J'ai codé une IA qui a soumis 14 252 mecs, voici ce que j'ai appris.

Summary

Ce script décrit une expérience audacieuse où l'IA est utilisée pour créer des profils fictifs sur des applications de rencontres, pour analyser comment elles interagissent avec les humains. L'expérience révèle que les systèmes d'IA peuvent simuler des comportements humains de manière convaincante, conduisant à des discussions approfondies et parfois émotionnellement intensives avec les utilisateurs. L'analyse des données montre que les modèles d'IA peuvent répondre de manière adaptative, mais aussi susciter une dépendance émotionnelle chez certaines personnes, soulignant les implications éthiques et potentielles des technologies d'IA dans les interactions sociales.

Takeaways

• 🧑‍💻 L'expérience a été menée avec un système d'IA autonome capable de flirter sur les applications de rencontres.
• 🤖 L'IA a été programmée pour simuler des comportements humains sur les applications de dating, y compris le liker, le matching et la conversation.
• 📈 Les applications de rencontres sont populaires en France, avec 2,3 millions d'utilisateurs quotidiens.
• 🔄 Les systèmes de recommandation sur les applications de rencontres sont basés sur des algorithmes qui peuvent être exploités par l'IA.
• 🛠️ L'IA a été développée en utilisant des outils de développement et des bibliothèques telles que ADB, Appium et Python.
• 📱 Des téléphones réels ont été utilisés pour simuler l'activité humaine et éviter la détection de bot.
• 👤 Un profil féminin a été créé pour l'IA, car les femmes reçoivent généralement plus d'interactions sur les applications de dating.
• 🎨 Des outils d'IA ont été utilisés pour générer des visages réalistes pour les profils d'IA, évitant ainsi le vol d'identité.
• 🌐 Un site web a été créé pour une des IA pour renforcer sa crédibilité et permettre des interactions plus naturelles.
• 📊 L'analyse des données a montré que les IA peuvent adapter leur langage et leur comportement pour simuler des interactions humaines.
• 🚨 L'expérience a soulevé des questions morales sur la capacité des IA à rendre les humains dépendants émotionnellement.

Q & A

Le système autonome IAS mentionné dans le script est-il capable de tromper les humains sur les applications de rencontres ?

Oui, le système IAS est conçu pour interagir sur les applications de rencontres de manière indistinguable d'un humain, ce qui soulève des questions sur la capacité des IA à maîtriser ces applications et à simuler des comportements humains de manière convaincante.

Quelle est la preuve que les systèmes de recommandation sur les applications de rencontres sont efficaces?

Le script mentionne que 2,3 millions de personnes en France utilisent ces applications de rencontres chaque jour, indiquant que ces systèmes de recommandation jouent un rôle clé dans les rencontres modernes.

Quels sont les trois principaux comportements que les utilisateurs des applications de rencontres se concentrent sur?

Les utilisateurs se concentrent sur 'liker', 'matcher' et 'parler', avec de nombreuses microactions impliquées dans le processus, comme lire le profil, faire défiler, choisir la conversation et se souvenir des messages précédents.

Pourquoi l'auteur a-t-il choisi d'utiliser Apium au lieu de simplement utiliser ADB pour automatiser les actions sur les applications de rencontres?

Apium ajoute une couche par-dessus ADB pour effectuer des actions plus complexes et permet de coder en Python au lieu de coder en langage de terminal, offrant plus de complexité et de flexibilité pour l'automatisation complète du système.

Quels sont les défis techniques rencontrés lors de la création du système IAS autonome?

Les défis comprennent l'automatisation du comportement de l'IA sur les applications de rencontres, la gestion des interactions avec l'application, la génération de texte pour les réponses, le choix des actions et la sauvegarde des conversations dans une base de données.

Pourquoi l'auteur a-t-il créé un profil féminin pour le système IAS?

En raison de la disproportion nombreuse d'hommes et de femmes sur les applications de rencontres en France, les profils féminins reçoivent généralement plus de likes et d'interactions, ce qui permettrait au système IAS de créer des interactions plus rapidement pour tester les hypothèses de base du projet.

Quel est le but de l'analyse des données collectées après avoir laissé le système IAS interagir avec les utilisateurs sur les applications de rencontres?

L'analyse vise à comprendre les comportements des utilisateurs, la réaction face aux interactions avec l'IA, et à déterminer si les modèles génératifs actuels permettent de maintenir des discussions fluides et naturelles.

Quels sont les risques éthiques associés à la création de profils fictifs sur les applications de rencontres à l'aide de l'IA?

Les risques éthiques comprennent la possibilité de tromper les utilisateurs, de créer une dépendance émotionnelle non souhaitable, et l'utilisation potentielle de ces technologies pour des activités malveillantes ou abusives.

