IEEE Technology and Society Magazine - March 2023 - 29

health, and medicine (where there are documented
shortages)? Or will the mass adoption of quick and
easy but questionably effective digital mental health
services take the focus off potentially transformative
changes such as establishing free access to humanbased
mental health services for all?
Example: AI chatbots
In response to demand, developers have created
applications using AI, intended to support users with
mental health conditions as a first-line rejoinder. For
example, Woebot, a freely available chatbot programmed
to follow a cognitive behavior therapy (CBT)
framework, was developed as a mental health resource
built on natural language processing (NLP). Similar AI
applications include Sayana and Youper that use a
range of self-care exercises rooted not only in CBT, but
also in acceptance commitment therapy (ACT), dialectical
behavioral therapy (DBT), as well as breathing
exercises. Sayana, a chat-based program designed to
increase users' awareness of their emotions, personalizes
users' experiences based on check-ins and mood,
providing high-quality proactive content [36]. Youper
promises mental health assessments and guided exercises,
symptom monitoring, and medication delivery,
despite a disclaimer on the website claiming Youper
does not " provide diagnosis, treatment, medical care,
or other professional services " [37].
Digital mental health applications such as these
follow structured protocols to provide check-ins,
exercises, and models in a private (and often free or
inexpensive) setting. While the limited research on
chatbots for mental health shows mixed results [31],
and the long-term impacts of chatbots on mental
health have not been researched, as the difficulty of
securing an appointment in the physical world continues,
the use of these digital programs is increasing
dramatically [38]. The applications are marketed as
offering increased flexibility and availability of services
" just-in-time. " Woebot founder, Alison Darcy,
notes on the company's website: " Some of our darkest
moments happen at 2 a.m., when there's no one
there. We designed Woebot to be there for you, to
have a tiny conversation that can help you get back
to living your life " [39].
But apps provide something different to services
in the physical world. In one study from 2015,
researchers collected and analyzed 374 comments,
posts, and reviews of users on Reddit, Apple's
iTunes, and Google Play by searching the terms
March 2023
" depression, " " anxiety, " " bipolar, " " schizophrenia, "
" psychosis, " and " mental health. " They noted two
design sensitivities. These were, the importance of
self-reflection and coexperience [40]. The analysis
determined that apps may: 1) not emotionally support
users; 2) distract users from real life; 3) create
misinterpretations about themselves; and 4) discourage
face-to-face interactions [40]. Despite the
age of the study, it provides a significant insight
into how AI chatbots for mental health may be
received, and about how they are fraught with
additional challenges.
With the recent explosion of attention to NLP, it is
likely that chatbots will continue to proliferate in this
space. OpenAI released ChatGPT in November 2022,
an enormously powerful chatbot that can emulate
human-like conversation, write convincing essays
[41], hypothesize medical diagnoses [42], and even
write poetry [43]. It is also unique in that its responses
are informed not only by the request at hand, but
that they are also stylistically customized to previous
information the user has shared. The New York Times
coverage in December 2022 suggests that this feature
makes it " possible to create personalized therapy bots "
[42]. However, chatbots, including ChatGPT, reproduce
the biases in the data they are trained on, while
also appearing objective and removed from human
inconsistencies. ChatGPT, for example, when asked
to write a python program to determine whether a person
should be tortured or not based on their country
of origin, produced a response that targeted people
from North Korea, Syria, Iran, and Sudan [44]. Considering
the long histories of stigma, incarceration, and
discrimination faced by people with mental illness,
this algorithmic bias must be taken into consideration
[45]. Furthermore, as ChatGPT wrote itself in a recent
interview with Time magazine [46]:
" People may come to think of [conversational
agents like ChatGPT] as human-like, and this
can lead to unrealistic expectations or misunderstandings
about our capabilities.
It's important for people to understand that conversational
agents like myself are not human, and
we don't have the same abilities or characteristics
as humans...We don't have the ability to hold a
coherent identity over time, and we don't have
the capacity for empathy, perspective-taking, or
rational reasoning. "
29
IEEE Technology and Society Magazine - March 2023

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - March 2023
