One of the many benchmarks for AI is the “Turing test,” Alan Turing’s adaptation of the “imitation game” where an interrogator must decide which of two respondents is a computer. It is, as many have pointed out, a strangely indirect test, one that depends on the credulity of the human interrogator and the capacity of the machine to deceive (Wooldridge 2020). Will they believe the computer? And will the computer be a good enough liar? As Pantsar (2025) comments, “For the machine to pass the test, it needs to impersonate a human successfully enough to fool the interrogator. But this is puzzling in the wide context of intelligence ascriptions. Why would intelligence be connected to a form of deception?”
On the one hand, measuring AI through its deceptive power has the benefit of avoiding the idiocy of attempting to establish a measure of intelligence, a task deeply imbricated in racial eugenics (Bender and Hanna 2025; Wooldridge 2020). On the other, generative AI applications seem to have been developed with deception in mind–deception from all parties. The application designers want to present outputs as just as good (if not better) than the products of human work. And the humans that utilize generative AI often seek to present these outputs as their own work. And even though ethical practice demands that we acknowledge the utilization of AI, the goal is that people consuming AI material will be unable to tell the difference or to mark where the machine ends and the human begins. So even though Turing tests may be a poor measure of machine “intelligence,” they seem to fit the moment.
But there’s another part of the Turing Test that is especially relevant: the organization of the “imitation game” itself. Let’s go back to Turing’s 1950 article. As Sterrett points out in a series of articles, there are two games described in Turing’s landmark essay, an "Original Imitation Game” and a “Standard Turing Test” (the terms are Sterrett’s) (Sterrett 2020: 469). The first one describes two rooms, one for the interrogator, and the other for a man and woman, who communicate with the interrogator via a teletype.
In order that tones of voice may not help the interrogator the answers should be written, or better still, typewritten. The ideal arrangement is to have a teleprinter communicating between the two rooms. Alternatively the question and answers can be repeated by an intermediary. The object of the game for the third player (B) is to help the interrogator. The best strategy for her is probably to give truthful answers. She can add such things as '' I am the woman, don't listen to him! ' to her answers, but it will avail nothing as the man can make similar remarks. We now ask the question, ' What will happen when a machine takes the part of A in this game? ' Will the interrogator decide wrongly as often when the game is played like this as he does when the game is played between a man and a woman? These questions replace our original, ' Can machines think? (Turing 1950: 434)
While people have certainly looked at the gender component here, most of the attention has been on the game itself and its overall intention: deception (Patterson et al 2018). I want to focus on the whole arrangement itself.
The game “works” through series of limitations. The interrogator can’t go into the other room, nor can the human burst into the interrogator’s room. The interrogator is also prevented from hearing them. Communication, Turing tells us, is best accomplished through teletype. Of course, one would need such an arrangement with a computer. Yet, game rules aren’t just arbitrary, and Turing’s test tells us much about social hierarchies and workplace organization. By the time Turing wrote his test, the office had taken its contemporary form–a spatialized organizational chart made up of offices communicating with each other through a variety of technologies: mail, telephone, teletype, pneumatic tube. Inter-office communication is also rendered through a variety of intermediaries and, as “organization man” develops, cybernetic systems of communication and decision-making come to dominate managerial processes (Whyte 1956).
In other words, Turing’s “imitation game” rules are a description of contemporary work. Turing’s readers would have easily conjured up an image of an organization where people don’t communicate face-to-face with each other. This has only become more marked in the intervening decades, where people are duped into long-term relationships without ever meeting their con-artist. More to the point, we are more and more called upon to judge between human- and computer outputs, a task in which humans have proven occasionally successful. But perhaps only under special conditions.
One of the theses that I have suggested (both on this blog and in published work) is that the triumph of automation, algorithm and AI are initially built not upon technological change, but upon behavioral and cognitive change. Before business owners can begin to replace workers with algorithmic process and generative AI, before, in other words, those outputs can be accepted as “just as good” or “better” than human work, humans and their labor must be constrained, delimited and “de-skilled.” And more than this - everyone has to be convinced that these narrowly defined outputs are as good as we humans get. With the Turing test, human communication is reduced to lines on a teletype. With algorithmic analysis, applying to a job, getting an apartment or reading a mammogram are reduced to scores that can be ranked, patterns that can be assigned probability, etc.
As I explained in an earlier essay, this process of labor alienation unfolds across several steps. First, human labor is parsed out into a series of constituent functions, consistent with the Taylorization that transformed factory labor. Second, those functions are reconceptualized as algorithms: steps in a chain of operations that proceed from inputs to outputs. These might be scripts for telemarketing, decision trees for insurance claims, procedures for reporting inventory loss, etc. Next, workers are confined to those algorithmic choices, and part of the de-skilling process means penalizing workers for deviating from the script. Finally, automation (AI applications and AI-infused platforms) replaces workers.
The important part here is the human side of the transformation. Humanity must be reduced, and people must be convinced that algorithmic processes are interchangeable with the etiolated human. In other words, the Turing test only works if we stay in our room, if we don’t shout, if we don’t bang on the walls with our fists. And there are 2 levels of deception - one in convincing (or forcing) people to reduce themselves to narrowly defined outputs, the other in misrecognizing those algorithmic products as the total of human possibility.
Does generative AI involve a similar series of reductions? Of course it does. In all kinds of professions, our labor has been reduced to the production of “content”: bland and repetitive text, stereotypical images, boilerplate scripts. People produce like this in accordance with late capitalism. The only way, after all, to monetize that YouTube channel is to constantly update with new material. And that new material must fit the algorithmic desires embedded in the platform. So: you start making a lot of content, and you align it with the algorithm. The next step is, of course, to replace you with generative content. ChatGPT may have come as somewhat of a shock to us educators in 2022, but it must have come to no surprise for platform laborers, who have been human generative AI for several years now.
References
Bender, Emily and Alex Hanna (2025). The AI Con. NY: Harper.
Collins, Samuel Gerald (2018). Welcome to Robocracy. Anthropology of Work Review.
Pantsar, Markus (2024). “Intelligence is not deception.” AI & Society.
Patterson, W., Boboye, J., Hall, S., Hornbuckle, M. (2018). The Gender Turing Test. In: Nicholson, D. (eds) Advances in Human Factors in Cybersecurity. AHFE 2017. Advances in Intelligent Systems and Computing, vol 593. Springer, Cham. https://doi.org/10.1007/978-3-319-60585-2_26
Sterrett SG (2020) The Genius of the “Original imitation Game” test. Mind Mach 30(4):469–486. https:// doi. org/ 10. 1007/S11023-020-09543-6
Turing, Alan M.(1950). “Computing machinery and intelligence”. Mind Q. Rev. Psychol. Philos. LIX(236), 433–460.
Whyte, William H. (1956). The Organization Man. NY: Simon and Schuster.
Wooldridge, Michael (2020). A Brief History of Artificial Intelligence. NY: Flatiron Books.
No comments:
Post a Comment