During the 2019 Tech Festival in Copenhagen, Denmark Klint Marketing‘s CEO Taylor Ryan covered how Ai generated content will break the internet. We received a lot of feedback and questions about AI and how it will affect social media companies. See the presentation below.
With so many interesting questions flooding in, we decided to write about the most interesting ways Artificial intelligence and social media companies are becoming intertwined.
The impact of artificial intelligence on social media companies It is a recurrent topic nowadays that creates perplexity and skepticism among marketers.
Is AI taking over social media? If so- then how?
Ai and Social Media – It’s already here
In this article, we will explore various Artificial Intelligence (AI) and Machine Learning (ML) applications. As well as the impact that these new technologies are having on social media companies, and potentially will have in the near future.
What is Artificial Intelligence and Machine Learning?
The definitions from Merriam-webster for Artificial Intelligence (AI) is:
“AI is a branch of computer science dealing with the simulation of intelligent behaviour in computers or the capability of a machine to imitate intelligent human behaviour”.
Regarding Machine Learning (ML) we will borrow the definition from Expert System
“Machine Learning is an application of Artificial Intelligence that provides systems with the ability to automatically learn and improve from experience without being explicitly programmed”.
Natural Language Generation (NLG) is the most interesting part for content creators.
It is a software process that with dataset input can provide written or spoken narrative.
Basically it gathers a huge amount of data and provides humans with text easy to understand.
How does AI work?
We are not going deep in a talk about engineering and software development. Nevertheless here is a basic explanation:
AI uses Machine learning to try to imitate human intelligence. With the use of algorithms and historical data, AI can make predictions.
This is only a theoretical explanation of Machine learning.
However, at today’s current level of progress, AI in social media can be a powerful help for social media companies and not only.
How are AI and ML used in Social Media?
According to this AI in the social media market statistics, the worth of the market is estimated to be around 2,197 million $ by 2023. This creates an enormous amount of opportunities for SoMe companies.
The applications of AI in social media companies are vast. Big firms have been using AI for a long time, spending money acquiring small companies and improving their existing platforms.
You will be surprised how your favorite social media apps are using Artificial intelligence and machine learning.
Examples of how Artificial Intelligence is used on Social media platforms:
- Analyzing text
- Analyzing pictures
- Detecting spam
- Avoiding propaganda
- Deciding content flow
- Social Insights
- Data gathering
Facebook has advanced Machine Learning and Artificial intelligence
Facebook is the leading social media platform when it comes to cutting edge technology that re-purposes user data across hundreds of thousands of experiments broken down into millions of accounts.
The company is the most popular social media network in the world with 2.4 billion users. Users can upload pictures, watch videos, engage in groups and many other functions.
Each day Facebook collects tons of data. This data brings value to the company and gives responsibility to Facebook.
How does Facebook handle crazy amounts of data?
Here is where artificial intelligence comes in handy. Facebook is covering multiple areas where AI and ML performs.
How is Facebook using Machine learning and Artificial intelligence?
Picture analyzing by Facebook
Uses Machine Learning to recognize your face in photos.
Facebook is using face recognition to find users in a picture that they are not aware of. This also helps you to find Catfish (People creating a fake account with your profile picture).
Also since the algorithm has a text explanation of your picture it can help people with visual disabilities in telling them what is on a picture.
The company stated that the face recognition feature is disabled normally, but can be opt-in in the settings.
Facebook Machine Learning Analyze Text
Facebook created an ai tool called DeepText. This tool helps the company to recognize the meaning of conversations on the platform. Understanding the topic can lead to a more accurate advertisement to the users.
Facebook’s AI saves lives by Preventing Suicide
With the same tool, Deep Text, Facebook can recognize, for example, posts that would represent suicidal thoughts.
There is around 1 death of suicide every 40 seconds in the world, and Facebook can do a lot to prevent it.
With the aid of an analysis based on human moderators, Facebook can send ads with suicide prevention materials to these specific users.
Ai helps Facebook to deal with Bad content
Again with the same tool, DeepText, Facebook is flagging automatically bad content posted on the platform. After recognized by the AI, human interactions take place to actually understand the content.
What is defined as “Bad content” by Facebook?
