Roku

Roku is credited alongside Amit Desai in a conference recording about how AI assistants, wearables, and robots should respond when uncertain. The supplied source describes Desai’s experience with voice interfaces at Roku, but provides no independently verified company overview or current employment confirmation. This profile therefore focuses on the credited talk.

The presentation separates model accuracy from decisions about whether to act, ask for confirmation, or decline a request. Its music-request example holds accuracy at 79 percent while reducing user burden by changing those decisions. Desai argues that confidence thresholds should reflect the different costs of mistakes and interruptions.

Desai introduces the Outcome User Cost Heuristic, or OUCH, to compare the burden of possible outcomes. Adding confirmation creates another response option that can reduce that burden further. The talk also applies the approach to television interactions, where presenting choices visually changes the costs involved; the source does not establish a deployed Roku product.