How HACKERS Are Using AI & Facial Recognition on Tinder | Mashable Originals

How HACKERS Are Using AI & Facial Recognition on Tinder | Mashable Originals


– If I asked you to have a guacamole party with me, would you do it? If she would say yes,
then the bot would smile and playfully say okay do women love anything more than avocados? (suspenseful music) Oftentimes when there were dull evenings, I would try to find dates. Whenever I had free time I was on the toilet or something
and start swiping. And it’s a massive time sink. I thought there had to be some kind of way to automate this process. That’s the reason why I started working with this Tinder bot. It was on a github,
and once you open it up it looks like Facebook Messenger almost. It’s a web interface for your Tinder. You can chat with the girls. You can see their user icon on the left. You have the conversation on the right. And you also have the
terminal window running where you can see what the bot is doing. And with some customizations
on the script, you could also start a
conversation automatically. And take positive and negative sentiments. And from thereon, the bot would
start a mini conversation. The second step is where you actually train the bot to know your preferences. In the background, there is an elementary machine learning method running. (suspenseful music) It detects things in
the picture like color. Hair color, eye color, facial structure for the cheek bones, and so forth. With all these machine learning images that are being input into the system, the system actually gets to
learn what you like or dislike. And can actually start
making this decision for you without you having to interact with it. So yeah, the more pictures
you swipe yes or no to, the more accurate the AI or
the bot can decide for you. If it needs to move forward
with this match or not. – The overall challenge
with a lot of males on Tinder or any other dating app is that, okay they won’t get responses because their profiles or pictures
are not good enough. I was in that camp where
hey, I wasn’t really getting that many dates through dating apps. I felt like I was putting
a lot of time into it. So, I was like okay, why
don’t I try to develop an AI that would automate the process for me. So, on a simple level, you
can use machine learning to make predictions about certain things. The idea behind it is to kind of replicate the neurons in your brain to make the most intelligent predictions based on as much data as possible. (upbeat music) What deep learning can do is it can actually learn the edges of the faces or the different shapes inside an image. So, I was able to extract a lot of data that I had already liked myself. And use that to train the model. However, one of the problems was that I actually probably dislike most of the profiles that I see. So, then the algorithm only learns the profiles that I dislike. I had to get more data
of people that I liked. I ended up just using a scraper and scraped a bunch of Google data. I trained it with images of
females that I found attractive. For the bot I had a automated message that I can change every
time I got a match. And then once I got a match then I can start a conversation there. – If I want to build an
artificial intelligence that learns my interests in women, I have to first collect a
lot of data about women. I developed a method
that teleports my profile to different locations in Switzerland. Just fetched all data for
all women in that area. And then categorize the women systematically into categories. I like the woman or I dislike the woman. In like 48 hours I fetched
40,000 images of women in my area so that it can figure out which women I like in my area and which I don’t. I then used a Tinder API to
get people in my actual area. And used my pre-trained
artificial intelligence to swipe on Tinder in an intelligent way. Also taking into account
some of the meta data. Like the university of the person. The location of the person. Or the land from the profile
description it provided. – Hinge came along and
they make you actually send a message based
on a piece of content. It does end up that there’s
five photos that everyone has. There’s a dog photo, there’s
a holding my nephew photo. There’s a hiking photo. It’s literally like you can categorize it. How do you optimize that interaction? And the idea I had in my head
was to call it a Witty Bot. ‘Cause I was watching
the banter in the box. To imagine in a world where you don’t target letters and words to make meaning. You start with your intent
and your like (hums)? I like her. Based on whatever the photos are that can auto-write that entire box. I mean, don’t even think of it as a bot. Think of it as your friend
who’s your concierge, who’s gonna find you a date. – It’s a funny story, my
girlfriend she was one of the matches that my bot created on Tinder. And my bot actually started
a conversation with her. Very random at a party a few
months after I met my current girlfriend and I was
immediately attracted to her. But she didn’t know that
I was the guy from Tinder. And I didn’t know that my bot had a conversation with her in the past. And I told her the story
that I’m now telling you, and she’s like, oh yeah, I remember! And you said that, that, and that. That was exactly the conversation
from the bot in the tree. – The Tinder bot I built
has a 90 to 92% ratio in which it swipes correctly for me. This is not a very good
performance for my personal usage. I could easily have message
women as well using my bot. But I didn’t implement that feature. – I was thinking, okay I’ll see how many matches I can collect. The last time I checked the
matches were around 1,200. I think in swipes maybe 10,000,
20,000 times at one point. The bot was having maybe 200
conversation at the same time. And I think (chuckles)
at Tinder they noticed. And then they banned me of
course from the platform. – The response when I told