In the company guidelines, we can see a few of the things which are flagged as bad content by Facebook:
- Nudity or sexually suggestive content.
- Hate speech, credible threats or direct attacks on an individual or group.
- Content that contains self-harm or excessive violence.
- Fake or impostor profiles.
Check the Facebook community standard to understand more.
Facebook is Mapping Population Density
Through the aid of AI, satellite images and population data, Facebook is mapping the entire world population density. The company is mainly doing it for humanitarian purposes.
Free to download, thanks to this no-profit organizations can better understand the population density, especially in rural places.
In this gif, you can see how the tool works. First, they take aside locations that couldn’t contain buildings. Then it ranks each remaining location with the likelihood that it could contain buildings. The highest likelihood is shown as blue dots. Each one of them is given the population from census data, which is shown as glowing.
In the end, the company inserts the distributed population data on the locations.
Facebook’s AI can do automatic Translation
AI is also adapted in facebook to translate automatically post in different languages. This is helping the translation to be more accurate and personal.
For example, translating from German to English, the tool works in a way of combining three different language models. German to English + English to German + English model.
Inserting data in each model will rank the translation, until the end where adding all the rank will give the most likely translation.
Instagram with AI
Instagram is a social media platform we all know well. Users can upload pictures and videos of their lief and share them with their friends and followers.
The platform is used by individuals and businesses as well as pets and fictional characters.
Each day Instagram is processing content generated by 1 billion active users. Additionally Instagram ahs to ensure users receive the content they like and engage with.
Managing all the information manually is impossible. Instagram has developed algorithms and AI models to ensure the best possible platform experience for its users.
How is Instagram using Artificial intelligence?
Instagram’s Ai is deciding what gets on your feed
The video and photos sharing company Instagram, Uses AI in the Explore feature.
The posts you will get on your personal Explore feature is suited to your interests based on the likes and account you follow.
Through an AI system based they created a ranking system that “extract 65 billion features and makes 90 million model predictions every second.”
This is fundamental for Instagram, the gigantic amount of data they are collecting can help in showing users what they like and will like.
[An interesting fact about this topic, on Instagram they develop a computer made influencer, Lilmiquela, which has currently collected 1.8 million followers.]
Instagram’s AI is improving target advertisement
Through an accurate study of the data gathered, Instagram can keep track, for example, of which posts users engage the most or search preferences.
After, AI helps Instagram in making target advertisements for companies based on all the data.
AI helps Instagram in Filter spam
Ai can recognize and remove spam messages from users’ accounts in 9 different languages. Also with the aid of the DeepText tool from Facebook they can understand the context in most of the situations to do more accurate filtering.
Instagrams AI fights against Cyberbullying and Bad content
Being the social media where most of the Cyberbullying happens, according to this study, Instagram battle against bad content and bullies is fundamental.
While Facebook and Twitter rely mostly on reports from users, Instagram has been using AI extensively to automatically check content based on hashtags from other users. Once something that does not follow the community guidelines is found, the AI removes it instantly.
What is defined as bad content by Instagram?
We already cover Facebook community guidelines on bad content, and since Instagram is owned by Facebook the guidelines are more or less the same.
Twitter and AI
On average users post 6,000 tweets per second on the platform. AI is a necessity when dealing with such a great amount of data.
How is Twitter is using artificial intelligence?
Like its social media competitors, Twitter is using AI to filter tweets looking for bad content. Since the entire platform relies on live tweets, the AI must be fast and accurate in blocking doubtful content before it spreads.
But this is not the only way the company has implemented and used AI.
Twitter and AI for tweet recommendations
Artificial intelligence helps twitter to detect fraud, propaganda and to remove hateful accounts.
Twitter implemented AI firstly to give users a better UX (user experience) capable of finding interesting tweets suitable individually.
How does the Twitter recommendation algorithm work?
The algorithm works in an interesting way, it learns from your actions on the platform. Every tweet is ranked to decide how likely is the tweet to be of interest to the individual user. The ranking model considers different factors, for example:
- The actual content of the tweet
- If it has a video
- If it has a picture
- Number of retweets
- Number of likes
Furthermore, It also checks if you had previous interaction with the creator of the tweet itself and how interested you could be to the creator of the content.
It also considers your history of engaging with what type of tweets and how similar are the tweets that the platform wants to recommend.
