The Promise of Artificial Intelligence in Supporting an Aging Population

Introduction: Why a Focus on Aging Adults?

The number of people in the United States aged 65+ will increase to about 98 million by 2060, with the fast-growing cohort of the “oldest old” (85+) projected to number 14.6 million by 2040 (Administration for Community Living, 2020). While population aging represents a great achievement, it also presents challenges for our healthcare and social support systems, and the economy. Although many older adults enjoy healthy and productive lives, increased age is an independent risk factor for the development of diseases and chronic conditions such as heart disease, diabetes, and arthritis. In 2018, ∼52% of community dwelling old adults living in the United States had one chronic condition, 26% had multiple chronic conditions, and ∼34% had some type of disability such as a vision, hearing, or mobility impairment. Chronic conditions and disabilities can negatively impact the performance of everyday activities and ultimately threaten independence. About 21% of those 85+ need help with everyday activities as do about 8% of those 75–84 years of age (Administration for Community Living, 2020).

The incidence of cognitive impairments also increases with age. Currently, about 6.2 million adults have Alzheimer’s disease and unless a cure is found, this number is projected to increase to 12.7 million by 2050. About 17 million adults have Mild Cognitive Impairment (MCI) and ∼11% have Subjective Cognitive Impairment (SCI), the self-perceived worsening of memory or thinking (Alzheimer’s Association, 2021). Cognitive health is critically important to functional abilities and independent living. Normative age-related changes in cognition, such as declines in memory and processing speed can negatively impact everyday activity performance, new learning, and the ability to adapt to changes in contexts and life circumstances. New learning is a critically important aspect of independent living given the continual development of and changes in everyday technology applications.

Recently, attention is also being directed toward social isolation and loneliness as a public health risk for older adults as changes in life circumstances, such as retirement, loss of partners or friends, financial circumstances, health issues, and mobility challenges make older people vulnerable to becoming isolated and potentially lonely. Recent data indicate that approximately 24% of community dwelling older adults are considered to be social isolated and 43% report feeling lonely (National Academies of Science Engineering and Medicine, 2020). Social isolation and loneliness present significant health risks and have been linked to cognitive declines; lower quality of life; a heightened risk for physical and mental health problems; functional declines; and mortality (Aylaz et al., 2012; Ellis & Hickie, 2001; Fratiglioni et al., 2000; Heinrich & Gullone, 2006; Holt-Lunstad et al., 2010). Data from the English longitudinal study also indicate that loneliness is a significant, independent predictor of dementia (Rafnsson et al., 2020). In fact, a recent National Academies of Science Engineering and Medicine (NASEM, 2020) report that focused on social isolation and loneliness in older adults pointed to the need for more intervention research to help ameliorate problems with isolation and loneliness and suggested that technology applications hold promise as an intervention approach.

Another concern associated with the aging of the population is the availability of support for future generations of aging adults. Currently, family members provide the majority of support for older adults. Estimates indicate that about 18 million individuals are providing care for an older adult such as a spouse, relative, friend, or neighbor. Importantly, this estimate does not include nursing home residents (NASEM, 2016). For example, about 60% of adults aged 85–89 years, a fast-growing cohort, receive help from a family caregiver because of a functional limitation or a health problem; those needing some of help increases to 76% among those aged 90 and older (Freedman & Spillman, 2014). Caregivers engage in a wide array of activities such as care coordination, provision of instrumental and emotional support, and assistance with medical activities (e.g., medication management). In fact, family caregivers are often considered as key members of an older adult’s care team. However, changes in family and social structures, such as geographical distances among family members, couples deciding to remain childless, and individuals deciding to remain single, coupled with an aging population, raise questions regarding how care will be provided to future cohort of older adults. The recent National Academies of Science, Engineering, and Medicine on family caregiving (2016), notes that given current demographic trends future cohorts of older adults needing help will likely exceed the capacity of family caregivers. There is also a shortage of paid care workers to meet the needs of older adults (Stone & Harahan, 2015).

In the following section, we provide examples of how AI applications can provide support to older adults. As the field of AI is broad and encompasses many application areas and new developments in this domain are continually emerging, this is not intended as an exhaustive review of AI or the AI and aging literature. Our objective is to illustrate how AI applications can be used to support older adults. We also discuss factors that influence the successful uptake of these applications by aging adults.

Artificial Intelligence and Aging Adults

We begin with a simple definition of Artificial Intelligence (AI) to guide the remainder of the discussion. In simple terms, AI refers to computer or machines that are programmed to think and act like humans, recognize objects, understand language, make decisions, solve problems, and perform functions. Machine learning, a subset of AI, involves methods of enabling an algorithm to learn from datasets or updating an algorithm based on new data (for more rigorous definitions of AI and ML see Russel and Norvig (2010) and Alpaydim (2010), respectively). Machine learning algorithms can be used to make predictions about future outcomes, classify images or text into predefined or automatically generated categories, recommend things like films or songs based on past preferences, and even teach a computer to learn to perform a particular task (such as playing a video game). The performance of an AI algorithm depends on the model selected, available data, and the selected input features used to predict an outcome. Supervised learning approaches require labeled datasets, such as diagnosis of an impairment versus unimpaired and clinical notes to train an algorithm to independently classify data and accurately predict outcomes (e.g., spam email). A common application of this approach is image recognition, which is an important feature of AI embedded home security systems.

