November 14, 2023 • 29 mins
Join host Graeme Klass as he sits down with aspiring data scientist Teena Sahu, the mind behind an AI software tool that can identify signs of depression. Learn about the intricate web of machine learning and brain-computer interfaces (BCI) that underpin Teena's groundbreaking work. Discover how her innovative approach to AI technology is bridging the gap between the digital and emotional worlds, shedding light on the profound connection between technology and human well-being.
This discussion will leave you with a deeper appreciation for the profound impact AI can have on our lives.
Learn more about how Intel is leading the charge in the AI Revolution at Intel.com/AIperformance
See omnystudio.com/listener for privacy information.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:08):
I know I need to get ready for work, but
I'm just so so tired. Maybe I can skip work
to take a nap justice once. What's that? Oh, my
depression monitors detecting early signs of a depressive state. I
thought it was just tired, but this might be a
bigger issue. Let me set in an appointment with my therapist.
(00:40):
Hey there, I'm grain class and this is technically speaking
an Intel podcast. The show is dedicated to highlighting the
way technology is revolutionizing the way we live, work and move.
In every episode, we'll connect with innovators in areas like
artificial intelligence to better understand the human centered technology they've developed.
Mental health care solutions remain underinvested in many communities around
(01:04):
the world, yet so many suffer from issues that they
don't even know. They have a lot of these conversations
around health care, hinge on making therapy more accessible to
those in need. However, it can be difficult to determine
that one is experiencing depression or mental health crisis. Artificial
intelligence is at the forefront of many different advancements in healthcare,
(01:26):
but today we are going to dive into how it
is working to make mental health care more accessible. To everyone.
In order to do that, I have to introduce you
to a special guest joining me now is Tinasawho. Tina
Saho was a high school student when she started exploring
coding and engineering with AI. She never considered herself to
(01:47):
be overly interested in technology to start. However, she took
her natural curiosity and eventually was invited to be a
part of Intel's AI for Youth Pilot program, where she
developed a tool that uses AI to detect and predict
patterns of depression with around eighty percent accuracy. Since then,
she's been awarded the Calm Trust Fellowship for Women. She
(02:08):
currently attends to Ayabata College at the University of Delhi,
where she's completing a bachelor's in computer science and exploring
more opportunities to use AI as a tool for mental
health and more.
Speaker 2 (02:19):
Welcome to the show, Tina, Hi, thank you for inviting
me to this show, and I'm feeling more than privileged
to be a part of this podcast. So numbers stay
to everyone.
Speaker 1 (02:33):
I'm really interested in your story of how you got
into the STEM field, what prompted your interest in that
field and also in AI.
Speaker 2 (02:42):
To begin with, till my tenth class, I was a
student or a person who was always against technology, like
I always found in fact that the negative impacts that
technology brings and whatever ethical concerns are there, they cannot
be solved and they are simply too much to be
on the side of technology. But I remember in twenty nineteen,
the AAFO Youth program was launched and it was launched
(03:03):
in our school, which is Salvangal Senior Secondary School. So
when I participated in that program, I got to know
what artificial intelligence is, how many astonishing and stounding possibilities
AI can unlock and has already unlocked. And therefore, this
program pivoted my you know journey, my career, my professional journey,
and my cademic journey from being a non tech student
(03:24):
to becoming a tech student finally pursuing computer science as
my graduation.
Speaker 1 (03:29):
Okay, was there a particular topic or person that really
did spark that interest? What was the particular topic that
actually got you really fired up?
Speaker 2 (03:40):
So in my school there were two teachers who actually
you know, molded my thinking and they actually infused this
critical thinking aspect in me and they opened me to
the world of possibilities that science, that technology and engineering offers.
And I would like to take their name as well
as a you know, token of respect. So it's one
that Mam and Serbiam they empowered me. They encouraged me
(04:02):
to think beyond what I see.
Speaker 1 (04:04):
And that led you to the Intel's AI for Youth program.
What sort of projects have you worked on? Are you
working on right now?
Speaker 2 (04:14):
This isn't my participating in this program. I build Happiness Guru,
which is a model that predicts depression. Apart from that too,
I was also a part of Utul Tinkling Labs. Through
the Detail Lab of our school. I basically got to
know about Intel's AfOR Youth program only and in there itself,
I build a few projects and one of them was
Happiness Guru.
Speaker 1 (04:34):
And with the Happiness Crew, is that related to the
depression detection research you've been doing? What kind of prompted
you to look in that direction in the field of
depression and then using technology? Because generally, speaker, we don't
associate technology with treating depression or even detecting depression. So
(04:57):
what was the spark for you there?
Speaker 2 (05:00):
So while I was in this program, I was in
Class ten and my Plus ten results were out and
they were not as much as I expected, and I
went into a phase of depression because I associated myself
worth with the Marxi score, so my own personal experience
of dealing with depression. And then at that time, you
know that society rates were very alarming among the youth,
(05:22):
especially aged between fifteen to twenty nine. And we found
that the driving forces behind you know these when these
societies were you know, peer pressure, overburdening academics, financial stress,
and too much expectations that we have from youth, you know,
especially if you talk about teenage and someone who is
in between eighteen to twenty five. So on researching, we
(05:42):
found that these societ rates are very alarming, they're very distressing,
and these are the leading cause of the depression or stress.
And thereby we thought that we must come up with
some solution that can basically help us predict which person
is going through depression, and that to in a very
human friendly manner, not making someone uncomfortable with the kind
of procedures or with the kind of system we have.
(06:03):
So these were the I would say, the enablers that
led our team building this solution.
Speaker 1 (06:09):
And in terms of the happiness grew app can you
just explain how it actually works, you know, to try
and detect the early signs of depression.