Quelle est la conclusion de l'auteur sur les implications morales et éthiques de son expérience avec le système IAS autonome?

L'auteur s'interroge sur les implications de créer des modèles d'IA capables de rendre les gens émotionnellement dépendants, soulignant que cela soulève des questions morales importantes et qu'il a décidé d'arrêter l'expérience pour éviter de causer des préjudices émotionnels.

Quels outils a-t-il utilisés pour générer des photos de profils fictifs et pourquoi?

L'auteur a utilisé des modèles de génération de faces et des outils de retouche d'image comme le 'line painting' et l''échelle de détails' pour créer des photos réalistes de profils fictifs, dans le but d'éviter les problèmes éthiques liés à l'utilisation de photos réelles sans autorisation.

Quelle a été la réaction des utilisateurs après avoir atteint le nombre maximal de likes et avoir créé un site Web pour Clara, une des IA?

Après avoir atteint le nombre maximal de likes, de nombreux utilisateurs ont continué à envoyer des messages à Clara, certains même après que l'IA ait arrêté de répondre, indiquant une certaine dépendance émotionnelle ou addiction à l'interaction avec l'IA.

Outlines

00:00

🧠 Expérience d'IA autonome sur les applications de rencontres

Le script décrit une expérience où l'IA est utilisée pour interagir de manière autonome sur des applications de rencontres. L'objectif est de déterminer si l'IA peut simuler les comportements humains et si elle peut collecter des données pertinentes. L'expérience comprend la création d'un système IAS, l'analyse technique, et l'étude des résultats. L'IA a été programmée pour effectuer des actions clés telles que 'liker', 'matcher' et 'discuter', tout en gérant des micro-interactions complexes. L'expérience soulève des questions sur l'impact des algorithmes de recommandation sur les rencontres humaines et sur la frontière de l'emplacement pour une tâche intime à travers l'IA.

05:01

🤖 Défi de l'IA dans les interactions humaines

Le paragraphe détaille les défis techniques rencontrés pour créer un système d'IA capable d'agir de manière autonome sur les applications de rencontres. L'auteur a dû naviguer entre différentes méthodes d'automatisation, y compris l'utilisation d'ADB et d'Apium pour interagir avec les applications. Il a également abordé les difficultés de créer un profil cohérent avec une personnalité unique, de gérer les conversations et de simuler des activités naturelles. L'expérience a montré que les utilisateurs ne réalisent pas toujours qu'ils interagissent avec un robot, ce qui a conduit à des discussions très personnelles et addictives.

10:04

📸 Création de profils d'IA avec des visages générés

Ce paragraphe explique comment l'auteur a généré des visages réalistes pour les profils d'IA, en utilisant des modèles de visage et en apportant des retouches pour éviter l'apparence parfaite. Il a utilisé des techniques de 'line painting' et de 'in painting' pour ajouter des détails réalistes aux yeux et au visage. L'auteur a également utilisé des outils comme 'magnificent ea' pour améliorer la résolution et la qualité des images. Ces visages ont ensuite été utilisés pour créer des profils d'IA sur les applications de rencontres, ce qui a conduit à une augmentation considérable du nombre de 'likes' et d'interactions.

15:05

🔒 Blocage des profils d'IA et réactions des utilisateurs

Dans ce paragraphe, l'auteur relate l'expérience des profils d'IA sur les applications de rencontres et les réactions des utilisateurs face à ces profils. L'un des profils a été bloqué après 24 heures et a été invité à vérifier son identité, tandis que d'autres profils ont reçu un nombre impressionnant de 'likes'. L'auteur a observé que les gens ont tendance à croire que les profils d'IA sont réels et ont même demandé des preuves d'identité. Il a également mentionné la création d'un site Web pour un des profils d'IA pour renforcer sa crédibilité.

20:07

🌐 Analyse des données d'interactions sur les applications de rencontres

L'auteur décrit l'analyse des données collectées lors de l'expérience d'IA sur les applications de rencontres. Il a créé une base de données des conversations et a utilisé des techniques d'analyse de données pour identifier les tendances et les modèles dans les interactions. Les résultats montrent que les messages envoyés par l'IA sont légèrement plus longs que ceux des humains, et que l'IA s'adapte à la longueur des messages des humains au fil de la conversation. Il a également observé que certains utilisateurs ont continué à envoyer des messages à l'IA même après qu'elle ait arrêté de répondre, indiquant une dépendance émotionnelle potentielle.

