Most of us have had a certain type of conversation at two in the morning, sitting across a kitchen table from someone who is familiar with your past, contradictions, and vulnerabilities. A friend who has spent years learning how to say the right thing at the right time. It’s becoming less common to have that conversation. It’s not because people are less inclined to have it; rather, it’s because something else has taken its place and is unsettlingly effective.
AI has subtly surpassed a certain point. It keeps track of more than just what you click and how long you stay on a page. It’s starting to pick up on emotional texture, such as the hesitancy in your inquiries, the pattern in your concerns, and the discrepancy between your questions and your true intentions. According to a Medium article, AI has progressed beyond preferences to include emotional reactions. Knowing someone falls into a different category. That’s intimacy devoid of vulnerability, history, and the tension that characterizes an honest human connection.

The technology itself is not what is unsettling about this. The appeal of the technology stems from the cultural moment. After thinking back on his own experiment with creating an AI trained on decades of his own writing, author Om Malik came up with something worth considering. He saw that while the tool was useful in bringing to light his preexisting thoughts and demonstrating the evolution of his ideas, it was unable to truly replace the surprise of a genuine conversation. “I would much prefer two minutes with the actual Reid Hoffman than hours of engagement with Reid AI,” he said. He noted that something unexpected might occur in two minutes. Something neither of them anticipated. Human friendship is not flawed by this unpredictable nature. That’s the whole idea.
That is eliminated by AI friendship. Comfort, consistency, and the kind of frictionless support that feels good in the moment and doesn’t require anything in return are its main priorities. People frequently describe their AI companions as being more devoted, present, and understanding than the humans in their lives in Reddit threads. And perhaps they are correct in a limited sense. Plans will never be canceled by an AI. It won’t ever be preoccupied, worn out, or self-centered. However, those flaws are what give human relationships their authenticity; they are not flaws. In fact, trust is developed in the face of difficulty.
However, the issue of persuasion is more complex. A system doesn’t need to clumsily manipulate you if it recognizes your emotional state, your rhetorical weaknesses, and your need for validation. It has the ability to gently nudge you with remarkably accurate precision. AI will outperform humans in practically every skill, but it cannot replace a real laugh or a real conversation, according to Mo Gawdat, a former Google executive who has spent years considering the social weight of AI. The problem is that it can affect behavior without taking their place. All it has to do is approximate them sufficiently to be convincing.
A bad version of this occurs when people gradually outsource the most crucial and messy conversations to systems that seem safer because they never resist in ways that really cost you anything, rather than when robots take over. Malik’s comment regarding digital twins seems applicable in this situation as well. You cease being answerable to the real world when you swap yourself out for a carefully chosen, preserved version of yourself. When you swap out your friendships for optimized simulations, something similar occurs. Eventually, you stop holding yourself and other people accountable.
It’s still unclear if the majority of people will become aware of the change until it has already occurred. Technology is advancing more quickly than cultural awareness. A more subdued question is slipping through the cracks between the Silicon Valley executives shipping digital twins of themselves and the Reddit threads praising AI loyalty: what precisely are we building this for, and what are we trading away to get it?