Unsupervised Learning approaches use, analyze, and cluster unlabeled datasets (e.g., the data is not categorized as impaired or unimpaired) to help discover patterns or groups of data that share some common characteristic.

Deep learning (DL) is a subset of machine learning that relies on layers of artificial neural networks (ANNs) to learn features from raw data, automatically. ANNs were originally inspired by models of how the human brain works (Rosenblatt, 1958). Deep learning models require large datasets. An example of a deep learning algorithm is a recurrent neural network, which is used for applications such as Siri or image captioning (IBM Cloud 2020). Finally, natural language processing (NLP) refers to how computers can be used to understand and manipulate natural language, such as speech or text to perform to perform desired tasks (Chowdhury, 2003). For example, NLP might be used to examine speech patterns to model changes in cognitive functioning (Graham et al., (2020) or to detect social isolation and loneliness among older adults (Badal et al., 2021).

Clearly, AI systems vary widely in complexity and functionality. Common applications include expert and decision support systems, disease mapping applications, smart assistants, search algorithms, virtual reality (VR) applications, artificial intelligent robots, and digital assistants such as the Amazon Echo or Siri, and self-driving cars. AI systems and applications are increasingly being directed toward healthcare and providing support to aging adults to help foster independent living and enhanced quality of life. AI applications can also aid in the clinical care of older patients and hold the potential to develop precision models that are both personalized and conceivably more accurate than traditional clinical care using vast amounts of real-world multimodal data about patients. In the following section, we review examples of these applications and research examining AI applications and aging adults.

Design Considerations for Aging Adults

As noted throughout this discussion, older adults represent an important user group when considering the design of AI technologies. AI systems that do not consider the needs, preferences, and abilities of older users will likely fail in their ability to support this user group and reach their full potential. In addition, failure to consider older adult users in the design process may results in systems that are underutilized by this population and may result in user errors with serious consequences.

Although there is substantial variability among older adults in needs, preferences, and capabilities, there are normative age-related changes in abilities that have relevance to the design of AI systems. It is beyond the scope of this article to describe these changes in detail so only some of the relevant changes will be highlighted (see Czaja et al., 2019 and Boot et al., 2020 for a more detailed discussion of this topic). Age-related changes in sensory systems such as vision and audition that have relevance for design. For example, due to changes in visual acuity and visual processes, older adults have a decreased ability to perceive detail, adapt to changes in levels of illumination, perceive color, and have a heightened sensitivity to glare. These changes have significant implications for the design and placement of displays and text-based instructional materials. Age-related losses in auditory acuity make it more difficult to hear sounds of higher frequency and distorted speech especially under noisy conditions. Age-related auditory changes have implications for the design of alarm and speech systems. There are also numerous age-related changes in cognition that have implications for the design of AI systems. These include changes in working memory, speed of processing, aspects of executive functioning, and attention. In general, older adults tend to process information more slowly, have challenges with working memory and prospective memory (remembering to do something in the future) tasks, have greater difficulty managing multiple tasks simultaneously, and tend to take longer to learn novel skills, which can impact the amount of training and support necessary for the use of an AI system. Systems that place high demands on working memory or have inconsistent functionality can also negatively influence the performance of older adults. Aging adults may need more training and instructional support than younger adults and it is important that training and instructional programs are designed to accommodate the needs of older learners (see Czaja & Sharit, 2013). Despite these changes in cognitive abilities, it is important to note that many of these changes can be mitigated through compensatory strategies and environmental supports (e.g., checklists and reminder systems) and in fact, as discussed, AI applications such as cognitive coaching systems may help support age-related cognitive supports. Aging is also associated with plasticity and older adults are willing and able to learn to use new technologies. Finally, physical capabilities may change with age. In general, system designers should consider age-related changes in physical dimensions (e.g., height), movement control, strength, and endurance.

Accommodating age-related changes in the design process requires an iterative user-centered design approach that actively involves older adult users in the design process. Including users in the design process involves identifying user groups; understanding their needs, characteristics, and preferences; and involving representative users in the actual design and evaluation of products and systems. As noted earlier, the older adult population is very diverse and varies on numerous dimensions including abilities, preferences, needs, skill level, prior experiences, and socio-demographic characteristics and thus when including older adults, it is important to include diverse samples of aging adults. Our group has found that aging adults are generally quite willing to engage in technology-based usability research, if they are informed about the importance of their input to the design of technology applications, the potential value of these applications with respect to independence and quality of life, and the protocol for participating is convenient and flexible. For example, this might involve conducting a study in a community center or providing participants with transportation to and from the research setting.