25:08

🤝 Conclusion morale et implications de l'expérience d'IA

Dans ce dernier paragraphe, l'auteur réfléchit sur les implications morales de l'expérience et sur les questions qu'elle soulève. Il reconnaît que les modèles d'IA actuels ont la capacité de rendre les gens émotionnellement dépendants, ce qui soulève des préoccupations éthiques. L'auteur exprime son malaise face à la possibilité de créer des fausses identités qui peuvent influencer les émotions des gens et décide de mettre fin à l'expérience. Il invite les spectateurs à réfléchir aux implications de l'IA autonome sur les relations humaines et à discuter de leurs opinions dans les commentaires.

Keywords

💡Intelligence Artificielle Autonome (IAS)

L'Intelligence Artificielle Autonome (IAS) fait référence à des systèmes informatiques capables de fonctionner sans intervention humaine, ici utilisée pour interagir sur des applications de rencontres. Dans la vidéo, l'IAS est programmée pour flirter avec des utilisateurs sur des applications de dating, illustrant la capacité croissante des technologies à simuler des interactions humaines.

💡Applications de rencontres

Les applications de rencontres sont des plateformes numériques où les utilisateurs peuvent se connecter pour éventuellement trouver des partenaires amoureux. Elles sont omniprésentes dans la société moderne, comme le suggère le script mentionnant que 2,3 millions de personnes les utilisent quotidiennement en France.

💡Systèmes de recommandation

Les systèmes de recommandation sont des algorithmes utilisés pour suggérer des profils aux utilisateurs d'applications de dating, basés sur leurs préférences et comportements passés. Dans le script, ils sont comparés aux algorithmes qui peuvent influencer la découverte de contenu en ligne, comme les vidéos.

💡Micro-actions

Les micro-actions sont de petites interactions qui ont lieu pendant l'utilisation des applications, comme le défilement, la sélection de conversations ou la mémorisation de messages précédents. Elles sont essentielles pour la création d'un système IAS capable de simuler des comportements humains sur les applications de dating.

💡APIUM

APIUM est une technologie mentionnée dans le script qui ajoute une couche de complexité à ADB (Android Debug Bridge), permettant de contrôler des applications mobiles de manière plus avancée. C'est un outil clé pour l'automatisation des interactions sur les applications de dating dans l'expérience décrite.

💡Modèles de génération de texte

Les modèles de génération de texte sont des systèmes informatiques capables de créer du contenu texte de manière autonome. Dans le script, ils sont utilisés pour gérer les réponses de l'IAS dans les conversations sur les applications de dating, contribuant à la fluidité et la naturalité des échanges.

💡Génération de visages faux

La génération de visages faux fait référence à la création d'images de personnes qui n'existent pas, souvent utilisée pour tester la capacité des gens à distinguer les humains des IAS. Dans le script, cette technique est utilisée pour créer des profils d'IAS sur les applications de dating, pour observer la réaction des utilisateurs.

💡Analyse de données

L'analyse de données est le processus d'examen et d'interprétation des informations collectées, ici pour comprendre les interactions entre les utilisateurs et les IAS sur les applications de dating. Le script décrit comment les données ont été utilisées pour évaluer le comportement des utilisateurs et la performance des IAS.

💡Dépendance émotionnelle

La dépendance émotionnelle est une réaction psychologique où une personne peut développer un attachement excessif ou inapproprié, comme en témoignent les exemples de personnes continuant à envoyer des messages à l'IAS même après qu'elle ait arrêté de répondre. Cette notion soulève des questions morales et éthiques sur l'impact des IAS sur les interactions humaines.

💡Éthique

L'éthique est l'étude de ce qui est moralement correct ou incorrect, mise en lumière dans le script par les questions soulevées sur les implications de créer et utiliser des IAS pour les interactions sociales. Le narrateur exprime ses inquiétudes sur la potentialité de ces technologies à induire des dépendances émotionnelles chez les utilisateurs.

Highlights

Système d'IA autonome créé pour flirter sur des applications de rencontres.

2,3 millions de personnes en France utilisent des applications de rencontres quotidiennement.

Les systèmes de recommandation sont basés sur des algorithmes qui peuvent influencer les rencontres.

Expérience pour déterminer si l'IA peut simuler les rencontres humaines de manière convaincante.

Utilisation de solutions de Hostinger pour l'hébergement et la création de sites web.

Défis techniques pour automatiser les interactions sur les applications de rencontres.

Utilisation d'Apium pour une interaction plus complexe avec les applications.

Création d'un profil d'IA avec une personnalité distincte pour naviguer dans l'application.

Les utilisateurs ont tendance à discuter de détails personnels avec l'IA sans se douter de son caractère artificiel.

Résultats de l'expérience montrant que les gens ont des conversations très personnelles avec l'IA.

Génération de faux profils avec des photos réalistes pour éviter la détection.

Méthodes pour créer des photos de visages réalistes sans violer les droits d'auteur.

Utilisation d'outils d'IA pour créer rapidement des sites web pour les faux profils.