Speaker 2 (06:19):
First of all, it's our web based application. While building
this project, the queue that we took, you know, to
build the entire model was that after our research, we
got to know that a person's vocabulary can be a
mirror into their mental state. And taking this as the queue,
we build this project which tries to analyze the emotional
(06:39):
quotient of a person of a user through their facial
expression and then their textual responses that the user is
going to provide to the AA machine. So the working
of the project is divided into three steps. The first
step is emotion detection stage, and in this stage you
basically need to stand in front of your laptop or
whatever device you are using this web application, and then
(07:00):
it detects your current mode, whether you're happy or sad,
you're neutral, angry. Then the next step is that user
is asked to answer nine questions and there's a scale
of relevance and then they need to select how much
relevant or how much they are able to relate this
to this situation. Then, after these two steps, a threshold
score is generated which gives the initial lead. If the
(07:21):
person is stressed or not, and if the score is
below the threshold that we have said, the person is
predicted as happy, while in the other case, user is
taken to the third step, which is the final step.
And this step consists of four descriptive questions which he
or she can use as a platform to went out
all his or her feelings and thoughts. So whatever answers
user will give to these four descriptive questions, these answers
(07:44):
will be used as a basis of classification. Then the
machine will predict whether the user is depressed or not.
So this AI machine, whatever you know input we are
giving in this step. There's a model namely SVM, which
is support vector machine. It's a non contextual classification model.
It is basically used to classify things. And then we
are using this model on the kind of you know,
(08:04):
language or keywords that are used in the answers. And
then accordingly the results are given out that whether the
person is stressed or not, and if the person is stressed,
automatically the person is consulted to the concerned authorities or
counselor otherwise the person is predicted as happy or not stressed.
Speaker 1 (08:22):
Detecting and treating mental health is something with which many
societies around the world struggle. According to the World Health Organization,
approximately two hundred and eighty million people in the world
suffer from depression and more than three hundred million are
living with anxiety. Many people with these mental health conditions
exhibit some symptoms as children or young adults, but based
(08:46):
on guidance from the US National Institute of Mental Health,
depression can only be diagnosed once an individual exhibits the
five major symptoms of depression every day, all day for
a minimum of two weeks. Imagine how we could help
people earlier if we were able to identify depression with
the help of AI tools like the Happiness Guru model.
(09:10):
How does one actually create that model? What data is
needed to train that model so that it can get
that output.
Speaker 2 (09:20):
So basically, whenever we build any project, we were taught
this thing in the program itself that there's a whole
project cycle that needs to be taken into you know,
account while we're building any project. So the first step
that comes into the AA project cycle is problem scoping.
So we have problem statements, we have a stakeholders, and
we have our ideal solution as well. Now comes data
acquisition so basically to make this project work the way
(09:42):
it is working right now, data was collected you know,
anonymously through offline and online surveys and across five different
schools across India. So during these surveys, we briefed students
in the school what this survey is about and then
they were asked to fill out that form which contained
descriptive questions. Now, these descriptive questions that we selected, these
were validated by a team of psychiatrists and counselors and
(10:04):
then with the help of this survey process, we were
able to develop an authentic data set of seven hundred
plus centuries where the students basically wrote whatever they felt
during that time and you know, went out their thoughts
in that survey. The responses were labeled the on the
scale of A two D, with A being least sever
like perfectly healthy mentally and to D being needing immediate
(10:26):
support from professionals and family. And this was done with
the help of our school counselor, Ishitan Atara, So she
helped us in you know, laboring these responses and then
this data was used to train that SVM model that
I was talking about that is a part of step three,
So we need to convert this offline data into a
digitized format because that's how model gets trained. So we
(10:48):
did that, we started classifying, and then we trained the
SVM model. Apart from that, there's one more thing that
has went into this. The step one which I talked
about is about, you know, recognizing whatever current emotion the
user has, whatever their emotion is currently while they're using
So this is turned with the help of library basically
fast a dot Vision. So fast a dot Vision is
(11:10):
a library that is used for computer vision tasks. And
then we have trained this module using a data set.
So this data set consisted of two thousand rows i
would say, which consisted of facial expressions of different people,
like there were videos and images of people from different
genders and heritages of different backgrounds, and then they were
classified as happy saturn nedle to train our module, which
(11:33):
was fast a Dot Vision.
Speaker 1 (11:36):
What Tina is describing in her design philosophy is very
interesting because in a way it mirrors processes used by
psychiatrists and counselors to identify depression in young people at schools. However,
in her system the effectiveness is amplified. Oftentimes people experiencing
depression are not able to recognize the symptoms in themselves,
(11:59):
and for young people particular, having access to a professional
who could observe and identify the science is not guaranteed.
For cultural, social and economic reasons, mental health is largely ignored.
I can see the benefit of an automated system being
used to identify it and how that can help those
with reservations around mental healthcare take that crucial first step.
(12:22):
You're listening to technically Speaking an Intel podcast will be
right back. Welcome back to technically Speaking an Intel podcast.
I'm here now with Tina. So, So, in terms of
your research or next phase, do you think these sorts
(12:47):
of wearable devices or things that can detect people's emotions,
do you see a future where that could be a
possibility where we could get in early in terms of
detecting depression.
Speaker 2 (13:01):
Yes, there can be. In fact, there's been a rise
in it lately. Like I've been following up the news
around this, and I've got to know that there was
some institute in New York itself which conducted a study
which basically built an machine learning model that took the
data of thousands users, and then this model was able
to tell whether a person was mentally healthy or not.
(13:22):
So we need to understand how this works for us
to fall like, we are basically collecting data points in
terms of different variables, and these variables are like you know,
what at our pulse rate, what is our heart beat?
And I mean different things that can be measured by
these devices, by these variables to find the relation between
someone's mental health and whatever data points we are collecting.
(13:43):
So there's a possibility that in the coming year we
can lead mental health care services. Apart from this, a
similar thing that strucks to me right now is that
brain computer interface. I mean well, brain computer interface is
a machine that actually helps us to control a device
or machine using our brain. So if something of that
sort can be infused with machine learning, and then if
(14:05):
we can build some solution that is oriented towards solving
mental health problems that exist, that is oriented towards providing
more healthcare services, like those that accessible enough and affordable
as well, So I think majority of problems can be
solved in this area.
