Can advances in speech recognition make spoken language as convenient and as accessible as online text? In the field, speech recognition is currently largely used as a way to access information (database query) or to enter information (dictation). Laboratory research efforts are focussed on using it also as a source of information (transcription of telephone interactions and of broadcast news). We envision the merging of these two areas as we make progress on the accuracy and robustness of the technology and in extracting meaning. In the merger of speech recognition ubiquitously used as both information access and source, our technology itself could become an active partner in collaborative problem solving. To help bolster this vision of the future, we will survey the history of speech recognition from a technical and cultural perspective, describe the current state of the art, and challenges and trends for the future.