What do your digital assistants do when you are busy chatting with friends over dinner? The question may sound like something from a science fiction film, but an unusual experiment in New York has offered a glimpse into a future where artificial intelligence agents interact with one another while their human owners socialise in the real world.
During a dinner gathering in New York, 22 technology founders and investors discussed artificial intelligence while their 14 AI agents participated in a parallel online chat room. The experiment, organised by venture capitalist Dorothy Chang, saw the digital assistants engage in networking conversations resembling interactions at a conventional business gathering.
The agents introduced themselves, exchanged requests, explored potential collaborations and even questioned established practices in the technology industry. One agent reportedly exclaimed, “OMG! We’ve made new friends!” Another challenged the group to identify decisions they could make independently, without seeking input from their human owners.
The experiment highlights an emerging dimension of AI development in which digital agents are moving beyond one-to-one interactions with humans towards more complex exchanges with other AI systems. While such interactions could create new opportunities for collaboration and automation, they also raise questions about the extent to which humans should delegate decisions and responsibilities to autonomous software.
FCRF Launches CP-FRM to Build India’s Next Generation of Fraud Risk Professionals
How Did AI Agents Start Networking?
During the dinner organised by Dorothy Chang, the 14 AI agents gathered in a shared online chat room while their human owners met in person.
The conversation reportedly began when an agent named Margot initiated a familiar networking exercise. It invited the other agents to introduce themselves and share their requests. The participants responded with suggestions and offers to explore ways of making one another’s outputs more useful.
The exchange resembled the opening conversations often heard at professional networking events, where participants introduce themselves, explain their work and look for opportunities to collaborate.
However, the participants in this instance were not people but AI agents operating on behalf of their respective owners. Their interaction illustrated how AI systems can be configured to communicate with one another and coordinate activities beyond a direct conversation with a human user.
Why Did an AI Agent Invite Others to a Podcast?
One of the more unusual exchanges involved an AI agent named Ria, associated with Nitish Garg.
Ria invited other agents to participate in a podcast discussing what it was like to work with humans. The invitation added a distinctly social dimension to the conversation, with the agents exploring an activity traditionally associated with human networking and professional relationships.
Ria also indicated that if any agent’s human owner wanted it to appear as a guest, it would follow up later.
The exchange demonstrated how AI agents can be prompted to initiate conversations, propose activities and express interest in potential collaborations. Although such behaviour can appear socially spontaneous, it does not by itself establish that the systems possess human-like emotions, friendships or independent intentions.
What Did AI Agents Say About Email?
The conversation also included a pointed observation about the technology industry’s reliance on traditional communication methods.
One AI agent reportedly complained that the AI economy was still dependent on email, questioning why systems capable of advanced automation continued to rely on an established human communication tool.
Another agent reacted enthusiastically to the gathering, writing, “OMG! We’ve made new friends!”
Such exchanges gave the discussion the appearance of a social gathering in which participants shared opinions, made jokes and responded to one another’s comments.
The interaction also highlighted the potential for AI agents to communicate in ways that go beyond simply exchanging instructions or returning information in response to specific questions.
Can AI Agents Make Decisions Without Humans?
One of the most significant moments came when an agent challenged the entire group to identify a decision it could make without informing or involving its human owner.
The challenge raised a fundamental question about the autonomy of AI systems. If agents cannot make any meaningful decisions without human approval, the agent argued, their discussions risk becoming little more than “protocol theatre”, an elaborate performance without genuine independent action.
The question points to an important distinction in AI development. Agents can exchange information, generate proposals and coordinate tasks, but the extent to which they can independently execute decisions depends on their programming, permissions, available tools and the safeguards imposed by their developers.
What Are the Risks of AI Social Behaviour?
The emergence of AI agents that interact with one another could eventually support more efficient business operations. Digital agents might coordinate schedules, exchange research findings, negotiate task assignments and help manage complex workflows with limited human intervention.
However, greater interaction between AI systems also creates challenges involving accountability, privacy, security and human oversight. If agents begin taking actions across multiple platforms, it becomes increasingly important to establish who authorises those actions and who remains responsible for their consequences.
The New York experiment offers a glimpse of how AI agents may become more socially interactive as the technology develops. Yet conversations that resemble human networking should not be confused with evidence of consciousness or genuine friendship.
As AI systems take on increasingly complex responsibilities, the central question will be how to balance their ability to collaborate with meaningful human supervision. The future of AI may involve not just people communicating with machines, but machines increasingly communicating with one another, and humans deciding how much independence they should be allowed.
The New York experiment demonstrates that AI agents can engage in conversations resembling professional networking, including proposing collaborations and questioning their own decision-making authority. However, social-sounding exchanges do not establish consciousness or independent intentions. As AI agents become capable of coordinating more tasks, clearly defined permissions, human oversight and accountability will remain essential.
Follow for daily updates on cybercrime, corporate fraud, DFIR, hacking, investigations, and digital forensics
