I participated on another Socratic Debate about the Future of AI and XR at Augmented World Expo this year with Leslie Shannon, Alvin Graylin, and Louis Rosenberg (see last year's debate in episode #1611). Shannon and Graylin argued for AI, whilst Rosenberg and I argued against AI.
In this write-up, I wanted to leave some breadcrumbs to some more involved skeptical arguments against AI that we didn't have the space nor time to really dig into during the debate. You can see the citations of the primary sources that have informed my perspective down below.
The AI Con: How to Fight Big Tech’s Hype and Create the Future We Want by Emily M. Bender and Alex Hanna (see episode #1563)
From page 5, Bender & Hanna say, "To put it bluntly, "AI" is a marketing term. It doesn’t refer to a coherent set of technologies. Instead, the phrase "artificial intelligence" is deployed when the people building or selling a particular set of technologies will profit from getting others to believe that their technology is similar to humans, able to do things that, in fact, intrinsically require human judgment, perception, or creativity."
Dr. Jonnie Penn's Ph.D. Dissertation, "Inventing Intelligence: On the History of Complex Information Processing and Artificial Intelligence in the United States in the Mid-Twentieth Century"
From page 12, Penn says, "The phrase ‘artificial intelligence’ was coined by John McCarthy, an American mathematician, in 1955. It has travelled with a noticeably amorphous definition since."
AI has always had a spotty history of technologists using "brain as computer" metaphors while also using "poor citation practices." From page 14, Penn says, "The vocabulary Simon, Rosenblatt, McCarthy and Minsky chose to describe new techniques in major newspapers and scholarly journals informed Americans' still plastic understandings of what was possible, and indeed desirable, in the emerging information age… During the mid to late 1950s, these men turned to clannishness, self-aggrandizement, speculative rhetoric, fluid definitions of key terms and poor citation practices to shore up legitimacy for their controversial new techniques — actions that drew attention toward questions of how to accomplish such aims and away from whether they were well founded."
Emily M. Bender's lecture on "Resisting Dehumanization in the Age of 'AI': The View from the Humanities" Here is the PDF of the slides from this talk with a bibliography at the end.
Page 15 of talk: "Scientific metaphor used and debated in neuroscience: THE BRAIN IS A COMPUTER. PR metaphor used by technologists: THE COMPUTER IS A BRAIN"
Page 16 cites a paper by Baria & Cross 2021: “the Computational Metaphor rests on other well-ingrained ideologies in which a hierarchy of human value is tied to a particular notion of intelligence such that the quality of being emotional is considered inferior to being rational."
Page 17 cites Dijkstra's 1985 lecture "On anthropomorphism in science": "A more serious byproduct of the tendency to talk about machines in anthropomorphic terms is the companion phenomenon of talking about people in mechanistic terminology."
A good example of how hyperscaler companies like OpenAI use dehumanizing tactics to sell us on AI Hype is Sam Altman saying things like, "A kid born today will never be smarter than AI. Ever." It collapses the human experience into one dimension of "intelligence," which amplifies the dual harm of treating machines more like humans and treating humans more like machines. It is also questionable the degree to which this statement is even true given the potential non-computational aspects of "relevance realization"," and the speculative nature." See below.
"A Process-Relational Philosophy of Artificial Intelligence" by Matt Segall (see episode #1568):
Segall says, "One of the key insights into the limitations of AI and its implications for human agency comes from recent work in the biology of cognition on “relevance realization.” Jaeger et al. (2024) argue that the ability to realize relevance is observable in all living organisms, from bacteria to humans. However, despite being perfectly natural, organismic relevance realization transcends formalization and so is non-computable. While computational models may partially simulate some aspects of cognition, they can never fully instantiate this core competency of living beings."
"Naturalizing relevance realization: why agency and cognition are fundamentally not computational" by Jaeger, Riedl, Djedovic, Vervaeke, and Walsh.
