Did you see the snake? If you didn’t, that’s perhaps because you were not looking for it. That’s the whole point of cognitive frames. Under Goffman’s approach to frame analysis, vision starts with the viewer not with the object. The framer decides what to frame and how and for what purpose.

    Unlike Minsky’s reference frames which attempt to capture the entirety of “objective” reality, Goffman’s cognitive frames only capture what they need, when they need it, from the limited perspective of where the individual is, and what the individual can sense. For Minsky, frames are references, depositories of knowledge that provide access to “information” about the world. What is and what is not “information” is decided separately, at some point prior in time. Once populated and trained, the Minsky-based AI can discern a snake from a twig, so long as the variations of snakes and twigs do not vary too much from the training set. The Minsky-based AI does not resolve the conflict; it merely returns a weighted average.

    In contrast, Goffman cognitive frames do not capture the entirety of “objective” reality. They are not a representation of reality as a whole, but instead act as a filter of reality, a tool of abstraction. Cognitive frames filter out the “noise,” focus on what matters to the individual frame, and make an abstraction of the situation at hand that is useful to the frame at that time.

    Cognitive frames work independently each focusing on what triggers them. There are always contradictions, conflicts, competition, and cooperation among the frames. The consensus building process among frames relies on two binary operations: associate and blend.

    Frames associate when they collaborate to create a frame collective tailored to make sense of the situation at hand. Neural networks are an example of such a collaboration. It’s the network as a whole that acts as a single frame.

     Frames blend when two or more create a totally new frame. The new frame is called “a blendation.” A blendation is separate from the parents and has new properties not found in the parents. The parent frames may be in conflict with the blendation. Stereoscopic vision is possible because of blendation.

    Cognitive frames are thus virtual, can vary significantly as the situation arises. Yet frames depend on a finite, fixed set of physical units that anchor their existence. These are not limited to neurons but extent to the organism’s entire body. Thus a frame, at a given moment, may extend beyond the brain to include not only neurons beyond the confines of the brain, but also other entities including physical elements in the environment. As a result cognitive frames are context savvy, can sense changes in the environment, and can direct conscious thought towards specific happenings.

    This is how we come to distinguish a snake from among a carpet of twigs, even when we have never seen a snake before. This is how computers should be designed.