women that I built an AI that automated Tinder was that they were actually very impressed by it. I think most of the women
I’ve gone on dates with were not in say an
engineering type of field. So, to them they thought it was a very magical thing that I had built. – I went on a date with
someone who was like a chief marketing officer
at one of the dating apps. And I just was like, what do you think about this kind of idea
that I’m working on? And (chuckles) I showed her. She’s like, if you give this to the world you’re gonna manipulate everyone! And [bleep] guys are gonna
trick women, and blah, blah. I was like, whoa, I
didn’t give it to anybody. I’m just working on it. (suspenseful music) – I, as someone on
Tinder, did I consent to having my data taken off Tinder? And used in this way? I love what could possibly go wrong? That’s my favorite design question as a design researcher
and design educator. So, this isn’t just an
issue of dudes being weird and trying to date as
many people as possible. Which is creepy and Tinder
should not allow creepiness! It’s also a heightened level of inappropriateness, if you will. Because someone is taking your data, creating a machine learning
model around data in aggregate. None of the women or
people consented to that. How will it feel as a woman who’s never interacted with that person? Or a person who’s on the site who ends up just in their data scraping? How will that person feel
being in their data set? If any of this data could be accessed, think of all different kinds of domestic violence issues
that could arise out of this. Maybe no one knew you were on Tinder. And now this big data set’s out there. And people can access it. Maybe you get a stalker. What if people are scraping
men seeking men or women seeking women in countries
where homosexuality is illegal? What do you do? What do you do if you’re
a marginalized person? You’re part of the LGBTQI community. And you’re in a country
where this looked down upon. Well, then this poses an
extreme security threat. And that shouldn’t be lessened
in this conversation, right? That is a problem. – Tinder just doesn’t try at all. Their API is so easily accessible. I can just teleport myself
all over Switzerland and fetch data and pictures of everybody
on whole Switzerland. That don’t have to try. They just offered the
data for free basically. So, it’s very easy to hack it and it’s very easy to data mine from it. – These are real security vulnerabilities that platforms need to take seriously. And Tinder, as a dating app, needs to view itself as a platform. – A creep tries to abuse
the scripts we write to get himself dates, then
yeah, of course he can do that. And that’s true, but the other side of open sourcing our script is that Tinder has the possibility to see how we spooked their API and how we hacked the Tinder. The real problem is that Tinder doesn’t do anything to
prevent us from hacking it. – I’m of the mindset that bots
should always be disclosed. In a setting like Bumble
or Tinder or Hinge where you aren’t expecting
to engage with a bot. You’re expecting to
engage with a real person. (suspenseful music) It’s like, that’s a great way
to push users off the site. Especially if they’re
getting a high frequency of interactions that they’re not used to. Which looks and feels like spam. – Yeah, I’ve seen that
men or women are using artificial intelligence to
just really run numbers. And even bots to communicate for them. I think that it’s only one more reason that I’m not currently on any dating apps. – Because I’ve got concerns that any time that you’re not the one
who’s personally writing your responses or evaluating
a potential match. – I found that a little disheartening because I feel like I have done the work. – If your intention is
to find a match and date, there’s a better way to do it. If your intention is to use software to feed my ego, to feel
better about myself, there’s a [bleep] of software,
ways that you can do that. Match with everyone, swipe,
swipe, swipe, swipe, swipe. Ding, ding, ding, ding, ding! That’s a very shallow,
hollow kind of world that I don’t know I would wanna do. – Yeah, partially why I
stopped using it for dating was because I had a really met, I didn’t get somebody that I really clicked with through an AI. (suspenseful music) – As a woman, I would not be interested in an automated dating app that
did the algorithms for me. Or used AI. Because to me that takes away
the personal touch of it. I’d rather have one match in a month. – Hardly on the edge of morality to even swipe on women using a bot. And to actually write
them automated messages. To me it was just a bit too much. – I see the issues, and I
totally agree with them. There’s no laws yet, so
it’s kind of like Wild West. I believe there’s gonna be regulations and things that have to be traded around the space in the
next two to three years. That will dictate how everyone
has to play in the space. Realizing the bot space is knowing if it’s a machine or a human
I think is very important. (suspenseful music) The reason I didn’t put it
out there is not as much about the assistant writing and
a bot that can help you, it’s more about disclosure
and the efforts around that. It’s giving the power
of better words to men who otherwise maybe wouldn’t say that. Or aren’t that good of a person. For the good that I think it can do, if not disclosed or packaged correctly or used with the wrong intention. Words have power and I think
it can be a dangerous thing. – And this is kind of the problem with Silicon Valley’s
hacker or “fix it” mindset is they get hyper-focused on one problem. And the problem this
was optimizing dating. But what they didn’t take
a step back and look at was what could possibly go
wrong if other people with different agendas can use this?

12 Comments

Leave a Reply

Your email address will not be published. Required fields are marked *