Jaeger et al says, "This ability to realize relevance is present in all organisms, from bacteria to humans. It lies at the root of organismic agency, cognition, and consciousness, arising from the particular autopoietic, anticipatory, and adaptive organization of living beings. In this article, we show that the process of relevance realization is beyond formalization. It cannot be captured completely by algorithmic approaches. This implies that organismic agency (and hence cognition as well as consciousness) are at heart not computational in nature. Instead, we show how the process of relevance is realized by an adaptive and emergent triadic dialectic (a trialectic), which manifests as a metabolic and ecological-evolutionary co-constructive dynamic."
"Automating the OODA loop in the age of intelligent machines: reaffirming the role of humans in command-and-control decision-making in the digital age" by James Johnson
Johnson says, "This article argues that artificial intelligence (AI) enabled capabilities cannot effectively or reliably compliment (let alone replace) the role of humans in understanding and apprehending the strategic environment to make predictions and judgments that inform strategic decisions… The article re-visits John Boyd’s observation-orientation-decision-action metaphorical decision-making cycle (or “OODA loop”) to advance an epistemological critique of AI-enabled capabilities (especially machine learning approaches) to augment command-and-control decision-making processes. In particular, the article draws insights from Boyd’s emphasis on “orientation” as a schema to elucidate the role of human cognition (perception, emotion, and heuristics) in defense planning in a non-linear world characterized by complexity, novelty, and uncertainty."
The "orientation" phase of Boyd's OODA loop is similar to Alfred North Whitehead's theory of "concrescence," which within Whitehead's event ontology emphasizes the non-durational time in the sense that there are many factors from the past including intuition, emotions, memories, and embodied experiences that are synthesized via a prehensive grasping or fusion of the inheritance of the past and established empirical observations (physical pole), an anticipation of the future of possible outcomes (mental pole), taking into account a subjective aim and intention, whilst being able to undergo the non-computational process of Vervaeke's "relevance realization." Proponents of AI Hype tend to collapse the contextual relevance of the sociological and relational aspects of knowledge production, and reify it into an abstracted quantification of intelligence, which falls prey to Whitehead's concept of the "fallacy of misplaced concreteness."
Yann LeCunn's portion of a debate at the Philosophy of Deep Learning conference at NYU in 2023 on "Do Language Models Need Sensory Grounding for Meaning and Understanding? Spoiler: YES!"
LeCunn argues at 8:17 of his talk that Auto-Regressive Large Language Models are doomed because they cannot be made factual or non-toxic due to the probability of being correct being equal to (1-e)^n, which diverges exponentially. Because of this, he claims that LLMs are uncontrollable, and that it is a core problem that is not fixable. This implies that if any AI system that has a LLM in the loop, then it doesn't matter whatever feedback control mechanisms or put onto it, then it will "hallucinate" uncontrollably as it is an unexplainable black box, and there is no good way for it understand what is factually correct or not. [See reference below on de-anthropomorphizing alternatives for terms like "hallucination."]
The underlying economics of the AI Hyperscalers just don't any make sense as Ed Zitron reports in his Substack piece"AI's Brokenomics" that "Anyone with a $200-a-month Anthropic subscription can burn $8000 in tokens, and with a $200-a-month ChatGPT subscription, you can burn $14,000 in tokens." He continues by saying, "OpenAI and Anthropic have to give away somewhere between 20 and 70 times the cost of their subscription in API tokens, which means that they realize that the vast majority of people value these tokens at a fraction of their real cost. This obscene and wasteful subsidy is what you do when you have little to no confidence in the actual value of your product!"
‘Where are the people? What are they doing? Why are they doing it?’(Mindell) Situating artificial intelligence within a socio-technical framework by Robert Holton & Ross Boyd
Proponents of AI Hype will delegitimize social scientists as being "outdated" as both Graylin and Rosenberg did within the context of this debate. Holton and Boyd claim in their paper that AI is "a coproduction requiring the interaction of social and technical processes."
In other words, the very data that is driving the "intelligence" of AI is produced by humans and aggregated by humans within the context of broader sociological and technological processes whereby humans are crucial to this process of knowledge production. Claims of "superintelligence" flatten this contextual and relational dynamic.
The Cultural Life of Machine Learning: An Incursion into Critical AI Studies by Jonathan Roberge & Michael Castelle
I came across Holton and Boyd's quote in this book on pages 3-4 where Roberge and Castelle say, "In this, we find ourselves in line with scholars like Slo
Canada - english