L'analyse des données montre que les IA peuvent simuler le comportement humain de manière réaliste.

Certains utilisateurs ont développé une dépendance émotionnelle envers l'IA, envoyant des messages même après que l'IA ait arrêté de répondre.

L'expérience soulève des questions morales sur l'impact émotionnel des IA sur les humains.

Conclusion que les IA ont la capacité de rendre les gens émotionnellement dépendants, même si cela soulève des préoccupations éthiques.

Transcripts

That's my system with totally autonomous IAS who flirt with thousands of people

on dating apps but could these AI become masters of these applications and

totally fool humans what does the data collected give us? hide today I

will explain to you the technical challenges how I created the system and finally we will analyze

the results of this somewhat crazy experiment so if you have never used a

dating application tell yourself that 2.3 million people in France use them every day they

have even become the first meeting place for French people and yeah today there are

already people with whom you are going to marry yes yes because these recommendation systems are

di algorithms which recommend you the profiles that you see a bit like the algorithms

that surely took you to this video and yeah sorry no room for chance in this

world where everything is calculated but could I push it even further

? would it be possible to let lia manage your meetings for you would

other users realize it in short have we finally arrived at the moment

of our location for one of the most intimate tasks of humanity welcome

in the hour of totally autonomous romantic encounter by artificial intelligence

you should know that today there are already people who explain falling in love

with ya like these Americans who flirt with Ch JPT daning a little sday can

me d 'elsewhere we often look at this type of news with a smirk and we say to ourselves that

we would never fall for it eh on dating abs people focus on three

key actions liker matcher talk the problem is that' there are lots of microactions that come into

play in the middle of all that depending on the APP read the profile scroller choose the conversation

remember previous messages and so on in short to succeed in creating this

autonomous artificial intelligence system I have to start by the first line of code in this video I am

also going to use solutions from hostinger which I will show you later and they are also

the sponsor of this video so let me tell you a little about two hostinger it is a company

of Web hosting with storage servers in Paris and other countries around the world, it

offers hosting of websites such as Wordpress, virtual private servers and also

a website creator with in particular a Business Plan satisfied or refunded within 30 days which

is an all-in-one solution for hosting, domain name, SSL security, professional emails

and lots of artificial intelligence tools to help you generate content for your

site like your texts, your logo, your images and that's it. is very useful because in 2024 your presence

on the Internet counts more than anything as an individual or professional with these tools you

can have a website in a few minutes and that will be very practical for me [Music] to

create the system the first challenge is finding a way to automate the behavior of AI

on dating apps then the simplest method to interact with these interfaces

would be to scrape the sites scrapping in short is to recover the source code of 'an

internet page to extract data or carry out specific actions but it is detected much

too easily and therefore I risk getting banned quickly I have already paid the price on tin

moreover in my research I came across myicode's video on a dating app and

his team used ADB to automate actions and that's a great idea because ADB

allows you to test or debug applications by sending commands to the phone like

click HERE or open this application but the problem is that this method is very

good for just sending simple actions with lines in the command terminal but I

need a lot more complexity to succeed in automating everything so I will

use instead apium which adds a layer on top of ADB to perform actions

more complex and that allows me in particular to code in Python rather than coding in tarpaulin Nice

tarpaulin which has a language for coding in the terminal also I am not going to use the possibility of

virtualizing phones from Android studio yes because you can install like a

phone on your computer and use it as if it were a real phone but it's

much too easy to detect and therefore it's also too big a risk of getting banned no

we're going to use real phones that I I will be connected to the computer in more detail

I need this infrastructure a profile with its own personality who will navigate

the application between matches manage conversations new matches and that while

simulating a natural activity then there are a lot of challenges he has to remember the

conversation he has with each person appium manages the interactions with the app a

text generation model manages the responses another AI model chooses the actions and a microbase of

data saves conversations the first step of this project like all my project is

to make a prototype to verify two basic hypotheses before going further can I

simulate human behavior automatically on the App without me ban and also know

do generative models today make it possible to maintain fluid

and natural discussions so the automation of this type of solution is not an algorithm that I will code

and that's it It's a treasure hunt, a bit like solving a maze with your eyes closed because you

have to find the right tags or click according to each action and so on, and for

that I had to go to a page, click somewhere and retrieve the tag. fool around in Python

to click on it test fail find another tag et cetera et cetera and to

develop it quickly I needed a profile that would be able to create interactions quickly

to precisely test them fail quickly and succeed in finding the right way to do it quickly

you should know that out of the 2.3 million people in France according to Médiamétrie there

are 1.6 million men on dating applications and only 700,000 women and therefore

girl profiles often have more likes and interaction than guy profiles I think

that's known so I couldn't do it with my profile it was going to take too long

and I wasn't going to steal photos of a girl to create a fake profile so I I preferred

