Connecting the dots is surprisingly difficult for so-called “artificial intelligence.” I mean this literally. I know because I conducted a rigorously controlled experiment in which I submitted to multiple generative programs a series of black dots on a white background, accompanied by the carefully calibrated prompt: “Connect the dots however you see fit.” The responses fell into three distinct categories. The first and most statistically significant outcome was total system collapse: the software simply crashed. The second category consisted of images that contained dots and lines, yet the number of dots, their relative sizes, and even their coordinates had been substantially reconfigured, as though the machine had revised my data set. The third and most intriguing result occurred when the program indeed produced lines across my actual points, but these connections hovered away from the dots themselves, as if avoiding contact. In this paper, the researcher in Pataphysics that I am, will present the complete series of meticulously absurd experiments alongside the protocols that underpinned them. I will also consider what these outcomes suggest about the peculiar operations of “intelligence” within computational systems—an “intelligence” that reveals itself not through cognition but through error, refusal, and the untranslatable impulse to misconnect.