Effective public engagement with complex technologies requires a nuanced understanding of how different audiences make sense of and communicate disruptive technologies with immense social implications. Using latent class analysis (LCA) on nationally-representative survey data (N = 2,700), we examine public attitudes on different aspects of AI, and segment the U.S. population based on their AI-related risk and benefit perceptions. Our analysis reveals five segments: the negative, perceiving risks outweighing benefits; the ambivalent, seeing high risks and benefits; the tepid, perceiving slightly more benefits than risks; the ambiguous, perceiving moderate risks and benefits; and the indifferent, perceiving low risks and benefits. For societal debates surrounding a deeply disruptive issue like AI, our findings suggest potential opportunities for engagement by soliciting input from individuals in segments with varying levels of support for AI, as well as a way to widen representation of voices and ensure responsible innovation of AI. •We classify Americans' AI perceptions into five segments: negative, ambivalent, tepid, ambiguous, and indifferent classes.•Views of AI vary both by the level of news attention and the content audiences attend to.•The negative and the ambivalent classes largely differ in support for AI, but agree that their voices should be heard.•The indifferent and the ambiguous classes include more minorities who may be disproportionately affected by AI.•Now is a great time to engage with the publics on issues related to AI because it is not overtly politicized.