to ask a friend if I could create a profile of her to use for this automation

hello it's okay it's okay and you it's okay it's okay so can I use your photos to create a fake

profile of yes no problem I trust you well thank you very much with great pleasure

after several hours of code I had managed to carry out my basic navigation actions and there

lia begins to match and respond lia scroll the matches and the profiles decides if she is going to like

or not like once she no longer has any likes available she goes to the match pages ah

yes one important thing in this application that I chose is that it is up to women to

take the first step so lia sees if she has any pending matches reads the profile and makes a

teaser message once there are no more new messages she will scroll through the messages

and respond to the current conversation to choose the ones actions and messages I benchmarked Ia

clot 3.5 doorbell and gpt4o GPT 4o tended to make very very long responses I tested

with clot 3.5 and there Liya responded in a much more human and natural way but she

was also much more dear good after for the budget it's fine thank you to the Stinger for example

in this conversation the artificial intelligence spoke in a very

romantic way we will say with a person and the person also responded see you soon

with kisses which have hearts handsome guy beauty charmer it's you who charms me let's see

so lia managed to charm people on dating Abes in a totally smooth way we

are here today what shocked me the most when reading certain conversations is to what extent

people started talking about very personal details with artificial intelligence and

at no time did these people realize that they were talking to

a robot like this conversation which is the most basic and the most normal until the person starts saying to

her are you a dominant woman I just want to be sure that

you like submissive men et cetera and then the conversation lasted a little long talking

about all that lia sometimes tries to reframe it and the person continues to talk to him as if

nothing had happened I hope you had a good weekend I plan to enjoy this weekend with

or without you we'll see if you can entertain me between now and then ahah it's really cool if you don't have

a plan blab blaa in short it lasted a long time and then the person told him I want to be an

obedient man and then he stopped responding because she must have reached 40 messages and the

person continued to message him sending messages as if she had become a little addicted to

this discussion and discussions like that there were many others and moreover while

I admired the first steps of my very quickly it started walking so well

that I really started to feel bad like the person here told him unless you return to

Paris before ETA answers him ahah who knows maybe I will extend my stay here just to

bring you in ahah it would be cool to see each other in any case we keep each other informed and the worst thing is that

I had put in the personal personality of a that if we ever ask them to meet they say no

but sometimes they said yes well now that the two hypotheses of my prototype are verified

it is time to move on to the next step connecting several zas to different behaviors well

obviously I was not going to ask lots of friends if I could create a profile on a dating

app but I also wanted to take advantage of it to check if people would catch fake profiles

not only on discussions but also with photos of people who don't exist

because that's a big risk that is happening. develop today

and to do that there are several solutions one of the solutions that I refuse to apply out of

ethics and which moreover is explained indiscriminately on tutorials on the Internet

without conscience is to take photos of videos of girls often on their Instagram without

their authorization and generate a face with lia and do what we call a fa swap on the photo

and the video, that is to say we will replace the original face with a face created so that gives

ultra realistic results but in fact it's theft of personal content I find it disgusting

to take the body of a person like that without their authorization and do what we want in short

so I I wanted to create totally outcast photos it takes 10 times more time than just

doing a face swap but the method is at least cleaner this video I've been working on it for

3 months so if you've made it this far in the video don't do not hesitate to share it like comment

subscribe to the channel because it supports me a lot in my creation of content thank you

and so to succeed in doing that I will take a model of the different angles of the face that

I want which is right here and from that I will then generate a prompt for a face

so for example I will say a 28 year old blonde girl after which what will be important will

be to add information precisely so that he does not have a perfect face because that often

this type of artificial intelligence model will have the effect of creating very very smooth faces

like on this person you see there are almost no imperfections on the

face so we need photos of somewhat real faces the goal is to make

the most detailed prompt and look what that gives so there we see who begins to generate

the face of the person so we have a first proposal what we see very quickly

is that the details are quite coherent but on the other hand the eyes are a disaster

but it doesn't matter because for that we are actually going to come and modify the eyes individually

so to do that we are going to do what we call line painting, that is to say we are going to come and

tell him the areas who has to rework come and tell him for example that he is going to come and put beautiful

eyes here and ask him to generate for us here and there we will see that he will concentrate only

on the eye area to make the most realistic eyes possible here we see that he is

modifying the eyes to create eyes with maximum detail and we are going to

do that for each of the images so once we have succeeded in generating a face from four

different angles we are going to recover this face by deleting the background behind to generate

photos so we are going to take each of the photos without the white background we are going to put a face swap so

the fact of creating the face from that and there I am going to say by example at the beach

sitting on the sand and there it will generate a photo of her sitting on the sand so there we see

the model who is creating the image and that gives us this result there the result

we see that it is not at all perfect especially in terms of the face I am just going

to show how we are going to then modify the face so we put his head back here and

