Behind the scenes of a contact tracing study
Transcriber: Amanda Chu
Reviewer: Peter van de Ven
Sydney-Pacific,
it’s a very active community,
and the vast majority of events
center around congregating in groups
and eating shared food.
In March,
the undergraduate dorms de-densified
and then the graduate dorms
relatively soon after.
In Sydney-Pacific,
our new mode of living
required that we not allowed to gather,
and our student life sort of halted.
The spark that became this study
actually originated
from the grad students in the dorm.
They said, “Isn’t there
some type of technology or means
to understand usage of our shared spaces
and risk associated with it?”
I thought it’s a good question,
so I sent a note to two different groups,
one at the Computer Science
and Artificial Intelligence Laboratory
and another in the Media Lab.
I just asked, “Is there something
that we can use that is useful,
actually, like in a dorm?”
And both responded back
immediately and said,
“Not with what we have right now,
but here’s an idea.
Take Bluetooth beacon sources
and install them
in each of the common spaces,
and then we could develop an app
that would receive
that signal on the phone.
From that, you have the potential
to back out distance and proximity
and therefore use of those common spaces.
I just asked a few questions
and had a lot of volunteers overnight.
We had this amazing team
of faculty, grad students,
and professionals at Lincoln Lab.
[Anthony Lapadula -
Researcher, MIT Lincoln Laboratory]
Anthony Lapadula: So I think everybody
had the same idea at the same time.
Everybody carries a smartphone -
“Oh, what can we use from the smartphone
to measure a distance?”
Bluetooth was kind of
an obvious first guess.
And then the question is,
Can you measure how long
you’ve been close to someone,
and how close have you been
to that person?
I don’t think anyone knew at the beginning
how hard it was going to be.
Bluetooth was not designed for this.
If you just stand close
to a Bluetooth beacon with your phone
and watch the signal strength,
it tends to bounce around quite a bit.
Their reflections are across the wall;
there’s multi-source;
there’s all kinds of problems.
One of the biggest things
that degrades signal strength
is the human body.
What really excites me about this project
is the possibility of collecting
lots of Bluetooth data
to help us build a good mapping
between signal strength and proximity.
But we already had a calibration protocol
for people to collect data
in a structured way;
however, it was probably a couple hours
to actually run the entire protocol.
There was no way that
we would get people to volunteer,
so we came up with a very streamlined
calibration protocol.
You should be able to run
through one of these in about 60 seconds.
The joke I always make is,
“You’re in the kitchen.
You just put your burrito
in the microwave.
You have a couple of minutes.
Go ahead and do a couple of calibrations.”
Abhishek Singh: So I’ve been working
mostly around the app development.
It really gets tricky
running these Bluetooth apps
because the Google
and iOS operating systems
are really stringent about Bluetooth code
running in the background,
so it requires some fiddling around
and some rigorous testing.
So with iOS,
whenever the app
goes to a background state,
the iOS operating system
makes it in a sleep mode.
The app is - although running
in background -
it’s not really doing anything,
and it’s not really trivial
to wake up this app
and make sure that it keeps recording
the Bluetooth data and keep sending it.
In Android, we can do this
for every minute,
but there is some trade-off
with battery usage.
It looks like 15 minute
is like sort of a sweet spot,
where the app can wake up
and listen to the chirps
and then again go to sleep.
Daniel Ribeirinha-Braga:
The tricky part is those sensors.
You could have an Android phone
that is five or six years old;
however, your phone does not have
a gyroscope or a barometer,
so we have to be able
to create a data model
so that even if you don’t have
that information,
our backend server can still ingest that
and do something
with the remaining information.
So being able to test
those cases is critical.
We’d just run a bunch
of emulators on our computers,
but that’s completely different
because a barometer and a gyroscope
don’t exist in a virtual machine.
In a computer, you need a physical device.
We all did a bunch of testing on the app.
Some people have old iOS phones,
some people have
different Android phones.
That’s how we go through it.
Everyone tested on their phone.