The higher the tweet scores in this ranking model, the more likely the user will see it on your page and in “in case you missed it” section.
Twitter is filtering tweets through AI
The company has been using AI to take down inappropriate accounts on the platform. In 2017 Twitter took down over 300.000 accounts connected to terrorism.
Or to take down accounts that go against twitter rules on platform manipulation and spam, as seen when they suspended 70 accounts Pro-Michael Bloomberg.
Twitter using AI to automatically make your pictures look better
AI is also used in the company to crop images automatically to make an image more appealing.
How does the Twitter AI Cropping tool work?
Pictures have been introduced in the platform in 2011. Since then, Twitter has been working on an algorithm capable of automatically cropping images in order to make them more appealing.
Firstly they made an algorithm that focuses on cropping the images based on face recognition, which brought up various problems. Mostly related to the fact that not every image has a face in it.
The algorithm was not perfect, if it did not find any face it would automatically crop the image from the center which often led to awkward pictures.
Introducing Saliency studies
So the company started using studies which focus on Saliency. Having a region on an image with high saliency means that an observer will most likely look at it when viewing it freely.
Basically with the use of this theory, the AI can predict and crop the image focusing on the most interesting parts.
The only issue with this Saliency technique is that it results to be too slow for the platform. The company wants to keep on cropping images without failing to let users post in real-time.
How did twitter manage to fasten the process with AI?
They use a technique called Knowledge Distillation in order to train a smaller network to imitate the slower but powerful network.
A large network is then used to generate predictions on a set of images.
Together with some third-party saliency data are used to train a smaller and faster network.
Finally, they developed a pruning technique to remove part of the neural network that did not contribute much to the performance.
Using these two models Twitter managed to crop media 10x faster than before. This allows them to crop the images as soon as they are uploaded.
The outcome is outstanding:
AI behind Linkedin
Linkedin is a platform that is known best for business connections.
AI behind Linkedin slowly powers everything.
Machine Learning works to learn what happens in real-time as the user is on the platform.
How Linkedin use AI?
Giving users job recommendations, profile connections and showing helpful content in the feed, and much more.
Linkedin AI and harmful content
Like Facebook and Twitter, AI is used on Linkedin in the background to check for harmful content.
The biggest challenge for Linkedin is detecting for example:
- Fake accounts
- Harmful content
What’s Linked-in’s way to find and delete bad content?
Firstly Linked in identifies a set of words and sentences to add in a blacklist. Once a profile was found posting about content in the blacklist they would delete the account.
Even though it seems to have a good outcome, the approach resulted to be a manual process after all.
There were chances of misunderstanding context which made it impossible to do it fully automatically.
Linkedin started to use Machine learning
To solve the issues related to identifying bad content, Linkedin started to use Machine Learning.
At first, the company labeled a set of accounts with “Inappropriate” or “Appropriate”. To find accounts that were removed due to their harmful content.
To better understand the context of a phrase, Linkedin implemented a type of deep learning machine called Convolutional Neural Network (CNN).
As an example of these types of deep learning, when the AI detects a bad word such as “Escort” it can learn that it is not always harmful since sometimes it can be used in contexts like “Medical escort” or “Security escort”.
Linkedin AI recruiting.
AI in Linkedin analyzes hiring patterns and job locations to give suggestions on the best candidates that suit your job posting.
The biggest challenge in this matter is that the AI has to understand and learn preferences from both the job seeker and the job poster.
Moreover has to unify different aspects such as skills required, roles preferred and/or job locations.
All this info and much more insights are gathered and connected in real-time thanks to the AI.
TikTok is training AI.
The newest social media app, TikTok is having a huge global success and it heavily relies on an AI platform as well. With the help of an algorithm, it quickly learns the user preferences in order to give personal suggestions.
How is TikTok is leveraging AI?
TikTok uses AI to train algorithms to produce more engagement. The more you engage with a certain type of video the more users like the more similar content you’ll end up seeing.
The TikTok algorithm monitors how and when users like comments and watch specific videos. The algorithm then determines what is unique to users’ content preferences.
After multiple watched videos we can notice results in a higher likelihood of seeing more content just like we prefer. This ultimately means that every time you open the app, the more likely the videos that will fit your taste.
