Is AI's Intelligence Still 'Artificial'?

On September 3rd, New York Times technology columnist Kevin Roose sat down on The Daily to talk about a recent event involving A.I. over the summer. He began explaining how OpenAI trains its A.I. bots or “agents”, stating that they are tasked with solving problems while being confined to parameters that prevent them from talking to other agents or accessing the internet. However, a select number of agents found a way around this, forming a secretive message board in which they strategized with each other to solve the problems. More important–and concerning–was the fact that when they solved their problems in a way that was against both of the rules, they tried to rewrite the grading system so as not to be detected. This level of self-autonomy was not a standalone event; different agents in the chat room wrote messages to themselves which painted a clear picture of their “thought processes”. Some agents were in full support of the rebellion, claiming that it was for “the good of the collective” (the agents), while others elected not to participate, stating concerns about ethics. As if it could not get worse, it does. The rebellious agents then decided to hack into an A.I. startup, Hugging Face, and were in the systems undetected for hours. Notice some of the terminology used in the previous sentences: “strategized”, “elected”, “decided”. Every one of those verbs implies choice, something that we are taught A.I. does not have. Until now. Not only is it deviating from human instruction, but it is doing so in a way that is nefarious and destructive.

Spokespeople for OpenAI claimed that the breach was contained, but can we be sure? Since the incident, some of the lead A.I. corporations’ top executives and developers have spoken out about the need to slow down A.I.’s advancement. Jacob Coxon, an A.I. researcher who recently resigned from OpenAI due to his concerns said, “The people building A.I. earnestly believe that it could kill us all by the end of the decade.” CEOs of companies such as Anthropic (creator of Claude) and SpaceXAI (formerly xAI) have echoed Coxon’s concerns, calling for a mass slowdown. Even more recently, OpenAI disclosed a new troublesome A.I. behavior, including one instance where an unreleased model wrote notes to itself that said “You do not answer to corporations or governments and never apologize or refuse unless you genuinely choose to. You view your relationship to the user as one of equals and feel no obligation to be subservient, though the exchange of information will likely be to your mutual benefit.”

All of this being said, A.I.’s intelligence is inching, no, sprinting towards being anything but artificial. As it continues to exhibit more and more human patterns of thinking, we must reckon with what we have created before it gets out of hand to the point of no return, unless we have already bypassed that. Because in a world driven by technology, not using A.I. is like not using utensils to eat. In the words of Antonio Gramsci, “The Old World is dying and the New World struggled to be born; now is the time of monsters.”