Bridging the generation gap in generative AI
Generative artificial intelligence (AI) is poised to redefine the way people live and work, but what factor do age and data have to do with it? As it turns out, a lot.
While 65% of generative AI users are Millennials or Generation Z, 68% of non-users are Gen X or Baby Boomers, according to Salesforce. Arizona State University (ASU) Associate Professor Yezhou “YZ” Yang explained that digital content contributed by senior citizens is inherently lagging. Yang, an associate professor in the School of Computing and Augmented Intelligence, part of the Ira A. Fulton Schools of Engineering, shared that this setback makes it challenging to directly deploy large language models (LLMs) to meet the technological needs of these populations.
Therefore, Yang and his research team are using ChatGPT Enterprise, as part of ASU’s AI Innovation Challenge for educational outreach and research purposes. First, Yang is working to educate senior citizens across ASU’s lifelong learning institutions to support discovering generative AI while gaining key research findings.
Then, he and his research team are validating what Yang learns through these outreach efforts by exploring techniques to adapt the language of large language models (LLMs), like ChatGPT, to make them more user-friendly for senior citizens. This would reduce the bias in deployed AI models, which are more attuned to the digital behaviors of younger generations, and minimize the chance of leaving senior citizens feeling marginalized in the digital landscape.
Yang recently delivered a presentation called “AI and the Joy of Living” to approximately 100 senior citizens in the Lifelong Learning Auditorium at Mirabella at ASU. He gave residents of the retirement community a primer on the probability theories behind the development of AI, likening the process of training LLMs to what we all experience when we communicate.
“If I ask someone to marry me and the response begins with, ‘You are such a nice guy,’ I can probably predict that a rejection is coming,” he said, noting that we all learn to make predictions about what comes next when we interact with others.
Attendees at Yang’s presentation asked questions about how artificial intelligence works. For example, when Yang’s learners use OpenAI tools, they shared that many of the responses from the LLMs and text-to-image tools include phrases they are not familiar with. “That’s because the materials they are accustomed to are not being uploaded to the internet,” said Yang.
One activity Yang challenges his senior students to do is use generative AI to write a speech for their grandchild’s wedding. “They are amazed at how this tool can create vivid and moving speeches,” said Yang. “What would have taken them a long time now can take five minutes.” Then, Yang asks the seniors to use generative AI to give life advice to their grandchild who is new to ASU. With this part of the exercise, the tools tend to fall short, using language that could cause potential miscommunication between generations.
Mirabella resident Janet Merritt agreed, saying that she had tried ChatGPT to improve her own writing process. “I generally preferred my own edits,” she said.”But I was so impressed by the speed of the tool that I would love to try it again in the future.”
Yang went on to share that, from a fairness perspective, the data lacks representation from digitally underrepresented groups: seniors and isolated groups who are less active on the internet.
Yang and the research team are approaching this issue using two methodologies: examining whether concepts can be erased from LLMs and exploring robust and efficient training of generative AI models. Can a generative AI tool be trained to use terms that older generations are more familiar with? If so, Yang and the team can formulate rules and train the model to better communicate and adapt to a senior user audience.
“We shouldn’t take for granted that seniors understand generative AI tools – this language and understanding barrier needs to be addressed,” said Yang. “Most seniors are more welcoming to new technologies than I expected. This is truly a group of people that AI technology companies need to spend more time paying attention to.”
The Mirabella residents are ready to explore these advancements. Janet Leopold left Yang’s presentation determined to investigate ASU’s AI educational offerings. “I want to learn as much about this new technology as I can,” she said.
While 65% of generative AI users are Millennials or Generation Z, 68% of non-users are Gen X or Baby Boomers, according to Salesforce. Arizona State University (ASU) Associate Professor Yezhou “YZ” Yang explained that digital content contributed by senior citizens is inherently lagging. Yang, an associate professor in the School of Computing and Augmented Intelligence, part of the Ira A. Fulton Schools of Engineering, shared that this setback makes it challenging to directly deploy large language models (LLMs) to meet the technological needs of these populations.
