big thing in AI language models: a massive one trillion-parameter transformer system. For example, you can go here and talk to a "philosopher AI" language model that'll attempt to answer any question you ask it. While these incredible AI models exist at the cutting-edge of machine learning technology.” The commentator also acknowledges, “it’s important to remember that they're essentially just performing parl or tricks. These systems don't understand language, they're just fine-tuned to make it look like they do.”¹⁵

In other words, AI essentially does not understand semantics, or meaning, so to speak. It simply follows syntactic rules to make logical inference and execute output as true or false. Let us also bear in mind, as Smith reminds us in the opening page in his 2019 book on AI titled: The Promise of Artificial Intelligence, Reckoning and Judgement, stating: “Neither deep learning, nor other forms of second-wave AI, nor any proposal yet advanced for third-wave, will lead to genuine intelligence.”¹⁶

Here the pursuit of genuine intelligence tacitly means the General Artificial Intelligence comparable to that of human mind.  This again returns us back to the age-old AI conundrum, which with human intelligence via a bio-chemical brain model as its logical paradigm, was and still is set up as the ultimate goal, imitating human cognition, perception, intention, while on the other hand an alternative take on the notion of intelligence which is independent of human enterprise, might well open up a more liberating frontier in AI research imagination and application. If AI (probably it then also requires another naming convention) can be viewed and measured as an autonomous and autopoietic technical reality as Gilbert Simondon had long ago advocated, or if we follow Alan Turing’s ambiguous intimation on gender and species neutrality when invoking machine intelligence, the dichotomy of us vs. AI, the imitation-induced competition and subordination between humans and AI machinery, the master / slave duality might no longer hold as a valid thesis or antithesis.

1. JoannaZylinska, AI Art: Machine Visions and Warped Dreams (OpenHumanities Press, 2020), p. 49.

2. Ibid., p.77.

3. Martin Heidegger, Nietzsche, edit.David Farrell Krell; trans. John Stambaugh, David Farrell Krell, Frank, A.Capuzzi(New York: HarperCollins, 1991), vol. 4,p. 28.

4. Eugene Thacker, After Life  (Chicago: University of Chicago Press, 2010),p. 3.

5. Brain Cantwell Smith, The Premise of Artificial Intelligence, Reckoning andJudgement (Cambridge: The MIT Press, 2019), p. 55.

6.http://cogprints.org/7150/1/10.1.1.83.5248.pdf,accessed 5/3/2022

7. https://academic.oup.com/mind/article/LIX/236/433/986238,accessed 5/3/2021

8. http://www.nyu.edu/gsas/dept/philo/courses/mindsandmachines/Papers/dennettcanmach.pdf,accessed5/3/2021

9. Nils j. Nilsson, The Quest for Artificial Intelligence (New York:Cambridge University Press, 2010), p. 53.

10. http://jmc.stanford.edu/articles/dartmouth/dartmouth.pdf,accessed 5/3/2021

11. Nilsson, The Quest for Artificial Intelligence, p. 49.

12. Cantwell Smith, The Premise of Artificial Intelligence, p. 37.

13. https://www.rand.org/content/dam/rand/pubs/papers/2006/P3244.pdf,accessed 5/3/2021

14. Cantwell Smith, The Premise of Artificial Intelligence, p. 19-20.

15. https://thenextweb.com/news/googles-new-trillion-parameter-ai-language-model-is-almost-6-times-bigger-than-gpt-3, accessed5/3/2021

16. Cantwell Smith, The Promise of Artificial Intelligence, p. xiii

新时线媒体艺术中心（CAC）成立于2013年，系国内首家致力于媒体艺术之展示、研究/创作及学术交流的非营利性艺术机构。通过展览、驻留、奖学金、讲座、工作坊及相关文献的梳理与出版，CAC为媒体艺术在全球语境中的论述、生产及传播开拓了一个多样化且富有活力的平台。CAC以批判地介入不断改变进而重塑当代经验的媒体技术来推动艺术创新及文化认知。

Established in 2013, Chronus Art Center (CAC) is China’s first nonprofit art organization dedicated to the presentation, research / creation and scholarship of media art. CAC with its exhibitions, residency-oriented fellowships, lectures and workshop programs and through its archiving and publishing initiatives, creates a multifaceted and vibrant platform for the discourse, production and dissemination of media art in a global context. CAC is positioned to advance artistic innovation and cultural awareness by critically engaging with media technologies that are transforming and reshaping contemporary experiences. 

www.chronusartcenter.org

更多展览以及艺术家信息请持续关注新时线媒体艺术中心微信公众号动态.

媒体垂询：media@chronusartcenter.org