It’s amazing how much data can be used just for the purposes of entertaining you on a millisecond basis.
Every day TikTok improves its algorithm to predict what will keep you using the platform for a little bit longer.
Do you need help editing on TikTok? Then give a look at our guide on how to upload edited TikTok Videos from pc to the app!)
Snapchat acquired two AI companies
The social media company started silently acquiring AI companies. It first started with the Ukrainian startup Looksery in 2015 in order to improve its animated lenses.
Then it acquired AI Factory to start enhancing their video capabilities.
How Snapchat is using artificial intelligence
Snapchat is using AI to recognize text in videos
The company uses AI companies to recognize text in the video which will then add content to the “Snap”. For example, texting “Wow” will automatically create a comic icon in the video.
AI for the Snapchat Cameo Feature
Or AI is used in Snapchat to edit someone’s face in a video. Once a user has a picture of themselves they can then insert it into a short video through the new Cameo Feature.
Youtube and AI
The major video platform company uses AI in different parts of their website.
Youtube’s AI is skimming through the videos!
An AI application is used to skim through all the videos that are being posted constantly in order to find objectionable content. This kind of ai must be powerful enough to check thousands of videos every day.
According to Forbes, 8,3 million videos were removed in the first quarter of 2019 and 79% were automatically flagged and identified by AI. With over three third of the videos didn’t have any views at all.
This shows how important is for Youtube to remove videos that are against the company policy guidelines.
Nevertheless, since the numerous time where AI misinterpreted the context, a human touch is still necessary.
AI in the Up Next feature, Youtube recommendations.
Another way in which youtube is using AI is in the “Up Next” feature. The algorithm is working in real-time to suggest the best video for you, even filtering among the recent uploads.
The initial algorithm used to be based on how many people clicked a video. While it was a smart idea, videos watched per user did not grow. So Youtube based the algorithm on how long people had spent watching a video.
This made poorly made videos dropped in views while the more high-quality videos grow since it’s extremely correlated to a longer watching time.
Introducing Google Brain
Google, the owner of the video platform decided to implement Google Brain.
Google Brain uses a technique called Unsupervised learning. The main characteristic is how easy for the algorithm to find relationships between different inputs.
For example, Brain algorithm started recommending shorter videos for mobile users and longer for users on Youtube’s TV app.
Pinterest uses AI to capture our imagination
AI is deeply used in Pinterest, the company co-founder, Evan Sharp, noticed that while users usually tend to spend time on other platforms when they come to Pinterest they invest their time to find answers to their problem.
Which opened a great challenge for the company to give valuable content as fast as possible.
Pinterest uses AI for discovering user’s intention
One of the main challenges for Pinterest is discovering the user’s intention when they are on the platform. Usually, when there is an input in the search field, 75% are three words or fewer. Which is a good start but still not a clear answer.
That’s where AI comes in help with the deep learning-powered search.
For example, when a user searches for “kitchen” the entire discovery path can go from pictures of a kitchen until recipes suitable in a small kitchen. It’s outstanding how far these searches can go and have a more deep meaning.
Pinterest uses AI for recommendations.
We already talked about how Facebook, Instagram, Linkedin, and Twitter use AI to give recommendations to users.
It is not surprising that also Pinterest uses AI for recommendations.
Developed with a mix of TensorFlow and PyTorch deep learning frameworks on Amazon Web Services (AWS), PinSage is a neural network used by Pinterest to give relevant recommendations.
How does PinSage work?
A huge amount of images (nodes) form a graph, then 18 billion of lines (edges) connect them together. The output is a deep detailed context for every image.
With all this information on images, Pinterest can give thematically similar images to users.
Pinterest Lens Camera Search powered by AI
In addition to the possibility to search for terms or words on Pinterest, the company added a feature called Pinterest Lens camera search.
Through the app, a user, can take a picture with the camera and uploaded it in the Pinterest search engine.
From here, Pinterest uses AI to recognize shapes and objects in the picture and suggest you related themes and products.
How does it work?
Pinterest engine breaks the image into segments, such as:
- Product category
It also includes a method to account for unclear images.
The process is easily understandable through this diagram from TechCrunch.