Therefore, Yang and his research team are using ChatGPT Enterprise, as part of ASU’s AI Innovation Challenge for educational outreach and research purposes. First, Yang is working to educate senior citizens across ASU’s lifelong learning institutions to support discovering generative AI while gaining key research findings.
Then, he and his research team are validating what Yang learns through these outreach efforts by exploring techniques to adapt the language of large language models (LLMs), like ChatGPT, to make them more user-friendly for senior citizens. This would reduce the bias in deployed AI models, which are more attuned to the digital behaviors of younger generations, and minimize the chance of leaving senior citizens feeling marginalized in the digital landscape.
Introducing seniors to generative AI
Yang is dedicated to teaching seniors and lifelong learners about generative AI tools. Through Osher Lifelong Learning Institute at ASU and Mirabella at ASU, Yang helps seniors become more comfortable with the tools and collects their opinions as they become familiar with the technology.Yang recently delivered a presentation called “AI and the Joy of Living” to approximately 100 senior citizens in the Lifelong Learning Auditorium at Mirabella at ASU. He gave residents of the retirement community a primer on the probability theories behind the development of AI, likening the process of training LLMs to what we all experience when we communicate.
“If I ask someone to marry me and the response begins with, ‘You are such a nice guy,’ I can probably predict that a rejection is coming,” he said, noting that we all learn to make predictions about what comes next when we interact with others.
Attendees at Yang’s presentation asked questions about how artificial intelligence works. For example, when Yang’s learners use OpenAI tools, they shared that many of the responses from the LLMs and text-to-image tools include phrases they are not familiar with. “That’s because the materials they are accustomed to are not being uploaded to the internet,” said Yang.
One activity Yang challenges his senior students to do is use generative AI to write a speech for their grandchild’s wedding. “They are amazed at how this tool can create vivid and moving speeches,” said Yang. “What would have taken them a long time now can take five minutes.” Then, Yang asks the seniors to use generative AI to give life advice to their grandchild who is new to ASU. With this part of the exercise, the tools tend to fall short, using language that could cause potential miscommunication between generations.
Mirabella resident Janet Merritt agreed, saying that she had tried ChatGPT to improve her own writing process. “I generally preferred my own edits,” she said.”But I was so impressed by the speed of the tool that I would love to try it again in the future.”
Why seniors are at a disadvantage
Yang explained that LLMs digest digital data that has been recorded, hosted, and collected on the internet since the 1980s and 1990s. Data creation began to boom in the late 2010s when short videos and social media became increasingly popular. “The availability of the data drives the success of generative AI,” said Yang. “But there is a caveat: the data is largely biased toward digitally active people; we call them Gen Z (younger generation).”Yang went on to share that, from a fairness perspective, the data lacks representation from digitally underrepresented groups: seniors and isolated groups who are less active on the internet.
Research to identify data gaps
In addition to educating seniors, Yang and his research team are conducting technological research to identify where the gaps are. Their challenge is to explore how to adapt the LLMs with a limited amount of well-formulated data to advance models and meet the needs of broader audiences while ensuring the technology is efficient, flexible, robust and secure.Yang and the research team are approaching this issue using two methodologies: examining whether concepts can be erased from LLMs and exploring robust and efficient training of generative AI models. Can a generative AI tool be trained to use terms that older generations are more familiar with? If so, Yang and the team can formulate rules and train the model to better communicate and adapt to a senior user audience.
Education and research to advance social good
Not only is Yang examining how technology is used and evolves over the lifetime of learners, but he is also exploring solutions to make technology more accessible to those in need, which aligns directly with ASU's charter.“We shouldn’t take for granted that seniors understand generative AI tools – this language and understanding barrier needs to be addressed,” said Yang. “Most seniors are more welcoming to new technologies than I expected. This is truly a group of people that AI technology companies need to spend more time paying attention to.”
The Mirabella residents are ready to explore these advancements. Janet Leopold left Yang’s presentation determined to investigate ASU’s AI educational offerings. “I want to learn as much about this new technology as I can,” she said.