New research explores how AI can help speech-generating device users reclaim timely, humorous comments —without losing their own voice
INFO Assistant Professor Stephanie Valencia² demos her team's work. Photo via Spoken app.
A group of friends sits around a table. One is venting about their aging sedan and the constant appearance of its check engine light. One of them has a speech impairment and uses an Augmentative and Alternative Communication (AAC) or speech-generating device to communicate. Listening to the story, they want to contribute a lighthearted comment to show empathy and shared frustration. They know the exact sentiment, but typing it out word by word would take too much time and stall the conversation. Instead, they tap the “joke” mode on an experimental, AI-equipped AAC app. The AI, having listened to the conversation, instantly generates a suggestion. The user selects it, and the device speaks the line:
“At this point, your ‘check engine’ light is just a mood light.”
The group laughs. The friend grins and nods, “Seriously!” The AAC user hasn’t just delivered a joke; they’ve successfully participated in the shared, cathartic moment.
This scenario is at the heart of research by the University of Maryland College of Information Assistant Professor Stephanie Valencia², Cornell Tech computer science PhD student Tobias Weinberg, who is an AAC user himself, and Information Science Assistant Professor Thijs Roumen at Cornell Tech. They are exploring how AI can help rebalance the scales of conversational timing for AAC users.
The situation described above raises important questions about AI-mediated communication with AAC technology, and presents a tradeoff between user agency vs efficiency, on one end of the spectrum the user can type word by word which gives them full agency over their speech but is really slow, on the other hand a joke button that gives automatic jokes is very efficient but it is likely not your joke.
In their study Why So Serious? Exploring Timely Humorous Comments in AAC Through AI-Powered Interfaces, which was presented at a top-tier HCI conference (ACM CHI’25) in Japan, winning Best Paper Honorable Mention award and Jury Best Demo award, they explored this tradeoff with four different AI-powered AAC interfaces, situated in different points of the spectrum, each offering a different level of user agency. These ranged from a “Context Bubble Selection” interface, which gives users high agency by letting them select specific parts of the conversation transcript to base the joke on, to the “Full-Auto” version that generates a complete joke with a single click for maximum efficiency.
The findings revealed a nuanced picture of the tradeoff between agency and efficiency. While users valued control, the social function of a quick quip often made speed the priority. The Full-Auto interface was frequently favored because, for a brief humorous interjection, achieving the goal of timely participation was more critical than perfect control over wording.
“The key question here is the function that humor is serving,” explains Weinberg, underscoring this finding. “In this context, the joke serves as a medium to keep the flow of conversation… It is not so much about the joke itself.” As Valencia² notes, this spontaneous participation is incredibly difficult with standard AAC speeds, which only allow users to communicate approximately 12-18 words per minute.
But this efficiency comes with a profound risk: the AI doesn’t just assist—it influences. One participant in their study astutely observed that the AI both “matched my intentions and also influenced it.” This gets at a major ethical dilemma in AI-mediated communication. If an AI constantly shapes what you say, when does it stop being your voice?
For Valencia², this highlights the complex definition of agency in AAC. “Agency is a really big word,” she says. “In a social setting, agency is ‘how can you contribute what you want to contribute and advance your goals despite social constraints.’”
“When we talk about AI, it gets a little bit more complex because then you have another layer where AI is giving you suggestions that might be great to help you achieve that goal. But is it in the style that you like? Is it how you imagined it?”
The challenge, then, is to build AAC tools that empower rather than overwrite. The future of AAC lies in creating systems that act as a true partnership. The goal is to move beyond a tool that just provides a voice to a technology that actively helps users craft and project their authentic identity. Because for AAC users, the ultimate goal isn’t to sound perfect. It’s to sound like themselves.