With the input of an image, in this case, shoes, the AI breaks down all the different parts until it recognizes exactly the type of shoes.
Then based on a blend of results from Visual Search, Object Search, Image Search, and more query info, the algorithm gives the perfect recommendations.
Pinterest uses AI to reduce self-harm content
Joining the other main social media platforms, Pinterest has a deep interest in helping users avoid any self-harm.
With the aid of Machine Learning techniques they discover and hide any content which can lead to or encourages self-injury.
In this way, the company decreased by 88% the reports of these kinds of contents over last year.
In addition, Pinterest introduced a collection of “Emotional well-being activities” in their app.
Firstly the collection would appear only at the moment someone will look for something that shows they are feeling down, such as: “Stress relief” or “Sad Quotes”. But now any user can just type #pinterestwellbeing to access the content.
Tinder and AI
The most successful dating app Tinder has over 3.8 million users.
It’s not surprising that Tinder uses AI.
How Tinder uses AI?
The main scope of Tinder is to suggest perfect matches perfectly suitable for you.
At the same time, they have to be careful blocking harmful content that could bother users.
In both cases, AI is implemented in the company to help.
Tinder uses Machine Learning to help flag offensive messages and inappropriate behavior.
With very little research online it’s easy to find a variety of offensive and inappropriate behavior on Tinder. For example Tinder Nightmares account.
The company decided to turn to Artificial Intelligence in order to help people with this bad behavior. Together with Machine Learning, Tinder is screening potentially offensive messages.
How does it work?
When the Machine Learning, after screening every message, find potentially offensive messages will flag it. Once it’s flagged the recipient gets a question from the App, “Does this bother you?“, if the answer is yes, Tinder will bring them to the report form.
This is still not automated, AI can of course flag potential bad behavior but an offensive message for someone could be interpreted in different ways for someone else.
For as much intelligent an AI can be, it still cannot predict reactions from individual users.
How did they solve the issue?
A way to solve these issues is that the company is trying to train its machine-learning with a large amount of already reported messages. The scope is to find keywords and patterns potentially offensive.
The algorithm becomes more accurate every time a message is reported
A new feature is also coming to Tinder, called Undo, which will ask users if they would like to take back messages containing bad behavior before it’s sent.
Tinder’s goal is to have a model that is completely custom built on tolerances and preferences of each user.
Tinder uses human-assisted AI to catch fake profiles
Photo verification is a new feature on Tinder to assure that the user you are chatting with is the real person. This feature allows users to self-authenticate themselves through real-time posed selfies.
Then a human-assisted AI compares the selfies with existing profile pictures.
Once they are verified, the users will get a blue checkmark to trust their authenticity.
Tinder uses AI for the Super Likeable feature.
Super like is a feature in Tinder which allows you to become more visible to the user you super liked.
Tinder data says that using super likes it increase the match chances of at least three times.
The company did not explain in detail how the feature works, but the basic idea is that randomly during your swiping you get a chance of Super Likeable.
The feature displays 4 different profiles and you have the opportunity to give a free super like to one of them.
According to the blog post, Tinder uses Artificial Intelligence to propose you the most accurate profile based on your previous swiping.
“Try it out to see if we know you better than you know yourself. ;)”
How does Bumble use AI?
The former co-founder of Tinder, Whitney Wolfe Herd, created a dating app called Bumble.
The concept is the same, the main difference with Tinder is that during heterosexual matches, the girl will have to start the conversation.
Bumble puts the safety of its users from bad content as a priority.
Bumble’s AI blurs your inappropriate images.
One of the other differences with Tinder is that on Bumble users have the opportunity of sending each other pictures.
Users have to tap on every image since all are automatically blurred. Which appears to be not enough for the company.
With an AI tool, called Private Detector, Bumble is able to detect inappropriate messages and warns you. The company says that the feature is 98% effective.
And soon the Private Detector will also be able to detect shirtless mirror selfies and pictures of guns.
What are the benefits of Artificial Intelligence in Social Media Marketing?
From the big firms we can extract a list of benefits of AI in Social Media Companies, as for example:
- AI can help in improving the user’s experience
- AI can predict the user’s behavior
- Give a more personalized experience to the users.
- AI can help you gather valuable data.
- AI can recognize a delete bad content.