Speaker 1 (14:21):
Tina mentioning BCI or brain computer interface reminds me of
the conversation in episode three with Jaggedish and Lama. We
tend to think of BCI as human brains controlling the
function of a machine, like moving a mouse cursor or
controlling a robotic limb. However, Tina imagines a world where
(14:42):
our brains can simply inform machines on how to service us.
It is not so much that you would need to
even think about being helped, but the machine learning process
would allow a tool to remind you of a service
you need. It's almost like having a second brain. I
can't wait to see all of the medical applications this
open up to the world. Particularly through the pandemic and
(15:05):
post pandemic, there was a rise in mental health issues
which needed expert care. Now do you think that AI
can play a role in actually providing therapy for people
with mental health concerns. I recently read an article in
Time magazine about robot, which is a AI personal therapist.
(15:28):
I'd like to get your thoughts as to whether they
could actually provide useful advice for people to help manage
their depression and mental health issues.
Speaker 2 (15:39):
So, when we think of mental health care, you know,
the corner store of this is communication. It's not depending
on the procedures, but more on the communication. Like if
we know that therapist and the patient that there should
be a strong relationship between them. The relationship should be
good enough so that the patient can communicate with their
therapists and then the problem can whatever problem the patient
(16:01):
is going through. So like if you talk about therapists
in terms of air, So there are chadbots which are
coming up, like robot and new par So these chadbots
that are increasingly being used to offer advice and a
line of communication for mental health patients during their treatment.
So they can also help with coping up with symptoms
as well as they can look out for keyword that
(16:22):
could trigger a possible help that patient needs. So chargipity
can be used like a therapist. Like there have been
certain use cases, like I've been reading on Reddit and
there have been people who have been like sharing their
stories around how they use chargity as a therapist. So
when we see that chatbot can be used as a therapist,
it is like we are giving them some inputs and
they're basically doing sentiment analygies on the basis of textual
(16:45):
responses that we're giving to them, and then they are
basically modifying their answers to make it more human like
and that's how they can work as AI therapist. But
there are concerns around it as well. Like the first
thing that comes up with is reliability. How much accurate
of the solution that chatbot is providing us or any
tool that we have built as a form of therapist
(17:07):
is providing us. So first is reliability and then comes accountability.
What if you know, something wrong happens, Who's responsible for
all of this? But apart from this, the concern that
always struck me is that these are privately funded apps,
Like these are the apps that have been used at
commercial level. I mean, there are certain subscription charges that
need to be paid to use these apps. So I've
(17:30):
always had this view that once you start commercializing and
start making out profits from healthcare services, then things turn problematic,
you know, and when something as vulnerable and as volatile
as mental health is involved, I think we must be
very much cautious. We must be very much vigilant about
the kind of apps we are using and the kind
of tools that are coming in in terms of mental
(17:51):
health care services.
Speaker 1 (17:53):
And that leads me to if you are going to
be using these sorts of chat pots like chat GPT,
as you mentioned, to make sure that you're well aware
of who's got your data, what the privacy concerns may be,
and how you can make an informed decision. I like
to get your thoughts around that, particularly around privacy and
data security, and maybe you could start with how you
(18:14):
tackled it with your app.
Speaker 2 (18:16):
This is one of the main concerns that come up.
Like you also mentioned that whenever we are using such apps,
we need to be aware that what kind of data
we are feeding into it and what kind of formissions
we're giving to such a tool. But someone who is
going through a mental health problem mental illness, I mean
we cannot say that the person is healthy enough or
stable enough to be able to make a decision on this,
(18:39):
and therefore privacy concerns will come later in the stage.
But the first thing is that are we able to
make the patients familiarize with the kind of data they're
feeding into the apps and what are the consequences or
ramifications that this data can lead to.
Speaker 1 (18:54):
Yeah, because I actually heard some stories around people using
these chat says therapy and the concept of this transference,
so they're actually falling in love with the bots. There's
a similar experience with psychologists where patients fall in love
with the therapist. So that's just another potential challenge that
(19:17):
we all have to come to deal with if you're
going to start using these things.
Speaker 2 (19:21):
Yeah, they're a virtual entities that are coming into this
scenario and we are able to see them and they've
been living their own life. People are becoming so comfortable
with chatbirds now because definitely there's a lack of communication
that is happening, and ever since the pandemic gets strung,
this communication gap has increased, it has profoundly increased, so
people are finding way to escape this and then these
(19:41):
AI therapists come as a rescue and therefore people use
it blindly without being enough aware about what kind of
data they're feeling it and what kind of algorithms these
applications are using. Because we know that to these algorithms
may not be explainable, they're not transparent, so we have
to be aware about this as well. Literacy and education
is needed in these aspects as well.
Speaker 1 (20:04):
Yeah, just on that you talked about explainability and transparency,
do we just explain to the audience who may be
not so familiar with those terms when it comes to
AI models, what that actually means transparency.
Speaker 2 (20:18):
Okay, So there's a term that goes with algorithms, and
that's black box. So algorithms are like black blocks. We
know what is going out, but we do not know how
all of this is functioning, what is actually into the algorithm,
and what is the procedure and how on what basis
they're doing everything. Transparency is related to the kind of
data we're feeding it and the way we are using
(20:40):
it and how algorithm is working. To know this and
explainability means that any user, because there are two categories
of population who are associated with any AA system. The
first one are users and the second one are the
developers and stakeholders. So stakeholders must know that what kind
of algorithm it is and there should be transparency in it.
But when it comes to you user, AI systems and
(21:01):
those algorithms must be explainable enough. I mean users are
able to understand in a very human like language, that's
what this algorithm is doing.
Speaker 1 (21:10):
That's really good And as AI emerges as this tool
to help people struggling with their mental health, I'd like
a few more comments just around how you see it
working in tandem with the medical community to better serve
their patients and their communities. Do you have any thoughts
on how you know this tool can actually be used
(21:31):
together rather than a replacement.
