Helping the Helper: LLM-Assisted Problem Articulation for Older Adults Seeking Technology Support
This paper demonstrates that an LLM-assisted pipeline effectively mitigates communication barriers in older adults' technology support requests by rephrasing unstructured queries, which significantly improves solution accuracy and helper comprehension while introducing the first synthetic dataset of such queries.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the modern world, digital devices are no longer just conveniences; they are the gateways to banking, healthcare, social connection, and civic life. For many older adults, owning a smartphone or tablet is common, but the ability to use these tools effectively often fades when a new update arrives or an unfamiliar app appears. When technology fails, the path to a solution usually involves asking for help. This process relies heavily on communication: the person struggling must describe what is wrong, and the person helping must understand that description to offer a fix. However, the way older adults naturally describe technical problems often clashes with the rigid, precise language that computers and even human helpers expect. Aging brings changes to how the brain processes information and retrieves words, which can make explaining a glitch feel like trying to describe a dream while waking up. The result is often a conversation that goes in circles, leaving the older adult frustrated and the helper confused, sometimes causing the older adult to give up on the technology entirely.
A team of researchers at the University of Illinois Chicago set out to understand exactly how older adults describe these technology problems and whether artificial intelligence could act as a translator to bridge the gap. They began by asking twenty-seven older adults, aged sixty and above, to keep a diary of their technology struggles over eight weeks. Instead of forcing them into a lab, the researchers let them send messages via text, email, or WhatsApp whenever a problem arose, capturing the raw, unfiltered way people actually ask for help. The researchers found that these natural requests were often difficult to decipher. Some messages were filled with so much extra detail about family members or daily routines that the actual technical issue got lost in the noise. Others were too vague, missing critical details like what device was being used or what the screen looked like. Some requests included too many specific, irrelevant facts, while others left out the most important context needed to solve the puzzle. The researchers identified four main patterns in these confusing messages: excessive wordiness, too much specific detail that didn't matter, not enough detail to be useful, and missing information entirely.
To see if technology could fix this communication breakdown, the researchers built a system using a large language model, a type of advanced computer program trained on vast amounts of text to understand and generate human language. They designed this system to act like a patient, skilled helper. When an older adult sent a messy or unclear message, the system first analyzed it to figure out what was missing. It then generated simple follow-up questions to gather the necessary context, such as asking which device was being used or what the screen looked like. Once the user answered these questions, the system took the original messy message and the new details to rewrite the problem into a clear, concise sentence that a search engine or a human helper could easily understand. Finally, the system generated a step-by-step solution. The researchers tested this pipeline with two groups. First, they asked forty-eight younger adults, who often serve as the "tech support" for their older relatives, to read both the original messy messages and the rewritten, clear versions. The results were striking: the younger helpers understood the rewritten messages nearly every time, whereas they struggled with the original ones. They also reported feeling much more confident in their ability to help when the problem was clearly stated.
The researchers then tested the system directly with thirty-four older adults to see if the solutions the computer generated were actually useful to them. These participants read the follow-up questions the system asked and the final solutions it provided. The older adults found that they could answer the system's questions with high confidence and felt capable of following the instructions given in the solutions. The study showed that when the computer helped clarify the problem, the chance of finding the correct solution jumped significantly. Before the computer helped, the system found the right answer only about a third of the time. After the computer rewrote the question to be clear and complete, it found the right answer more than two-thirds of the time. This improvement suggests that the barrier was not the older adults' inability to use technology, but rather the difficulty of translating their experience into a format that current support systems could understand.
To ensure this work could help others in the future, the researchers created a new dataset called STAQ. Because it is difficult and expensive to gather thousands of real-life examples from older adults to train new computer programs, the team used their findings to generate a large collection of synthetic examples. These are computer-generated messages that mimic the specific communication styles, word choices, and confusion patterns the researchers observed in their real study. They verified that these fake messages looked and felt just like the real ones, capturing the same mix of too much detail and missing context. This dataset allows other scientists to train and test their own artificial intelligence tools to make sure they work well for older adults, rather than just for younger, tech-savvy users. The study concludes that older adults are not failing at technology; rather, the tools designed to help them are failing to understand how they speak. By using artificial intelligence to translate the natural, sometimes messy way older adults describe their problems into clear instructions, we can build a support system that respects their communication style and restores their independence.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.