Experts hope voice assistants can detect signs of dementia

‘Courtroom tests unreliable’

CONCORD, NH, Feb 17, (AP): An effort to use voice-assistant devices like Amazon’s Alexa to detect signs of memory problems in people has gotten a boost with a grant from the federal government.

Researchers from DartmouthHitchcock and the University of Massachusetts Boston will get a four-year $1.2 million grant from the National Institute on Aging. The team hopes to develop a system that would use machine and deep learning techniques to detect changes in speech patterns to determine if someone is a risk of developing dementia or Alzheimer’s.

“We are tackling a significant and complicated data-science question: whether the collection of long-term speech patterns of individuals at home will enable us to develop new speech-analysis methods for early detection of this challenging disease,” Xiaohui Liang, an assistant professor of computer science from the University of Massachusetts Boston, said in a statement. “Our team envisions that the changes in the speech patterns of individuals using the voice assistant systems may be sensitive to their decline in memory and function over time.”

John Batsis, a member of the team and associate professor of medicine at the Geisel School of Medicine at Dartmouth, said the system would help families better plan for care should someone develop a cognitive impairment.

“Alzheimer’s disease and related dementias are a major public health concern that lead to high health costs, risk of nursing home placement, and place an inordinate burden on the whole family,” Batsis said. “The ability to plan in the early stages of the disease is essential for initiating interventions and providing support systems to improve patients’ everyday function and quality of life.”

Batsis admitted this was a novel approach and that challenges lie ahead in developing a system he and the other researchers plan to eventually test in people’s homes.

System

The system, in theory, would aim to pick up changes in a person’s speech pattern, intonation and lexicon, he said. But researchers also would have to figure out how to make the system work for a myriad of languages, when there are multiple people speaking in the room, or when someone mumbles or doesn’t speak clearly.

“These are all pragmatic and practical issues,” Batsis said.

Should a system one day be sold commercially, researchers envision that patients, the family or caregivers would choose to enable the system on their voice assistant.

“A huge challenge is that of privacy,” he said. “You need to think about these things. Older adults that may be at risk or whose family member are concerned about this need to have buy in for that.”

Several experts, who were not part of the research, welcomed the focus of the research.

“Imagine if we had another tool to help diagnose this, and if that tool helped us detect it early,” Alicia Nobles, an assistant professor in the Department of Medicine at University of California, San Diego and the co-founder of the Center for Data-Driven

Health at the Qualcomm Institute, said in an email. She noted that detecting impairments early may be “crucial” to helping patients and their caregivers manage their care.

Sarah Lenz Lock, the senior vicepresident for policy at AARP and the executive director of the Global Council on Brain Health, also said the research looked promising.

“We need to assure that people’s privacy is maintained through the expanded use of technology in this way,” she said. “But speech patterns present a promising area for early screening of cognitive decline.”

Forensic science:

Courts are not properly screening out unreliable psychological and IQ tests, allowing junk science to be used as evidence, researchers have concluded. Such tests can sway judges or juries and influence whether someone gets custody of a child or is eligible for bail or capital punishment.

The scientists looked at hundreds of different psychological tests used in recent court cases and found that a third of those exams weren’t reviewed in the field’s most prominent manuals. Of those that were reviewed, just 40% were graded favorably. Nearly a quarter were deemed unreliable.

Variability

“There’s huge variability in the psychological tools now being admitted in US courts,” said Tess Neal, an Arizona State University psychology professor and co-author of the study published Saturday in the journal Psychological Science in the Public Interest.

“There’s a lot of stuff that looks like it’s junk and should be filtered out by the courts, but it’s not being filtered out,” said Neal.

Legal challenges to the validity of psychological tests occurred in less than 3% of cases, the researchers found.

“This paper is highly significant, in part because many people’s fates are determined by these tests,” said Dan Simon, an expert on law and psychology at the University of Southern California Law School, who was not involved in the research.

The new study is not the first critique of how science is used in the courts.

In 2009, the National Research Council released an extensive report on courtroom science that found that “testimony based on faulty forensic science analyses may have contributed to wrongful convictions of innocent people.”

The critique prompted calls for reform, and only partial progress has been made, said Simon.

“Courts are supposed to sift out the junk science from the good science, as laid out in the federal rules of evidence” – a set of national guidelines that require that “testimony is the product of reliable principles and methods.”

“But that’s not happening,” said Simon.

The new study examined 876 court cases in the US between 2016 and 2018, and found the most commonly used psychological test was the Minnesota Multiphasic Personality Inventory, which has generally positive reviews in the professional literature.