Speaker 2 (21:34):
Yeah, Basically, we always think that AI is a disruptor.
We have always thought of this any technology that comes,
but I've always believed that they are over here to
augment our capabilities and to supplement whatever you know, roles
are there. So I'm from India and the very first
thing that I mean I have to cover up is
that we need to educate people around mental health because
(21:57):
in India, the most instrumental impedt in terms of mental
health is lack of awareness and education. People do not
know what exactly depression is, what exactly anxiety and stresses.
They use it in a very casual way. And to
be very honest, mental health is something which is stigmatized
in India. So you know, if someone is suffering from
mental health issue, they are often labeled as lunatics or
(22:17):
crazy or possessed. So we need to educate people around
this first of all. So I believe my project it's
still it's working. I'm looking forward to deploying it into
as many schools as I can. Because we know that annealgorithm,
the more data we feed into it, the more accurate
it becomes. Its current accuracy is seventy seven to eighty person.
So we need to increase that accuracy first of all,
(22:39):
and then we have to take care of the data.
We need to have some regulations, we have some norms
and rules. We have to inform our users also that
the data that we're taking from them is in safe hans. Secondly,
I believe I will be changing the working of this project.
Currently it works on you know, facial recognition on current mood,
and that can easily be fabricated. I mean something that
(23:01):
is not reliable. That is not a thing that should
be taken into account while you are assessing someone's mental health.
So I think I need to eliminate this step and
replace it with something better. It could possibly be like
I find a BCI like brain computer interface, this technology.
I find it very interesting, so I can possibly couple
it with this and then I can, you know, find
some solution.
Speaker 1 (23:23):
Tina's recognition of the unsustainability of facial recognition is very valuable.
My mother always said the eyes are the windows of
the soul. But Tina understands that who we are has
a lot more nuance to it. This is so important
to how machine learning develops to become more inclusive. One
of the biggest concerns with AI is a distrust of
(23:43):
the machine's ability to understand humanity. What is great about
hearing Tina speak is that her work is rooted in
finding multiple ways to understand humans. This gives me a
lot of hope for what AI can be, and we
have people like Tina behind its development. Just to circle
back round at the start, we talked about the start
(24:04):
of your story and getting inspired by the AI Youth
program run by Intel. I'd like to get a sense
of in terms of your peer group, how much interest
is there in AI development and STEM. I guess in
your coh of friends and peers, is it something they're
interested in and do you see a trend growing or
(24:27):
are there's still more challenges for people to take up
that sort of role in their career.
Speaker 2 (24:32):
Whatever peer groups I have, they all of them are
quite interested in data science and machine learning. We know
the data is the new oil, so like there are
a huge number of job rules that have been coming up.
And since many of my friends and acquaintances we are
like financially weak, so all of us look towards earning
some skill set and becoming job ready, increasing unemployability rather
(24:55):
than you know, we do not focus on taking this
up on a longer run. So, I mean there's a
lot up in this because we know that machine learning,
artificial intelligence, deep learning and whatever technologies that they are coming up,
they hold the potential to change, to transform the landscape
of every industry. So if we take it up as
a profession, then we need to stay in it for
(25:15):
a long run. But there are a multitude of impairments
to it. So the very first one is like I
being a girl. So in India, like especially from the
place I belong to, girls are usually not encouraged to
take up STEM fields. So we need to overcome that
first of all. And then once we become employable, once
we become like financially stable independent, I mean, then talking
(25:36):
on a personal level, I can then you know, work
in this field, and then I can possibly work in
somewhere around mental health and machine learning. And therefore, in
the coming future I plan to you know, launch a
program to say which is shakti in STEM. So Shakti
is a Hindi word and a literal meaning. It means
feminine energy. Apart from this, it also has a different meaning,
(25:57):
like in India, Shakti is used to represent strong and
resilient young girls and women. So I would want to
launch this program Shuck teen Stem, which aims at educating
youngers who are based in rural areas who heal from
financially weaker and economically weaker and socially backward start of
the society and to educate them and to fuel their
(26:18):
aspirations to enter into STEM careers.
Speaker 1 (26:22):
Yeah, that's awesome because I mean, I have two daughters
and I'm really encouraging them to get into the STEM
side of things. And you know, anything to help anyone
get into coding and developing and actually creating something from
new is quite a exciting feeling. So thanks Tina for
joining us today. I really enjoyed that and I learned
(26:43):
quite a lot from this.
Speaker 2 (26:44):
Thank you, Thank you.
Speaker 1 (26:50):
Thank you to my guest Tina Sahu for joining me
on this episode of Technically Speaking, an Intel podcast. This
episode brilliantly highlighted the potential of AI in supporting those
facing mental health challenges. I firmly believe that within the
next decade will witness a surge in AI powered therapeutic
tools designed especially for the younger generation navigating life hurdles.
(27:13):
One heartening development is society's evolving recognition of mental health
as a genuine concern. I remember the nineties as a
fresh faced teenager. It was a time when such discussions
were almost taboo and laden with stigma. Yet there's a
pressing issue the shortage of well trained mental health professionals
to cater to the increasing demand. AI and tech can
(27:35):
serve as invaluable aids for these professionals, ultimately benefiting our
community at large. Tina's transition from technology skeptic to its
ardent supporter was a highlight for me as a father
of three. I'm hopeful not just about the job prospects
AI will offer them, but also the tech savvy liars
they will lead, with AI becoming second nature to them.
(27:56):
Observing the innovative solutions emerging from young minds like Tina's,
I'm convinced we're on the cusps discovering awesome new technologies, apps,
and remedies for many of life's challenges. Please join us
on Tuesday, November twenty eighth for the next two episodes
of Technically Speaking, an Intel podcast we'll be sharing two
(28:17):
special episodes exploring the future of transportation and how technology
like AI has already created modern day and mobility marvels
like flying cars and autonomous shuttles. Technically Speaking, was produced
by Ruby Studios from iHeartRadio in partnership with Intel, and
(28:38):
hosted by me Graham Class. Our executive producer is Moley Sosha,
our ep of Post production is James Foster, and our
supervising producer is Nikias Swinton. This episode was edited by
Cira Spreen and written and produced by Tiree Rush.