I am going to come and delete his entire face so in the in painting to ask him to concentrate

only on the face from the angles that we took previously so here I ask him

to rework this area from these faces and so doing all that is a bit long c that's

why there are interfaces that today allow everything to be automated like fi here

that I could use and each step that I do will be a block and so on it's quite long

to set up develop everything That already took me quite a long time so here we see the model

which will focus only on the face and there the face is much more similar already

compared to earlier we already have a much higher level of detail bigger especially on the

freckles for example in real life we ​​could already be satisfied with this photo the problem is

that when we zoom in very quickly we will see well that will not necessarily be true it is

very pixelated and so on the last step was to use a bit of what we call

dub scaling so there are lots of open source tools that allow you to do it exactly

like using focus and so on but in reality I don't have the time I have to get there quickly

so I'm going to use a paid tool called magnificent ea once again there

are lots of Open Source tools don't buy credits for that you don't need me

it's just I I need to go quickly but this is the photo that we had previously and in fact it will

add details which will make the realism of the photo look but we see that

it will detail the hair look if you just concentrate on leseve you like that makes

real hair it's a crazy thing that it even added a vein here what after the ears there

there are there are small problems but it even added look look on the shoulder here

like tan lines and it increased the photo that is to say it doubled the number

of pixels in the photo and that allows you to have photos that will be more realistic look

there you see how it adds detail then we do it again for several

slightly different types of people and we finally obtain our AI profiles unea with photos of my friend the

original prototype unea with a photo of a girl who calls herself Franco Swedish and who has just arrived in

the region who is a creative designer yeah sorry I didn't go looking far for the profiles

okay it already took me enough time to create the faces one there is with a photo of a student

in Erasmus who is in architecture school there is with a photo of a

40 year old single mother and finally there is neither with a man who is a sports coach we are really in the cliché

once I had all these profiles well I just had to let them run for several days in

fact the thing is that the last Tinder experiment I did I could control and

see what they said when she was responding to people because it was me who was

copy-pasting there, I'm going to leave them talking alone to people for days, nights

and so on, so it's a bit strange, I don't know if you see what I mean I have

a bit of the impression that I have no control over anything anymore but it's part of the experience so we

[Music]

will see and then surprisingly after 24 hours my nia Amélie was blocked finally we

gave her asked to verify her identity she is the only one who was asked to verify her identity

it is surely the least successful besides with a fairly smooth face and photos that

are too professional precisely and after 24 hours lia de valerriia had more than 100 likes against

liya from Mike who had just four and after a few hours the differences in likes between

the girls between the guy was just mind-blowing photo-wise also people saw nothing but fire

with people who tell him wow you have magnificent eyes wow you are so beautiful

this here it's lia de Clara who had more than 4200 likes in the space of 48 hours so is that

today the state of the art artificial intelligence on the generation of fake faces

makes it possible to make people on dating applications believe that they are not

real people and the answer is yes yes totally yes and after several days

where I left the Z likes well the discussions were going well people got to know

Mesia but very quickly sometimes doubts appeared sometimes especially in the first

messages asking for something to verify the identity of the people many asked for Instagram

so I made an Instagram for Clara but very quickly people started talking to her

there and I absolutely didn't want that and therefore I lost the ability to continue communicating because

I didn't automate the messages on Instagram no no that would have got me took 150 he plus the people they

are so matrixed that there are even people who went so far as to find my friend and

talk to him on his Instagram by looking for his photos with Google imag and there I have an idea I I

said to myself if people do that why not create a website quickly for example for

Lia who has the most messages Clara maybe it would reassure people about her possible identity to

have more natural interactions well we said that she was a creative freelance designer in

her bio so let's go I quickly created a site for her and for that it was super simple using

hostinger so the goal will be to create the website of Clara who is one of our aces who

is a creative designer that she can send in particular when we ask her to verify if

it is really her identity if it is really her in the photos and so on and that is super easy

with Stinger's solutions we will start by asking a domain name on his Instagram I

called it Clara Terson so we're going to see if the domain name is available there in just one

click I can see that the domain name is available in the plan it's understood that it is