And not only these, but AI can also help you in understanding human psychology. Therefore it can track multiple characteristics of your customer behavior.
- How much time do they spend online?
- Which is the favorite platform?
- Why do they use social media and for what?
- Where do they use social media?
The point is to answer these questions for thousands of users and combine them with your historical data.
The outcome will be that you will have an enormous amount of valuable data for your business that can be easily processed with the use of AI software.
If you want to know which AI tools for social media are the most suitable for your company and how to implement them, check our article on 50 AI Tools For Growth Hacking!
Furthermore, a customer service part where AI has already replaced humans is with Chatbots.
How does Chatbot’s in Social Media work?
In 1994, Michael Mauldin, creator of the first chatbot, Julia, coined the term “Chatterbot“.
A chatbot is a software that can conduct a conversation with humans.
The developer usually tends to make a great effort to make the chatbot appears as a normal human.
With the help of ML and NLG, they can find the most useful and right response for the customer questions.
Even though chatbot is still not able to pass the Turing test, some of them are really convincing.
Why Chatbots are the future?
Chatbots are a really powerful tool that can be useful for multiple reasons, for example:
- Customer service
- 24/7 presence on the website
- Consumer analysis
- Immediate response with multiple customers
- Gaining insights and consumer data
- Cost savings
And the response by customers is great, according to SalesForce chatbot statistics. If a customer has the option of filling a form or talking to a chatbot, only 14% will prefer the form.
More and more people are wondering and experimenting how far these bots can go, you can see some examples of how the Messenger platform from Facebook could be more interactive. The possibility of improvement is immense.
Social Media Companies are not alone in the run to AI.
On Google, for example:
- Google Vision API
Google software that can recognize different factors in a picture. When you give input an image what comes out is a code text.
In there you can find all the information about your picture. Such as, what is there and where is located in the picture through coordinates.
Even though it is not super accurate, it can also give you output, on a scale, of the emotions shown in the picture!
Furthermore, Google Vision is able to gather insights and to monitor your products on the internet.
Until now everything looks amazing but what are the downsides? And more specifically…
What are the risks of using a not fully operating AI?
Even if we listed some of the most interesting positive effects of Artificial Intelligence, as already mentioned, as per today, AI is not completely able to imitate human intelligence.
This leads to a huge missing piece of human behavior, AI cannot create emotional connections.
In 2016, the tech company Microsoft unveiled TAY a twitter bot as an experiment in, as the company described it, “conversational understanding”.
The bot is programmed to get smarter and learn through conversations with users.
Unexpectedly, Tay started to receive racist and sexists tweets. Consequently, Tay, being just a machine without any definition of good or bad, began to repeat the same feelings in its tweets.
This entire story happened in less than 24 hours, after which Microsoft decided to shut down the account.
Julia McCoy vs Articoolo
We already talked briefly about Articoolo, which also promise to “create unique textual content in a flash”.
So Julia McCoy proceeded with an initial input which was “content promotion” and after paying 1,90$ for the article she received the outcome. Which was
“Content promotion is a subject in the promotion universe relating to publishing important messaging to a specific audience white papers will be an example of this.”
The outcome did not make any sense. This is just to understand how long it is still the way for AI to take over a content creator job.
We believe that real images will still be a valuable part of the content, while the written part is being heavily taken over bu robots.
Nevertheless, when it comes to statistics and numbers, AI automatic content creation can be helpful and save a lot of time.
An example is Forbes, which uses Quill to write these kinds of reports. Which do not need any kind of empathy in their texts.
And this is what AI is still missing, emotional connection and empathy.
Conclusion – What awaits for the Future of AI in social media?
Regarding the question “Is AI taking over the social media companies?” the answer is no, at least not now.
Artificial intelligence is everywhere, it works silently in the background every time we open our computer or we interact with an App. It gathers data and learns from it.
It had a long way since the simple theory of reality. Yet as per today this technology still needs improvements.
As told during this article, AI can be a powerful tool in social media companies.
It can compute thousands of queries in less than a second, a task that would probably take ages for a human.
Nevertheless, users should be aware of how much data these companies are collecting on each individual.
Companies are getting data from everywhere and every individual user in real life will have an extremely deep impact on the content on which a person will engage and see in the future.