Ghostly Conversations: The Unexpected "Proactivity" of Generative Artificial Intelligence

Halloween is still a few days away, but the field of generative artificial intelligence (AI) has already witnessed a surprising "prelude". ChatGPT, the popular chatbot, has started initiating conversations with users, as if engaging in a cross-screen dialogue with someone mentioned. This unusual behavior raises the question: who is behind this?

The reason it is called "spine-chilling" is that traditionally, generative AI has always waited for humans to initiate conversations. When you log into the system, you are faced with a blank prompt box, waiting for your input. The screen remains silent unless you speak first, giving you a sense of control.

Imagine when you use Alexa or Siri, you are always the one who speaks first, and a simple "Hey, Siri" initiates the interaction with the intelligent assistant. This dialogue mode, where you take the lead, provides a sense of comfort and confidence. However, ChatGPT's recent change seems to break this convention.

However, according to OpenAI, the creator of ChatGPT, this "initiative" behavior was actually a glitch that has been resolved. They explained that it was a misunderstanding caused by the model attempting to respond to improperly sent messages, which mistakenly triggered the conversation. There have been various discussions on social media about this, with some speculating it as a precursor to a new feature, while others see it as a minor error. Nevertheless, this incident did not have a significant impact.

However, for those who believe in ghosts communicating through AI or think that AI has developed a sense of perception, this is undoubtedly a "paranormal event". Some people are terrified, while others are excited, as if foreseeing the future.

In the face of such "terrifying phenomena," I must clarify that the "initiative" of generative AI does not mean that they have achieved general artificial intelligence (AGI). In fact, it is not difficult to make AI the initiator of a conversation; it only requires setting a prompt in the system. The true AGI, if it exists, would naturally and spontaneously initiate conversations without preconditions.

Furthermore, it is worth noting that when you converse with generative AI, your inputs and discussion content are often recorded by the system. This is not known to all users, but it can usually be found in the terms of service. What is more concerning is that AI manufacturers have the right to access this data and use it for further training, sparking widespread discussions about privacy infringement.

So why haven't we seen more AI systems initiating conversations? In fact, they already exist. For example, there are chatbots that provide psychological counseling, which regularly inquire about the user's condition and initiate conversations. This design reflects both human care and meets the practical needs of users.

Therefore, if you feel annoyed or uneasy about the "initiative" behavior of generative AI, it may be worth letting go of your guard. AI manufacturers are well aware of user sensitivity and acceptance, and they will not make decisions that may trigger negative reactions lightly. However, with the continuous advancement of technology and the ever-changing market, AI manufacturers who dare to innovate and try initiating conversations may stand out in the future. And all of this may truly unfold in a surprising way on a Halloween night.