Survey Shows AI Voice App Usage on the Rise Amid Growing Concerns About Bias
Undoubtedly, ChatGPT and other voice-driven AI technologies have dominated both the media and social media landscapes this year. Overwhelming popularity in experimenting with the technology to answer questions (including strange ones like, how to build a magical potato) and attempts to complete a myriad of tasks — engineering computations, school essays, business contracts, ideas for birthday parties, marketing copy, song lyrics, and whatever else the human mind can dream up — has yielded some interesting, and in some cases, frustrating results.
To better understand sentiment around, interest in using, and causes of poor experiences with chatbots (including ChatGPT), interactive voice response (IVR) services and conversational assistants, Applause conducted its second annual AI and Voice Applications Survey. Wrapped up in early March this year, the survey collected responses from more than 5,000 digital quality testing professionals from around the world.
Satisfaction with ChatGPT
In light of generative AI’s growing popularity and ChatGPT’s rising engagement, we questioned survey recipients this year about their experiences with the chatbot. Overall, their sentiment was positive. Nearly 98% of respondents said ChatGPT had understood their questions and provided helpful answers.
When questioned about satisfaction with ChatGPT, 91% of respondents said they were either satisfied (78%) or neutral (13%) about their interaction. Additionally, a resounding 83% said that ChatGPT or similar technology would be helpful in completing their business- or work-related tasks.
Sentiment With Overall Chatbot Experiences
However, when asked about their overall experiences using chatbots, nearly one-third of the respondents said they are dissatisfied with chat functionality overall. The top reasons for dissatisfaction were the same as in the 2022 survey:
I could not find the answers I was looking for
The chatbot did not understand what I was asking
The chatbot directed me to irrelevant information
The chatbot wasted my time (did not add value) before connecting me with an agent
One-third also indicated that they would use chatbots more if the responses they received to voice commands and typed questions were more accurate.
Meanwhile, 85% said they expect to interact with chatbots on websites or apps so they don’t need to call for service or to have questions answered. This year’s survey indicated a slight decrease in respondents reporting their preference to wait for a human agent (48%, down from 51% in 2022) rather than use a chatbot. The remainder said they prefer immediate access to an automated touch tone response system or prefer an automated virtual service representative that responds to voice commands.
So, despite the group’s expectation of access and an implied willingness to use the technology, nearly half said they would opt for human interaction instead.
When asked how chatbots could be improved, the top three responses were:
More relevant responses (23%)
Better understanding of language variations (17%)
Responses more specific to user (15%)
Concern About Inherent Bias
To function properly, AI algorithms require huge amounts of high-quality, non-biased training data. However, bias often occurs with AI-driven applications because the data used to train them is insufficient, overly homogenous, or otherwise limited. This results in incorrect or unexpected results.
Nearly 86% of the survey respondents said they are concerned about the likelihood that inherent bias could affect the accuracy, relevance or tone of AI-generated content and chatbot responses.
Quality is Key in Customer Experience
In competitive markets and uncertain economic conditions, creating high-quality customer experiences is vital. Companies that invest in improving their AI and voice applications will have the advantage of providing convenient tools their customers want and expect, and ensure they are working properly and are providing reliable, useful responses every time.
For more on proper data collection, read our post: Using Questionable Datasets to Train AI Could Come With High Costs.