I know I need to get ready for work, but
I'm just so so tired. Maybe I can skip work
to take a nap justice once. What's that? Oh, my
depression monitors detecting early signs of a depressive state. I
thought it was just tired, but this might be a
bigger issue. Let me set in an appointment with my therapist.
(00:40):
Hey there, I'm grain class and this is technically speaking
an Intel podcast. The show is dedicated to highlighting the
way technology is revolutionizing the way we live, work and move.
In every episode, we'll connect with innovators in areas like
artificial intelligence to better understand the human centered technology they've developed.
Mental health care solutions remain underinvested in many communities around
(01:04):
the world, yet so many suffer from issues that they
don't even know. They have a lot of these conversations
around health care, hinge on making therapy more accessible to
those in need. However, it can be difficult to determine
that one is experiencing depression or mental health crisis. Artificial
intelligence is at the forefront of many different advancements in healthcare,
(01:26):
but today we are going to dive into how it
is working to make mental health care more accessible. To everyone.
In order to do that, I have to introduce you
to a special guest joining me now is Tinasawho. Tina
Saho was a high school student when she started exploring
coding and engineering with AI. She never considered herself to
(01:47):
be overly interested in technology to start. However, she took
her natural curiosity and eventually was invited to be a
part of Intel's AI for Youth Pilot program, where she
developed a tool that uses AI to detect and predict
patterns of depression with around eighty percent accuracy. Since then,
she's been awarded the Calm Trust Fellowship for Women. She
(02:08):
currently attends to Ayabata College at the University of Delhi,
where she's completing a bachelor's in computer science and exploring
more opportunities to use AI as a tool for mental
health and more.
Speaker 2 (02:19):
Welcome to the show, Tina, Hi, thank you for inviting
me to this show, and I'm feeling more than privileged
to be a part of this podcast. So numbers stay
to everyone.
Speaker 1 (02:33):
I'm really interested in your story of how you got
into the STEM field, what prompted your interest in that
field and also in AI.
Speaker 2 (02:42):
To begin with, till my tenth class, I was a
student or a person who was always against technology, like
I always found in fact that the negative impacts that
technology brings and whatever ethical concerns are there, they cannot
be solved and they are simply too much to be
on the side of technology. But I remember in twenty nineteen,
the AAFO Youth program was launched and it was launched
(03:03):
in our school, which is Salvangal Senior Secondary School. So
when I participated in that program, I got to know
what artificial intelligence is, how many astonishing and stounding possibilities
AI can unlock and has already unlocked. And therefore, this
program pivoted my you know journey, my career, my professional journey,
and my cademic journey from being a non tech student
(03:24):
to becoming a tech student finally pursuing computer science as
my graduation.
Speaker 1 (03:29):
Okay, was there a particular topic or person that really
did spark that interest? What was the particular topic that
actually got you really fired up?
Speaker 2 (03:40):
So in my school there were two teachers who actually
you know, molded my thinking and they actually infused this
critical thinking aspect in me and they opened me to
the world of possibilities that science, that technology and engineering offers.
And I would like to take their name as well
as a you know, token of respect. So it's one
that Mam and Serbiam they empowered me. They encouraged me
(04:02):
to think beyond what I see.
Speaker 1 (04:04):
And that led you to the Intel's AI for Youth program.
What sort of projects have you worked on? Are you
working on right now?
Speaker 2 (04:14):
This isn't my participating in this program. I build Happiness Guru,
which is a model that predicts depression. Apart from that too,
I was also a part of Utul Tinkling Labs. Through
the Detail Lab of our school. I basically got to
know about Intel's AfOR Youth program only and in there itself,
I build a few projects and one of them was
Happiness Guru.
Speaker 1 (04:34):
And with the Happiness Crew, is that related to the
depression detection research you've been doing? What kind of prompted
you to look in that direction in the field of
depression and then using technology? Because generally, speaker, we don't
associate technology with treating depression or even detecting depression. So
(04:57):
what was the spark for you there?
Speaker 2 (05:00):
So while I was in this program, I was in
Class ten and my Plus ten results were out and
they were not as much as I expected, and I
went into a phase of depression because I associated myself
worth with the Marxi score, so my own personal experience
of dealing with depression. And then at that time, you
know that society rates were very alarming among the youth,
(05:22):
especially aged between fifteen to twenty nine. And we found
that the driving forces behind you know these when these
societies were you know, peer pressure, overburdening academics, financial stress,
and too much expectations that we have from youth, you know,
especially if you talk about teenage and someone who is
in between eighteen to twenty five. So on researching, we
(05:42):
found that these societ rates are very alarming, they're very distressing,
and these are the leading cause of the depression or stress.
And thereby we thought that we must come up with
some solution that can basically help us predict which person
is going through depression, and that to in a very
human friendly manner, not making someone uncomfortable with the kind
of procedures or with the kind of system we have.
(06:03):
So these were the I would say, the enablers that
led our team building this solution.
Speaker 1 (06:09):
And in terms of the happiness grew app can you
just explain how it actually works, you know, to try
and detect the early signs of depression.
Speaker 2 (06:19):
First of all, it's our web based application. While building
this project, the queue that we took, you know, to
build the entire model was that after our research, we
got to know that a person's vocabulary can be a
mirror into their mental state. And taking this as the queue,
we build this project which tries to analyze the emotional
(06:39):
quotient of a person of a user through their facial
expression and then their textual responses that the user is
going to provide to the AA machine. So the working
of the project is divided into three steps. The first
step is emotion detection stage, and in this stage you
basically need to stand in front of your laptop or
whatever device you are using this web application, and then
(07:00):
it detects your current mode, whether you're happy or sad,
you're neutral, angry. Then the next step is that user
is asked to answer nine questions and there's a scale
of relevance and then they need to select how much
relevant or how much they are able to relate this
to this situation. Then, after these two steps, a threshold
score is generated which gives the initial lead. If the
(07:21):
person is stressed or not, and if the score is
below the threshold that we have said, the person is
predicted as happy, while in the other case, user is
taken to the third step, which is the final step.
