Recent advancements in the field of computer science have produced virtual assistants (VAs) capable of aiding those who suffer from mental illness in a more personalized and “natural” way.
In integrating artificial intelligence (AI) with treatment for mental illness, the ongoing challenge has been to keep the personalities of VAs consistent with the preferences of patients.
“Certain personalities or emotions within a virtual assistant appeal more to certain individuals,” Steven Feng, an undergraduate research assistant in UW’s David R. Cheriton School of Computer Science, said.
The demand for VAs in the mental health sector has seen a drastic increase in recent years due to limited human resources for people who require professional aid such as mental health therapy, being able to automate some of the aspects of treatment would be greatly beneficial.
It would reduce wait times for patients and make the process more affordable.
UW faculty of Mathematics researchers Feng, Jesse Hoey, an associate professor, and Aaron Li, research assistant in the Cheriton School of Computer Science, have pioneered a new way to overcome this hurdle with the development of program named SMERTI (pronounced Smarty).
SMERTI allows VAs to adapt to people based on the relationship and present situation by processing natural language and emotional signals.
This allows for the development of VAs that better connect with people they are used to help.
“Enabling virtual assistants, based on the situation, to tweak the words or sentences used to match the personalities could lead to on-demand chatbots available to talk to people with mental illness and cognitive disabilities whenever they require,” Feng said.
SMERTI works employing various AI tools such as similarity masking, entity replacement, and text infilling.
Its function is to take texts in their original, specific form and give it relevance to a variety of contexts by selectively substituting in the appropriate words.
For example, it would take a text like ““it is sunny outside; I know you hate to, but you must wear sunscreen” and change it to “it is rainy outside; I know you hate to, but you must bring an umbrella”.
The results of the research are truly groundbreaking.
As evidenced by an evaluation where the researchers rated the sentences created by SMERTI on its “fluency, sentiment preservation, and content exchange,” Feng said.
Feng explained that the new software outperformed all baseline models.
Most notably, the software had higher proficiency in replacing text to adapt to new semantics.
“What we were mainly focused on was the fluency and semantic exchange aspects to show that the task is possible and measure the emotional preservation, which was decently high,” he said.
The next step for researchers is to figure out how to maintain a virtual assistants personality or persona fairly stable rather than just preserving its emotional aspect in texting,.
The emotional aspect is far more complicated and would allow the virtual assistants to have a more consistent personality.