China has decided to put limits on a very specific type of AI: that designed to look like a person. Doubao (ByteDance) and Qwen (Alibaba) have announced that they will disable their custom AI agents before the regulations come into force imminently. Tencent had already done it with its Yuanbao model, but behind all these closures there is a much deeper idea.
The blackout of anthropomorphic AI already has a date. Doubao has already warned its users that its personalized AI agents will stop working on July 15 due to “product adjustments” and that three months later they will not be able to access the stored conversations with said chatbots. With Qwen things will be even faster: the agents will disappear on July 10, and the rest of the functions will do so five days later. These measures are taken before the entry into force of the new Provisional Measures for the Administration of Anthropomorphic Interaction Services with AI, recently published by the Chinese government.
Conversational chatbots, yes… The interesting thing about the regulations is that they do not regulate generative AI in general. Leave assistants out to work, study, research or serve customers. What it regulates are the systems that “simulate personality traits, thought patterns and human communication styles to provide a sustained emotional interaction.” The problem therefore is not that AI can be more or less powerful, but that it comes to simulate the behavior of a person.
…company chatbots no. The functions that disappear from these platforms are those that allowed a generalist chatbot to be turned into a fictional character, a friend or even a romantic partner. This personalization gave these AI agents a fixed personality, a memory and a characteristic conversational style that reinforced interaction with it. They were therefore not tools to replace the current search engine, but rather systems designed to trap the user in an increasingly personal and intimate relationship and conversation.
Beware of emotional dependence. Here the Chinese regulation does not try to identify dangerous risks such as those traditionally discussed, for example, with the European Union regulation. Instead what it monitors are interactions in which there is an emotional dependencea potential addiction and even manipulation by these systems. In fact, these systems are prohibited from fostering emotional bonds that could lead to irrational decisions. All of this is stopped by regulations to avoid the potential deterioration of personal relationships.
AI must analyze itself. According to the regulation, AI must intervene when it detects these types of situations. If a user shows too extreme emotions towards these chatbots or if it is perceived that there may be risks to their integrity, the AI should encourage them to seek help or even activate certain emergency protocols. You should also warn if you see excessive usage and remind the user to rest after two hours of continuous talk. In a certain sense, these regulations are reminiscent of the “digital welfare” mechanisms that have been imposed on mobile apps that try to limit the use of social networks.
Protection for vulnerable people. The measures pay special attention to underage users and also to older people. For example, these types of chatbots are prohibited from offering intimate virtual relationships, such as couples or artificial relatives, and they require parental consent. There is also an obligation to take precautions when the users are elderly, since this group is especially exposed to loneliness and a possible emotional dependence.
AI should stay just that: be an AI. Although there have been advances in the psychological impact of these models in their relationship with people, this is probably the first major regulation that does not attempt to control how powerful AI models are, but rather how capable they are. to establish personal relationships with users. This discussion will probably end up reaching Europe and the US, and is a touchstone for a technology that has certainly been infiltrating the personal sphere for some time.
Image | Andy Kelly
(function() { window._JS_MODULES = window._JS_MODULES || {}; var headElement = document.getElementsByTagName(‘head’)(0); if (_JS_MODULES.instagram) { var instagramScript = document.createElement(‘script’); instagramScript.src=”https://platform.instagram.com/en_US/embeds.js”; instagramScript.async = true; instagramScript.defer = true; headElement.appendChild(instagramScript);
was originally published in
Xataka
by
Javier Pastor
.
GIPHY App Key not set. Please check settings