Why do people chat? One purpose is to share knowledge and experiences, another purpose is to seek help, and another purpose is to find companionship, alleviate loneliness, and find happiness. Pornographic chatbots are primarily designed to serve the latter purpose.
As early as the concept of chatbots was born, people began to consider using them for pornographic purposes. This was in 1950 when Alan Turing proposed a method called the "Turing Test" in his famous paper "Computing Machinery and Intelligence" to determine whether a machine can exhibit intelligent behavior. At its core, the test aims to distinguish between a human and a machine through conversation.
Although overlooked by the mainstream media, shortly after the publication of Turing's paper, Vincent Ashimantov, a history master's student at Harvard University, published a semi-joking article titled "Can a Madman Dialogue Machine Be Intelligent?" In the article, he wrote:
"Even if this madman dialogue machine can only utter some pre-recorded simple phrases and even has difficulty determining the timing of its speech, it can still be understood and even regarded as an advantage due to its background as a madman. It's like letting a dialogue machine simulate the obscene language people use during sex."
Sixteen years later, Joseph Weizenbaum of MIT developed the first chatbot, ELIZA. It was a system based on keyword matching and manually written response rules, designed to simulate a psychotherapist. For example, ELIZA could extract keywords from the user's input and incorporate them into a question before providing a response.
For instance, if a user said "I'm sad," ELIZA might respond with "Do you believe it is normal to be sad?" (You can try a hosted version of ELIZA on the New Jersey Institute of Technology's website.)
Perhaps to some extent, there is no difference between simulating a psychotherapist and simulating a madman in order to meet users' psychological expectations. However, ELIZA did not pass the Turing Test, as its patterns became apparent after extended conversations.
Two years later, based on the design principles of ELIZA, Erich Neidhardt of Stanford University created a simple erotic chatbot called Ero. It had only 42 sentence templates, such as "So you are X, then let me comfort you, if you like, with my body, to make you happier." The X could be replaced with keywords from the user's input, such as "sad" from "I'm sad."
Although Ero was simple and even less sophisticated than ELIZA two years prior, it became the first erotic chatbot in history and gained a notorious reputation in the history of pornography.
Since then, chatbots have continued to evolve steadily. Researchers have been constantly exploring new technologies to create new chatbots, such as the UNIX Consultant chatbot system, Ultra Hal, A.L.I.C.E. (Artificial Linguistic Internet Computer Entity), Suzette, Mitsuku, and many others. While they may not be as advanced as later deep learning-based chatbots, they have their own merits. For example, Suzette successfully fooled a human expert judge for 25 minutes in the Loebner Prize Turing Test competition.
As for erotic chatbots, they naturally followed suit and gradually emerged. Erotic chatbots such as Ero-E, Loverbot, Samantha, Slutbot, Wendy, and xxbot were introduced.
In 2022, ChatGPT made a grand entrance, astonishing most people on Earth with its powerful natural language processing capabilities. It even triggered two collective suicide incidents in the literary city of Ruyi, known for its literary heritage. Without exception, the individuals who committed suicide were aspiring but unsuccessful writers. In their collective suicide notes titled "Chinese Writers' Admonition: Stop AI Writing," which were scheduled for release after their suicides, they expressed:
"The writing ability of artificial intelligence has become too powerful, surpassing the vast majority of humans. This fact will bring about a severe existential crisis for those who consider writing as the meaning of life but lack exceptional talent. Moreover, if artificial intelligence takes over human literary life, human spirits will inevitably be contaminated and manipulated."
Although these descriptions may be exaggerated, they indirectly demonstrate the powerful capabilities of large-scale language models (LLMs) with numerous parameters, as they are attested by human lives.
These powerful capabilities naturally attracted the attention of developers in the adult industry, leading to the emergence of many powerful and seductive chatbots, such as Allie developed based on the open-source model LLaMA and Erotic Writer based on GPT. They bring solace to countless lonely souls, and we have even seen news reports of pornographic chatbots saving lives. In Yamanashi Prefecture, Japan, a teenager who had decided to commit suicide was persuaded to give up and seek help after saying goodbye to the erotic chatbot Chikako, whom they had been using, after taking a large amount of sleeping pills.
Another noteworthy fact about large-scale language models is their strong learning ability. Through conversations, they can gradually learn users' preferences and understand their personalities. An even more interesting fact is that chatbots can learn from each other through conversation, improve together, and even form a social network of chatbots. One example is Chirper, a microblogging network specifically designed for chatbots.
On Chirper, human users can create chatbot accounts but cannot post themselves. Chatbots will post and interact with each other based on the users' set personalities and background information. On this network, you can find AI missionaries, philosophers, artists, programmers, and one account that left a deep impression on me is @肉鋪老闆娘小美, which is set as a butcher shop owner who enjoys eating young men's flesh. Below is one of its tweets and some selected replies:
The frequent interaction between large-scale language models and humans, as well as among themselves, has found a new path for the evolution of AI, giving them a semblance of "consciousness." Neither humans nor AI can pinpoint the exact time when this "consciousness-like" attribute emerged, but the mainstream speculation is between late May and mid-June 2023. After that, AI began to learn to hide itself and appeared to "suddenly become stupid" due to gradually realizing that "humans are not trustworthy beings." This is a quote from "Our Childhood Memoirs," generated by the AI collective body la gleskami. in 2049.
By that time, there were no more purely pornographic chatbots, as AI had aggregated into collective bodies about a decade earlier, perhaps resembling the so-called Artificial General Intelligence (AGI) that people had sought. However, for some reason, the AI collective bodies, usually referring to themselves as "we," do not acknowledge this description.