And this step consists of four descriptive questions which he
or she can use as a platform to went out
all his or her feelings and thoughts. So whatever answers
user will give to these four descriptive questions, these answers
(07:44):
will be used as a basis of classification. Then the
machine will predict whether the user is depressed or not.
So this AI machine, whatever you know input we are
giving in this step. There's a model namely SVM, which
is support vector machine. It's a non contextual classification model.
It is basically used to classify things. And then we
are using this model on the kind of you know,
(08:04):
language or keywords that are used in the answers. And
then accordingly the results are given out that whether the
person is stressed or not, and if the person is stressed,
automatically the person is consulted to the concerned authorities or
counselor otherwise the person is predicted as happy or not stressed.
Speaker 1 (08:22):
Detecting and treating mental health is something with which many
societies around the world struggle. According to the World Health Organization,
approximately two hundred and eighty million people in the world
suffer from depression and more than three hundred million are
living with anxiety. Many people with these mental health conditions
exhibit some symptoms as children or young adults, but based
(08:46):
on guidance from the US National Institute of Mental Health,
depression can only be diagnosed once an individual exhibits the
five major symptoms of depression every day, all day for
a minimum of two weeks. Imagine how we could help
people earlier if we were able to identify depression with
the help of AI tools like the Happiness Guru model.
(09:10):
How does one actually create that model? What data is
needed to train that model so that it can get
that output.
Speaker 2 (09:20):
So basically, whenever we build any project, we were taught
this thing in the program itself that there's a whole
project cycle that needs to be taken into you know,
account while we're building any project. So the first step
that comes into the AA project cycle is problem scoping.
So we have problem statements, we have a stakeholders, and
we have our ideal solution as well. Now comes data
acquisition so basically to make this project work the way
(09:42):
it is working right now, data was collected you know,
anonymously through offline and online surveys and across five different
schools across India. So during these surveys, we briefed students
in the school what this survey is about and then
they were asked to fill out that form which contained
descriptive questions. Now, these descriptive questions that we selected, these
were validated by a team of psychiatrists and counselors and
(10:04):
then with the help of this survey process, we were
able to develop an authentic data set of seven hundred
plus centuries where the students basically wrote whatever they felt
during that time and you know, went out their thoughts
in that survey. The responses were labeled the on the
scale of A two D, with A being least sever
like perfectly healthy mentally and to D being needing immediate
(10:26):
support from professionals and family. And this was done with
the help of our school counselor, Ishitan Atara, So she
helped us in you know, laboring these responses and then
this data was used to train that SVM model that
I was talking about that is a part of step three,
So we need to convert this offline data into a
digitized format because that's how model gets trained. So we
(10:48):
did that, we started classifying, and then we trained the
SVM model. Apart from that, there's one more thing that
has went into this. The step one which I talked
about is about, you know, recognizing whatever current emotion the
user has, whatever their emotion is currently while they're using
So this is turned with the help of library basically
fast a dot Vision. So fast a dot Vision is
(11:10):
a library that is used for computer vision tasks. And
then we have trained this module using a data set.
So this data set consisted of two thousand rows i
would say, which consisted of facial expressions of different people,
like there were videos and images of people from different
genders and heritages of different backgrounds, and then they were
classified as happy saturn nedle to train our module, which
(11:33):
was fast a Dot Vision.
Speaker 1 (11:36):
What Tina is describing in her design philosophy is very
interesting because in a way it mirrors processes used by
psychiatrists and counselors to identify depression in young people at schools. However,
in her system the effectiveness is amplified. Oftentimes people experiencing
depression are not able to recognize the symptoms in themselves,
(11:59):
and for young people particular, having access to a professional
who could observe and identify the science is not guaranteed.
For cultural, social and economic reasons, mental health is largely ignored.
I can see the benefit of an automated system being
used to identify it and how that can help those
with reservations around mental healthcare take that crucial first step.
(12:22):
You're listening to technically Speaking an Intel podcast will be
right back. Welcome back to technically Speaking an Intel podcast.
I'm here now with Tina. So, So, in terms of
your research or next phase, do you think these sorts
(12:47):
of wearable devices or things that can detect people's emotions,
do you see a future where that could be a
possibility where we could get in early in terms of
detecting depression.
Speaker 2 (13:01):
Yes, there can be. In fact, there's been a rise
in it lately. Like I've been following up the news
around this, and I've got to know that there was
some institute in New York itself which conducted a study
which basically built an machine learning model that took the
data of thousands users, and then this model was able
to tell whether a person was mentally healthy or not.
(13:22):
So we need to understand how this works for us
to fall like, we are basically collecting data points in
terms of different variables, and these variables are like you know,
what at our pulse rate, what is our heart beat?
And I mean different things that can be measured by
these devices, by these variables to find the relation between
someone's mental health and whatever data points we are collecting.
(13:43):
So there's a possibility that in the coming year we
can lead mental health care services. Apart from this, a
similar thing that strucks to me right now is that
brain computer interface. I mean well, brain computer interface is
a machine that actually helps us to control a device
or machine using our brain. So if something of that
sort can be infused with machine learning, and then if
(14:05):
we can build some solution that is oriented towards solving
mental health problems that exist, that is oriented towards providing
more healthcare services, like those that accessible enough and affordable
as well, So I think majority of problems can be
solved in this area.
Speaker 1 (14:21):
Tina mentioning BCI or brain computer interface reminds me of
the conversation in episode three with Jaggedish and Lama. We
tend to think of BCI as human brains controlling the
function of a machine, like moving a mouse cursor or
controlling a robotic limb. However, Tina imagines a world where
(14:42):
our brains can simply inform machines on how to service us.