We reported a bunch of bugs.
But there was a solid
two- or three-week period
where we were constantly
getting a list of tasks
and it’d be like,
“I saw this issue and that issue,”
but little by little,
we were able to really make some progress
to solidify the application.
Christopher Fourie:
So phones have complications.
Our tile-based solution is far simpler.
You have constant transmission power
because you no longer have to worry
about different device platforms.
It doesn’t require user inputs,
because you can just attach it
to your keys and forget about it.
So the entire solution is a lot simpler
and as a result, a lot more robust.
The device will constantly emit chirps.
We’ll record those chirps
using an external architecture,
and you can then figure out
which people were
in close proximity to each other.
So the tile solution allows us
to directly compare
the additional consequences
of using a phone-based environment
as opposed to something
that’s a little bit different.
Ilaria Liccardi: As a privacy researcher,
sometimes I try to think
of the worst case scenarios.
Sometimes I cannot even predict
what that will be,
because there are so many things
that could go wrong.
Especially some of the contact traces app
have leaked their entire email addresses.
So the fact that we’re using an ID
that only a few people have access to
and the fact that we are not
using location,
that was something
that we had discussed at length
because removing this kind of information
makes it more difficult,
but it does make
people’s identity more secure,
and we actually found
a good balance as a result.
We could have gone even further.
But with the balance that we strike,
I think it was good enough.
Sarah Chung: This whole process
is for the students, right?
I mean, there are research goals,
but part of it is
to benefit MIT campus in general,
and then much of it is by students
because they’re the boots on the ground.
They’re not only the boots on the ground
to set up all the infrastructure,
but they’re also the ones
who have to volunteer their own data
and download this app.
Without these student volunteers
advocating in the dorm,
this wouldn’t happen.
Without installers, this wouldn’t happen.
Without students registering,
this wouldn’t happen.
And then even after they register,
without them calibrating like,
also, it will not happen.
However, volunteers really
need to be invested in too.
They have to feel like
they’re getting something out of it.
And a lot of my thinking
has gone into how do we make this
more efficient and less work for them
so that they can enjoy
benefits with less cost
and they do what we hope
that they will do.
We shall see, right?
Because we’re just at the cusp
of rolling out Eastgate and Ashdown,
and we shall see.
Geeticka Chauhan:
As I talk with my officers
about different policies
we want to implement
or different events we want to run,
having the knowledge
that the contact tracing study
is going on in parallel
is very helpful,
and the officers that I’m involved with
are quite excited.
A lot of the questions
that other dorm presidents had
were primarily privacy concerns
and how the information
of the student would be kept
and how this information
would be used by MIT Medical.
Another thing that the dorm presidents
were really excited about -
they were thinking,
“Yeah, this is a really good thing
to start in the dorms,
but they were also asking Julie
if she’d be willing
to deploy this in the labs.
Przemyslaw Lasota:
The main objective
of the scientific protocol
was to evaluate
how digital contact tracing
would compare to manual contact tracing,
and we had three
secondary objectives in this study.
The first one was
understanding the trade-offs
between more privacy preserving
and less privacy preserving
digital contract tracing methods;
also, contributing calibration data
to help other researchers
understand the relationship
between Bluetooth
signal strength and distance;
and finally, trying to understand
user acceptance of different
digital contact tracing methods.
JS: I still am amazed where at the point,
we are just about ready
to launch the study at full scale
because the team that came together
was a set of people who had
literally never worked together,
and now we’re co-developing
software, hardware virtually.
And it’s been a dream team.
The complexity and scale
was surprising to me,
but you wouldn’t know it
based on how this has come together
in the course of six weeks
to two months.
We’re currently rolling out
to three of the graduate dorms,
but then three of the undergrad dorms,
as sort of a pilot or a test
of what we might be able
to do more broadly
when undergrads return
at a higher level in September.
The hope is that we can help
keep our community safer
through the fall reopening
and also provide valuable knowledge
on how digital
contact tracing technologies
can be used more broadly beyond MIT.