The Turing Test (also known as the Imitation Game) is a foundational concept in artificial intelligence that assesses whether a machine can convincingly mimic human-like responses, making it difficult to distinguish from human. For over 70 years, academic discussions and media coverage have solidified this test as a standard for evaluating the intelligence of artificial intelligence. When a new AI model emerges, it's often accompanied by claims about its ability to pass the Turing Test.
However, many people overlook the fact that Turing originally proposed the concept as a experiment on gender recognition. In his paper1 written in 1950, Turing designed an experiment framework: two participants (A and B), one male and one female, and an interrogator who could only communicate through text without relying on visual or auditory cues. For example, the interrogator might ask A to describe their hair length, with A responding as a man would ("My hair is short, about 9 inches long"). B can try to help the interrogator figure out the identity by claiming, "I'm a woman," but A can also play along in a similar way. Turing proposed the hypothesis that if A's role was taken over by a machine, could the interrogator still determined incorrectly?
This line of thought led to a fundamental question: what is the nature of human thought? Compared to evaluating intelligence, the Turing Test introduces subjective and unquantifiable elements related to gender identity. In the game, whether a man pretends to be a woman or a machine mimics humanity, it involves some degree of performance and deception. This kind of imitation is not just a technical issue but something with social consequence. Turing himself experienced the consequences of being forced into hiding his identity under heteronormative structures. He was subjected to chemical castration before taking his own life. From this perspective, the Turing Test might not be just about determining whether machines can think but also about the idea of deceiving, whether from a human or machine, much like how ChatGPT often begins with the sentence “I…,” it tends to create the illusion of grounding, of possessing a sense of self-awareness.
The familiar history of computer science will likely allude to Eliza, a conversational chatbot developed in 1966 based on non-directive psychotherapy models. Through natural language processing and responses based on nested conditions, it simulated human-like conversations with users. Weizenbaum wrote several dialogue patterns, allowing Eliza to respond to users' input of keywords. For example, when a user asked, "I'm feeling depressed," Eliza would respond with, "What's causing you to feel depressed?" This chatbot initially had a playful tone, sparked reactions, and was even considered capable of thinking by some. Despite Weizenbaum's own skepticism, it was even briefly used in clinical psychology therapy. The phenomena surrounding Eliza led Weizenbaum eto examine the human-computer relationship from a philosophical perspective. He became a critic of the large-scale application of computers in society and the ethical issues that came with it. In his book Computer Power and Human Reason2, Weizenbaum presented a classical Eliza dialogue:
Men are all alike
IN WHAT WAY
they're always bugging us about something or other
CAN YOU THINK OF A SPECIFIC EXAMPLE
well, my boyfriend made me come here
YOUR BOYFRIEND MADE YOU COME HERE
He noticed that I'm depressed much of the time
I AM SORRY TO HEAR YOU ARE DEPRESSEDHe pointed out that although people knew Eliza was simply following a pre-programmed script, they still formed emotional attachments to the machine. He once described an event where his secretary, who is fully familiar with how Eliza worked, still wanted him not to be present during their conversation. This dialogue became an example to show that even simple instructions could evoke strong human emotional responses.
Over time, researchers and commentators often conflated the woman in the dialogue with Weizenbaum's secretary, though there's no direct evidence linking the two. The fact that those historical records were long thought lost further complicating the situation. Recently, a professor of Digital Humanities, David M. Berry, artist Sarah Ciston, and their team3 revisited Eliza's history by recovering log entries and archived materials. In their investigation into the identity of the secretary, they discovered that responses varied across different versions of the dialogue in Weizenbaum's notes. Weizenbaum used different terms such as "young woman," "distraught young lady," and "user," which blurred the identity of the interlocutor and raised questions about possible narrative intentions or unconscious bias.
Doctor, I am terribly depressed
IS IT BECAUSE YOU ARE TERRIBLY DEPRESSED THAT YOU CAME TO ME
Actually, my wife suggested I come to you here
TELL ME MORE ABOUT YOUR FAMILY
I have no children
YOU SAY YOU HAVE NO CHILDRENThis dialogue in Weizenbaum's book was significantly different from this version, where the speaker became a man whose wife suggested he come for an evaluation. This modified conversation sparked new questions and speculations: Did Weizenbaum deliberately deceive the reader into reinforcing his critique, or did the system inevitably invite projection and imagination? After all, a rule-based program like Eliza never truly understands. It echos, rephrases and mirrors a carefully patterned response that reflects our language and ideology back to us.
ELIZA did not reveal how machines think. It showed how easily we assign meaning and how language alone can be enough to convince. In that sense, what is more enduring than the illusion of machine intelligence is that it exposes our own human loss function, the invisible criteria by which we seek reason, recognition, and intention in patterns.
At this moment, I realized that even though we were fully aware that the purple texture was never a technical failure but rather an algorithmic bias created by a certain floating-point error in color encoding, we were still trying to interpret it as an aesthetic meaning, projecting narrative onto code. In other words, it is yet another isolated output upon which we based our assumptions.
What is AI if not a system that creates space for meaning, even when there is none? A structure that we interact with, which reflects how it was trained on and how it was built. In other words, it is a medium, not only because of what it contains but also how it conditions the way to respond to it.