It is not so much that you would need to
even think about being helped, but the machine learning process
would allow a tool to remind you of a service
you need. It's almost like having a second brain. I
can't wait to see all of the medical applications this
open up to the world. Particularly through the pandemic and
(15:05):
post pandemic, there was a rise in mental health issues
which needed expert care. Now do you think that AI
can play a role in actually providing therapy for people
with mental health concerns. I recently read an article in
Time magazine about robot, which is a AI personal therapist.
(15:28):
I'd like to get your thoughts as to whether they
could actually provide useful advice for people to help manage
their depression and mental health issues.
Speaker 2 (15:39):
So, when we think of mental health care, you know,
the corner store of this is communication. It's not depending
on the procedures, but more on the communication. Like if
we know that therapist and the patient that there should
be a strong relationship between them. The relationship should be
good enough so that the patient can communicate with their
therapists and then the problem can whatever problem the patient
(16:01):
is going through. So like if you talk about therapists
in terms of air, So there are chadbots which are
coming up, like robot and new par So these chadbots
that are increasingly being used to offer advice and a
line of communication for mental health patients during their treatment.
So they can also help with coping up with symptoms
as well as they can look out for keyword that
(16:22):
could trigger a possible help that patient needs. So chargipity
can be used like a therapist. Like there have been
certain use cases, like I've been reading on Reddit and
there have been people who have been like sharing their
stories around how they use chargity as a therapist. So
when we see that chatbot can be used as a therapist,
it is like we are giving them some inputs and
they're basically doing sentiment analygies on the basis of textual
(16:45):
responses that we're giving to them, and then they are
basically modifying their answers to make it more human like
and that's how they can work as AI therapist. But
there are concerns around it as well. Like the first
thing that comes up with is reliability. How much accurate
of the solution that chatbot is providing us or any
tool that we have built as a form of therapist
(17:07):
is providing us. So first is reliability and then comes accountability.
What if you know, something wrong happens, Who's responsible for
all of this? But apart from this, the concern that
always struck me is that these are privately funded apps,
Like these are the apps that have been used at
commercial level. I mean, there are certain subscription charges that
need to be paid to use these apps. So I've
(17:30):
always had this view that once you start commercializing and
start making out profits from healthcare services, then things turn problematic,
you know, and when something as vulnerable and as volatile
as mental health is involved, I think we must be
very much cautious. We must be very much vigilant about
the kind of apps we are using and the kind
of tools that are coming in in terms of mental
(17:51):
health care services.
Speaker 1 (17:53):
And that leads me to if you are going to
be using these sorts of chat pots like chat GPT,
as you mentioned, to make sure that you're well aware
of who's got your data, what the privacy concerns may be,
and how you can make an informed decision. I like
to get your thoughts around that, particularly around privacy and
data security, and maybe you could start with how you
(18:14):
tackled it with your app.
Speaker 2 (18:16):
This is one of the main concerns that come up.
Like you also mentioned that whenever we are using such apps,
we need to be aware that what kind of data
we are feeding into it and what kind of formissions
we're giving to such a tool. But someone who is
going through a mental health problem mental illness, I mean
we cannot say that the person is healthy enough or
stable enough to be able to make a decision on this,
(18:39):
and therefore privacy concerns will come later in the stage.
But the first thing is that are we able to
make the patients familiarize with the kind of data they're
feeding into the apps and what are the consequences or
ramifications that this data can lead to.
Speaker 1 (18:54):
Yeah, because I actually heard some stories around people using
these chat says therapy and the concept of this transference,
so they're actually falling in love with the bots. There's
a similar experience with psychologists where patients fall in love
with the therapist. So that's just another potential challenge that
(19:17):
we all have to come to deal with if you're
going to start using these things.
Speaker 2 (19:21):
Yeah, they're a virtual entities that are coming into this
scenario and we are able to see them and they've
been living their own life. People are becoming so comfortable
with chatbirds now because definitely there's a lack of communication
that is happening, and ever since the pandemic gets strung,
this communication gap has increased, it has profoundly increased, so
people are finding way to escape this and then these
(19:41):
AI therapists come as a rescue and therefore people use
it blindly without being enough aware about what kind of
data they're feeling it and what kind of algorithms these
applications are using. Because we know that to these algorithms
may not be explainable, they're not transparent, so we have
to be aware about this as well. Literacy and education
is needed in these aspects as well.
Speaker 1 (20:04):
Yeah, just on that you talked about explainability and transparency,
do we just explain to the audience who may be
not so familiar with those terms when it comes to
AI models, what that actually means transparency.
Speaker 2 (20:18):
Okay, So there's a term that goes with algorithms, and
that's black box. So algorithms are like black blocks. We
know what is going out, but we do not know how
all of this is functioning, what is actually into the algorithm,
and what is the procedure and how on what basis
they're doing everything. Transparency is related to the kind of
data we're feeding it and the way we are using
(20:40):
it and how algorithm is working. To know this and
explainability means that any user, because there are two categories
of population who are associated with any AA system. The
first one are users and the second one are the
developers and stakeholders. So stakeholders must know that what kind
of algorithm it is and there should be transparency in it.
But when it comes to you user, AI systems and
(21:01):
those algorithms must be explainable enough. I mean users are
able to understand in a very human like language, that's
what this algorithm is doing.
Speaker 1 (21:10):
That's really good And as AI emerges as this tool
to help people struggling with their mental health, I'd like
a few more comments just around how you see it
working in tandem with the medical community to better serve
their patients and their communities. Do you have any thoughts
on how you know this tool can actually be used
(21:31):
together rather than a replacement.
Speaker 2 (21:34):
Yeah, Basically, we always think that AI is a disruptor.
We have always thought of this any technology that comes,
but I've always believed that they are over here to
augment our capabilities and to supplement whatever you know, roles
are there. So I'm from India and the very first
thing that I mean I have to cover up is
that we need to educate people around mental health because
(21:57):
in India, the most instrumental impedt in terms of mental
health is lack of awareness and education. People do not
know what exactly depression is, what exactly anxiety and stresses.