There are numerous methods for including older adults in design and evaluation of systems include surveys, focus groups, usability and pilot trials, observational studies, and randomized control designs. The choice of method depends on the stage in the design process and the design/research questions. There are also several important issues to consider when involving older adults in design research such as making the research environment as stress-free and comfortable as possible, making sure questionnaires and instructions are readable and that oral instructions are understood, and being mindful of testing session duration (especially with those in the older cohorts) and pacing schedules. It is important to minimize participant burden and fatigue. Other issues that need to be considered include how the study material will be delivered and in what context and logistical issues associated with data collection (see Czaja et al., 2019 for a more complete discussion of these issues).

Challenges and Conclusions

The aging of the population gives rise to the need to develop innovative strategies to support older adults especially those in the older cohorts. Clearly, AI has a great deal to offer in terms of providing support for and potentially improving the independence and quality of life for older people. Currently, there are still challenges associated with the use of these systems such as system complexity, privacy and security concerns, sampling and data management issues, and usability. For the benefits of AI applications to be realized, these systems must be available, useful, useable, safe, and reliable for diverse older adult user groups. We must also ensure equitable implementation of these systems and that training and technical support are available. It is also important to recognize that the user group for these systems may also include family caregivers and healthcare professionals. Access to reliable internet connections also remains a challenge for many older adults such as those in lower Socio-Economic Strata (SES) or in rural locations. Although the rate of technology adoption is increasing among older adults, there are still age-related gaps in the uptake of technology. This has vast implications given the increased diffusion of technology within the service sector and the healthcare arena. Policies and programs need to be in place to ensure equitable access to technology and training and technical support.

AI systems must also be designed using a user-centered design approach, which considers the needs and characteristics of the user. Adopting this type of approach requires a multidisciplinary team approach that includes system designers, human factors engineers, behavioral scientists, gerontechnologists, and active involvement of diverse user groups during the design process.

AI applications are changing at an enormously fast pace, and new systems and applications aimed at older adults are continually emerging. There are multiple opportunities for research in this area. Table 1 provides a summary of some general recommendations of needed research. It is beyond the scope of this paper to provide specific research recommendations for specific types of AI applications (e.g., identifying factors that influence the degree of presence in VR programs). In general, large-scale randomized trials demonstrating the benefits as well as the limitations of AI applications for older adults, family caregivers, and providers are needed. Research also needs to be directed to understand the benefits and effectiveness of AI applications in ecological contexts with diverse populations on health and well-being outcomes. These studies need to include metrics of cost effectiveness, feasibility, acceptability, and examine use of these systems over time. In addition, data on older adults needs to be included in AI data sets so that older adults are represented in machine learning training and test models. These data sets must capture the diversity of the older adult population and include data on aging adults with and without impairments. Finally, careful attention needs to be given to understanding the optimal balance between technology and human interactions and human expertise. For AI to achieve its true potential, there also needs to be greater interoperability (the ability of devices or systems to connect and communicate with other devices and systems) among different technologies and devices. Ultimately, we need to identify what applications to deploy and how to deploy these systems to best support an aging population. This is a ripe area for human factors engineering, given the focus of human factors on user-centered design, and designing systems to accommodate user needs, preferences, and abilities.

OPEN IN VIEWER

Table 1.

Summary of Recommendations for Future Research in Artificial Intelligence for Aging Adults.

Research is needed to
1. Identify AI applications that support the needs and preferences of aging adults for specific tasks/activities. This research should extend beyond the healthcare domain and include leisure, social, educational, and cognitive engagement
2. Identify how to best design features of AI applications (e.g., interface) for aging adults
3. Identify the needs, preferences, abilities, and training/support needs of aging adults with and without impairments with respect to AI applications. This necessitates including diverse, representative populations of aging adults in development and usability studies
4. Understand how to best implement AI applications with older adult populations in diverse contexts
5. Understand how to best implement and integrate AI applications among multiple user groups such as older adults, family member, and clinical care providers
6. How to integrate AI healthcare applications such as sensor and monitoring systems within clinical care environments and systems
7. Understand factors that influence the adoption and uptake of AI applications by aging adults
8. Understand the benefits and effectiveness of AI applications in ecological contexts with diverse populations on health and well-being outcomes over time. This research must include both large scale randomized trials and pragmatic trials
9. Include data from diverse aging adults in AI data sets
10. Examine the cost/benefits of AI support systems and other issues such as privacy and trust
11. Identify metrics that meaningfully assess the impact of the use AI among older adults on outcomes such as independence and well-being as well as metrics that reliably capture use of the AI applications
12. Examine AI applications that are efficacious in settings beyond independent living environments such as continuing care settings as well as how to best implement these applications within these settings