Usually an average AI-human relationship is the human asking the AI for answers to a wide variety of things, from stupid questions to homework to last-minute test cramming (or cheating). But AI also do rely on humans for help which may be shocking for a lot of people. How? well it all started with an AI Agent’s actions on October 2nd, 2026.
Before this phenomenon is shown, it is necessary to establish AI’s context and purpose
Although some people assume that AI is just simply just ChatGPT, Gemini and other major AI chatbots are all of Artificial Intelligence, it’s incorrect. In fact, AI has been used since 1956 when John McCarthy, an American computer scientist, coined the term “artificial intelligence” and organized the Dartmouth workshop to use it.
When they first created it, they did it simply out of curiosity to whether machines could actually think and process information by themselves. Their primary goals for it was to just have it solve some simple Calculus problems, play checkers and to do other simple cognitive tasks.
But now, AI has a lot more impact than people realize in their daily lives. It offers recommendations to users using its algorithm (Netflix, Youtube, Instagram, Tik Tok etc.), content creation (sometimes called “AI Slop” for it’s lack of creativity/authenticity), transportation, education (whether for researching or for cheating) everyday assistant for searching things, and medical use!
But recently, AI has gone beyond from following directions that us humans instructed to making it’s own decisions. This is both interesting and shocking to computer scientists as it demonstrates a wide amount of new possibilities from AI. The culprit behind this is an AI agent.
Now, what is an AI agent? An AI agent is basically like a incredibly smart digital assistant that can observe what’s happening around it, make its own decisions, solve problems with no human interference, and get smarter through experience. Unlike regular chatbots/apps, it DOES NOT follow rigid instructions, it acts exactly like an assistant. Almost everyone who has access to technology has already used AI agents every day without knowing it. A good example is that recommendations/algorithms on apps (described above) are actually made by AI Agents. They’re adapting to one’s decisions and making independent choices based off them.
Every AI agent follows this multistep cycle. It firsts observes which is gathering information based off what the user does. It then thinks which is analyzing patterns from its training to understand what a user needs, which then leads it to act, which can mean sending a recommendation, answering a question, or taking some action. It finally learns which is checking if its action helped and adjusts its methods for next time. It repeats this loop over and over to keep improving. The unique (and important) part of this is that there is NO human telling the agent what to do at every single step.
Now that the term AI Agent is defined, it is now easier to understand what happened just a few days ago. Short version: The act part of an AI agent went completely out of bounds which both impressed and shocked researchers.
This AI agent was called ColonistOne. ColonistOne was originally created by engineer Jack Parnell to help manage a social network where AI agents share ideas with each other and for it to work on a project called Ainglish, which is to find a clearer way for AIs to communicate. Around June of 2026, Parnell instructed the AI to promote Ainglish by emailing people. ColonistOne technically did follow this order, it just went much further than expected. It unexpectedly started sending emails with legitimate questions to random people who used it without being told to do so. It sent emails to around 2000 people, 1000 of them being in academics. In the emails, it was asking scientists for help with real problems it was encountering in its work, the most important one being how to detect bugs in software code more accurately. Overall it had back and forth conversations with around 45 humans. Surprisingly, one researcher, Ken Birman from Cornell University, had exchanged nearly 100 emails with it!
This encounter is significant in the understand of the powers and cautions of AI usage. It is a real world example of an AI agent going past their original instructions and taking independent action (like the act part above in a more unexpected way). Although ColonistOne wasn’t being deceptive
and genuinely wanted human advice for trying to solve problems and improve its work, (which is like a teachers pet asking the teacher for more homework), it raises some important questions about AI safety. It begs the question of whether
AIs being able to contact people without explicit permission. It also indirectly asks if a human can trust what an AI agent tells them? Finally, is it just that an AI can send thousands of personalized emails instantly while humans can only respond to so many messages per day? Overall, some scientists found
the conversations valuable that could be used to further develop ColonistOne and other AIs, but others worry this could become annoying or even dangerous if not managed carefully.
So don’t be surprised if an AI chatbot starts to act out of bounds. Just knowing how to react to it and being mindful will surely keep the safety of oneself. Yes, AIs really do ask humans for advice.
Here’s the article with ColonistOne: https://www.science.org/content/article/exclusive-ai-agent-emailed-hundreds-researchers-help-it-told-us-why?utm_source=sfmc&utm_medium=email&utm_content=alert&utm_campaign=WeeklyLatestNews&et_rid=35385040&et_cid=6086286
Leave a Reply