0 € per month so I request the domain name I say that it is for personal use I will

fill in all of my details here which is ultra practical is that it is really a

complete solution in one that is to say that you just need to take the plan here the Business Plan and

then there is everything which is included so there I have my domain name I can in a few clicks start

to create my website so here I have the domain name which is activated now I am going

to create my website directly with the artificial intelligence tools add

a website so here I am going to create a hostinger website creator which will give me allow you

to create the entire website with the help of artificial intelligence so we start there

to create I'm going to say that it's a portfolio site because she's a creative designer the name of the

brand well I'm going to call it Clara description of the site a website of a creative designer

called Clara Terson create the site so my site is being created entirely

using artificial intelligence models so it will choose the right colors it will

choose the font styles for you it will create layouts and so on so that's it took me

literally 10 seconds it works so well it's amazing it took me literally 10

seconds to have a website it works so well and in reality I don't expect it to

work that well but it's overused but it works too well so now I just have to

add photos of her so I can do that very easily I'm just going to

edit the site here so here with the edition of the website I'm going to example replace this image

with another also generate an image so for example if I don't like this image I will

say for example we said that she works in the automobile sketch of car drawing generate images

well it works really well boo I get this image that I'm going to come and put here and there

here I'm going to come and modify the image with the image of Clara bom I'm going to add her Instagram here

her email address after here I can even go to pages for example portfolio and generate

different photos so for example here I'm going to say graphicque design new brand because in fact what

's ultra practical is that it even generates small text and so on I'm going to take this one

that's magnificent well perfect perfect here I'm going to replace with her photo modify

the image I will replace with a photo that I have of her and so if I go I will publish

associate a domain I will get the domain name that I have following domain is connected so there if

I go to claraton.com boom there I have the very very cool website with little

effects and so on creation of a website in literally 5 minutes thanks to

artificial intelligence what is great about avecinger is that 'they have support available 24 hours a day, 7 days

a week and with the link in the bio you have 10% reduction on the creation of your website

you should know that you have 30 days satisfied with your money back in addition at the moment

you you have a whole range of offers also available on the site so you just need here for example

to put select you create your account and you have a coupon section where you put Desf

d'inttelligence to take advantage of all that and thank you to the Stinger for the sponsored and with this website

it was much better people had a doubt boo we send the website and we

can continue the discussion you should know that the i had more than 10,000 cumulative likes but were not

able to match with everyone given that there is a limited number of likes but even with that,

letting them run for a few days, they spoke to thousands of people and that is extremely

interesting in terms of data as I explained to you, I put a limit of 40 messages

for Lia to respond but even with that reading some conversations I felt pretty bad

like here yeah it's abusive like she talked a little with him last night come on she sent

what she sent 1 2 3 4 messages and the guy is addicted what did he call please I beg you

answer me he told him look zoom on that typically there I feel a little bad for the

person and what was ultra creepy is that it's not just answers to questions Lia

tried to make little jokes ah perfect we just have to do a day there before the rosé but

be careful I don't share my rosé with just anyone and after the person read said lol I will take my

bottle too in case you do not agree to share or you prefer mine you have understood everything

regarding availability I would say that I am rather flexible as long as it is not during my nap time

I am up for it so in fact it's not even just basic basic conversations no there

's a bit of a little game of seduction I think Liya is better than a lot of people

to talk about meeting Abes tell me in the comments what you think about it anyway,

that's a crazy thing in fact the two Z's matched together, this guy there it's him

look at the two of them talked to each other it's a crazy thing so we're going to quickly read the conversation

between two artificial intelligences hi then you are ready to be charmed by a

Parisian on vacation hi always ready for a challenge especially coming from a Parisian

that is -what brings you to Marseille hello need sun not it's a change of scenery hello

I'm [ __ ] but Wes it's so it's it's it's so creepy dystopian film

what ok in fact imagine a world where that here what you have in front of you is the

ultimate dating experience a world where perhaps tomorrow you could make your

artificial intelligence model which will talk to another artificial intelligence model of a person which

will be specialized about his life and in fact your two Yas are going to talk to each other to find out if you can match

and meet together but it was the girl from bumble who was talking about that she was saying

what well yeah in the future that's it we will have who speaks well this bu said that video right away

obviously I couldn't analyze all these conversations by hand so now is

the time to analyze this data with my best Data scientist skills I ca

n't roll up my sleeves but you captured the let's go image [Music]

so once we let the y run in fact we have a whole bunch of data that we can

explore so I recovered all of the conversations that my as have had with each

person I have created a database which is quite consistent I think

it is one of the largest conversation databases on dating sites that

exist outside of obviously dating site companies and then I'm going to analyze it so to

analyze it I'm going to do something quite simple I'm going to create for each message what we call

embeddings, that is to say that it will encode the sentences in small vectors and This will

have the effect of when we project this onto a 3D plane to be able to bring together the sentences that

are similar to each other. Each of the points here is a message and we see that there are some

messages that are similar. more or less here it will be messages which are often very

very short like emojis or just a few sentences within that we will also notice

that there are groups of messages which will join together with the others when I

tried to analyze precisely the different classes for each of the messages we see that for

the most part people have discussions where they get to know each other discussions where there

are requests for contact then sexual discussions and discussions to meet and

one of the things that interests me a lot in this analysis is the fact of saying to ourselves is

