Kind: captions
Language: en
Oh, hi ma'am. Wait, are you a Yes. Oh,
okay. I think I uh I think I met you
through like the Cambridge Community
Foundation, right? Yeah. Yeah. Yeah.
It's great. What if you could
immediately know the name and personal
details of any stranger you passed on
the street? Two undergraduate college
students developed a tool that can pull
sensitive personal information like home
addresses and phone numbers from
existing databases just by using a
person's face and without that person
even knowing it's happening. Anfuin and
Kane Ario are students at Harvard
University. They set out to build facial
recognition glasses using these smart
glasses built by Meta, the parent
company of social media sites like
Facebook and Instagram. Basically, they
wanted to be able to put on the glasses,
look at a random person, and identify
who they were. After just 4 days of
coding, they had a working prototype and
were trying it on students around
campus, even on strangers at a train
station. like minority stuff for like
Muslims in India at all or something?
Really?
Yes. Oh, I've read your work before.
Here's how it works. Once Anfo Recane
puts on these glasses, an LED light
turns on to indicate they're recording
and they start to live stream the video
feed to Instagram. Then a bot they built
takes screenshots of the video and scans
a facial recognition library looking for
a face. Once you look at someone, it
puts the face into this database that
returns links to where that person
appears online. That database is called
PIMIS, a facial recognition search
engine that scour the internet to find
websites where that face shows up. We
use some large language models to scrape
those websites, pull the information,
and then guess who it might be on those
websites. A large language model or LLM
is a type of artificial intelligence
like chat GBT that can understand and
summarize data into text that's similar
to how humans speak. The LLM suggests a
name to match the face and then that
name is used to find more information.
And then it would put it through a bunch
of databases to try to figure out like
home address, job, what they enjoy
doing, that type of thing. And then all
of that information, the photos, names,
addresses, and interests get sent to
their phone. From beginning to end, this
takes about 90 seconds. Anfu and Kane
use the free versions of these
databases, but they say if they had
unlimited access, they could have gotten
results in just a few seconds. The
project raises a lot of ethical
questions, but it's not illegal. Some
states like Massachusetts do have
stringent rules that restrict how the
government can use facial recognition
technology, but that is not the norm
across the country. And the databases
themselves can also impose limits. Once
the project went public, Pimise removed
Anfu and Kane's access because they say
the students violated the site's terms
of service by uploading photos of people
without their permission. But Anfu and
Kane's goal isn't for this to be used.
It's to show that we're already
vulnerable to this kind of technology.
We basically expose the fact that you
can go on the street and pull all of
this data. It was simply like a
technical demonstration. In a Google doc
accompanying their video, they encourage
and instruct people to remove personal
information from databases that contain
details like addresses and even the
names of relatives. All of these
databases that we use, they have opt out
systems where you can just go onto their
website, verify your identity, and then
you're completely scrubbed from those
databases. Facial recognition software
already exists, like the kind used to
verify your identity when you're going
through airport security or to unlock
your iPhone. And similar facial
recognition technology is being used by
some law enforcement agencies both at
the federal, state, and local levels to
identify individuals and support
criminal investigations. Though a few
cities like San Francisco, Boston, and
Portland, Oregon have laws specifically
prohibiting city officials like police
departments from using facial
recognition systems. But Anfu and Ka
showed private citizens can also use
facial recognition software to identify
strangers. They don't plan to take this
project any further. Instead, they
started working with data privacy
advocacy groups. As engineers, they say
they've done their job to bring
awareness and show that this is
possible. As facial recognition
technology continues to advance,
questions about data privacy and
security will continue to arise, and
many may start to use this technology
for more than just a demonstration.
[Music]