They use it in a very casual way. And to
be very honest, mental health is something which is stigmatized
in India. So you know, if someone is suffering from
mental health issue, they are often labeled as lunatics or
(22:17):
crazy or possessed. So we need to educate people around
this first of all. So I believe my project it's
still it's working. I'm looking forward to deploying it into
as many schools as I can. Because we know that annealgorithm,
the more data we feed into it, the more accurate
it becomes. Its current accuracy is seventy seven to eighty person.
So we need to increase that accuracy first of all,
(22:39):
and then we have to take care of the data.
We need to have some regulations, we have some norms
and rules. We have to inform our users also that
the data that we're taking from them is in safe hans. Secondly,
I believe I will be changing the working of this project.
Currently it works on you know, facial recognition on current mood,
and that can easily be fabricated. I mean something that
(23:01):
is not reliable. That is not a thing that should
be taken into account while you are assessing someone's mental health.
So I think I need to eliminate this step and
replace it with something better. It could possibly be like
I find a BCI like brain computer interface, this technology.
I find it very interesting, so I can possibly couple
it with this and then I can, you know, find
some solution.
Speaker 1 (23:23):
Tina's recognition of the unsustainability of facial recognition is very valuable.
My mother always said the eyes are the windows of
the soul. But Tina understands that who we are has
a lot more nuance to it. This is so important
to how machine learning develops to become more inclusive. One
of the biggest concerns with AI is a distrust of
(23:43):
the machine's ability to understand humanity. What is great about
hearing Tina speak is that her work is rooted in
finding multiple ways to understand humans. This gives me a
lot of hope for what AI can be, and we
have people like Tina behind its development. Just to circle
back round at the start, we talked about the start
(24:04):
of your story and getting inspired by the AI Youth
program run by Intel. I'd like to get a sense
of in terms of your peer group, how much interest
is there in AI development and STEM. I guess in
your coh of friends and peers, is it something they're
interested in and do you see a trend growing or
(24:27):
are there's still more challenges for people to take up
that sort of role in their career.
Speaker 2 (24:32):
Whatever peer groups I have, they all of them are
quite interested in data science and machine learning. We know
the data is the new oil, so like there are
a huge number of job rules that have been coming up.
And since many of my friends and acquaintances we are
like financially weak, so all of us look towards earning
some skill set and becoming job ready, increasing unemployability rather
(24:55):
than you know, we do not focus on taking this
up on a longer run. So, I mean there's a
lot up in this because we know that machine learning,
artificial intelligence, deep learning and whatever technologies that they are coming up,
they hold the potential to change, to transform the landscape
of every industry. So if we take it up as
a profession, then we need to stay in it for
(25:15):
a long run. But there are a multitude of impairments
to it. So the very first one is like I
being a girl. So in India, like especially from the
place I belong to, girls are usually not encouraged to
take up STEM fields. So we need to overcome that
first of all. And then once we become employable, once
we become like financially stable independent, I mean, then talking
(25:36):
on a personal level, I can then you know, work
in this field, and then I can possibly work in
somewhere around mental health and machine learning. And therefore, in
the coming future I plan to you know, launch a
program to say which is shakti in STEM. So Shakti
is a Hindi word and a literal meaning. It means
feminine energy. Apart from this, it also has a different meaning,
(25:57):
like in India, Shakti is used to represent strong and
resilient young girls and women. So I would want to
launch this program Shuck teen Stem, which aims at educating
youngers who are based in rural areas who heal from
financially weaker and economically weaker and socially backward start of
the society and to educate them and to fuel their
(26:18):
aspirations to enter into STEM careers.
Speaker 1 (26:22):
Yeah, that's awesome because I mean, I have two daughters
and I'm really encouraging them to get into the STEM
side of things. And you know, anything to help anyone
get into coding and developing and actually creating something from
new is quite a exciting feeling. So thanks Tina for
joining us today. I really enjoyed that and I learned
(26:43):
quite a lot from this.
Speaker 2 (26:44):
Thank you, Thank you.
Speaker 1 (26:50):
Thank you to my guest Tina Sahu for joining me
on this episode of Technically Speaking, an Intel podcast. This
episode brilliantly highlighted the potential of AI in supporting those
facing mental health challenges. I firmly believe that within the
next decade will witness a surge in AI powered therapeutic
tools designed especially for the younger generation navigating life hurdles.
(27:13):
One heartening development is society's evolving recognition of mental health
as a genuine concern. I remember the nineties as a
fresh faced teenager. It was a time when such discussions
were almost taboo and laden with stigma. Yet there's a
pressing issue the shortage of well trained mental health professionals
to cater to the increasing demand. AI and tech can
(27:35):
serve as invaluable aids for these professionals, ultimately benefiting our
community at large. Tina's transition from technology skeptic to its
ardent supporter was a highlight for me as a father
of three. I'm hopeful not just about the job prospects
AI will offer them, but also the tech savvy liars
they will lead, with AI becoming second nature to them.
(27:56):
Observing the innovative solutions emerging from young minds like Tina's,
I'm convinced we're on the cusps discovering awesome new technologies, apps,
and remedies for many of life's challenges. Please join us
on Tuesday, November twenty eighth for the next two episodes
of Technically Speaking, an Intel podcast we'll be sharing two
(28:17):
special episodes exploring the future of transportation and how technology
like AI has already created modern day and mobility marvels
like flying cars and autonomous shuttles. Technically Speaking, was produced
by Ruby Studios from iHeartRadio in partnership with Intel, and
(28:38):
hosted by me Graham Class. Our executive producer is Moley Sosha,
our ep of Post production is James Foster, and our
supervising producer is Nikias Swinton. This episode was edited by
Cira Spreen and written and produced by Tiree Rush.
Host
Graeme Klass
Popular Podcasts
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com