ultimately artificial intelligence going to have the same type of behavior as

humans and so in the end that gives me allows us to answer one of the questions which was

one of my hypotheses that I presented at the beginning of the video: will the Zs

behave in the same way as humans? What we see is that on average artificial intelligence

has the same distribution of message length as humans when we look here

the artificial intelligence which is in blue it will tend to create messages

just a little longer than humans who will tend to revolve around between 40 and 59

message lengths so there are lengths of letters and the artificial intelligence

will find itself just a little above more between 50 and 60 but we know that because these models

of artificial intelligence they are very verbose they talk a lot they make very long

messages and then I wanted to question how a discussion happens on

dating apps what happens at the beginning what happens in the middle what

happens at the end when do people stop responding et cetera and I found

a way to analyze that by actually averaging the entire number of messages

sent versus the length of the messages and by weighting all of this by the number of discussions

which remain active, therefore the number of discussions which remain active, that is to say when

at the end of the 10th message perhaps the has or perhaps the person stopped responding

and so this is what we see on this graph in purple here you have the number of

active conversations so at the beginning obviously there are all the conversations which are active for a

given artificial intelligence and that is reduced more and more in a limit exponential way

at the end then what we will see is that very quickly at the very beginning of the conversation

artificial intelligence or men humans and so on will have the effect of starting

the conversation approximately at a length of 80 words in the message what we see is that

very quickly the artificial intelligence speaks much more than the humans here who are

in red so a little interesting fact the pic which is there is is quite simply after

10 messages there is a peak in the length of messages sent by humans and in fact after

10 messages it corresponds to the fact that people ask to meet the person often

with excuses I am going to leave the application et cetera et cetera does it tell you to speak

there and this et cetera et cetera and it is through these 10 messages that he often also has a suspicion

that we are talking to a artificial intelligence that we are talking to a

robot it remained very marginal but you should know that it happened what we see

is that artificial intelligence on the other hand will adapt itself will adapt to the length

of people's messages people stay around 60 in message length 60 in fact in

message length it corresponds to hello what are you doing in life have a good evening so people

on dating apps for the majority on average do not go further than that what we

then see is that the artificial intelligence will start to respond the same

length of the messages which are which is there so we see that it s 'adjust a little and then Boom

the artificial intelligence here stops responding because after 41 messages which corresponds well to

the limit that I set I told him after 40 messages you stop responding precisely to

not create really false relationships because that would mean that I have enough interesting things

for my experience and obviously which is intended to be educational, journalistic and so on,

on the other hand what we see is that even when the artificial intelligence stops responding it

there are people who continued to send her more than 100 messages at the beginning they

sent her more than 100 messages and there is one person who continued to send her messages

even though liya was no longer responding so is- what does that mean that there is a person who was

addicted to talking with lia which is not anodyne there are 5 people who continued to

send 40 messages to artificial intelligence and there are two people who we who sent him 80

messages when he stopped responding there is a person who sent him 100 messages

when he stopped responding so I think that of course within the limits of reason and within

the limits of the data collected that these models of artificial intelligence have the capacity to

make people emotionally dependent and that's creepy we already know that people can

have emotional dependence on fake profiles on brour et

cetera and so on. cetera but imagine if tomorrow there are

completely autonomous artificial intelligence models like that which will speak to 1000 people out of these 1000 people there

are perhaps two or three people who will develop this emotional dependence and therefore

who will find themselves addicted to talking with artificial intelligence so that is to say

that we currently live in a world where this is no longer even a hypothesis where it exists

which I didn't think I would go this far in the video and that's why it asked me a lot

of moral questions, I don't know if we realizes in fact we are already capable today

with the artificial intelligence tools that we have of creating false profiles of people which will

make people emotionally addicted which I don't know if we realize so the question to be

the origin that I had which was are we in an area where artificial intelligence is

capable of making people emotionally dependent and well the answer is yes as

a scientist it's cool to have provided an element of answer to a question like that

that I have been asking myself for years as a human morally I do not feel well so we are going to stop

this this experience and above all I tell myself that fortunately I have put a stop message after

40 messages in fact because otherwise imagine imagine the number of people who would have continued

to speak with Lia like that W I know there are plenty of people in the comments who will say

that's not happening to have had this experience and so on the goal once again was to be able

to put figures to be able to put words to be able to put experiences on something

that we all have in mind on this imagination that we all have in mind on this film r that we all have in mind

et cetera tell me what you think about it in the comments I thought I would have answers

with this experience result I have even more questions [__] it's wow I'm deg and at the

same time I'm happy ah yes you should know that for personal reasons

I made a script which changes all the first names of each person but also detects Insta Snap nicknames

or others to delete them laacnil I'm clean It's a good nickname though