We live in a funny time. Technological progress is slowly but inexorably squeezing reality into the world of science fiction films. The world in which robots live among us and perform a huge pile of tasks.
The latest trend of artificial intelligence is neural networks. They can learn, memorize and process information, and then perform some actions based on their experience. Just like people. Almost.
Google, Microsoft, Yandex, as well as hundreds of institutes and startups around the world are engaged in research and development of neural networks. Something cool appears almost every month.
To begin with, let's see what robots have learned to do by the end of 2016, and what benefits we, internet marketers, can get from this.See also: New reality in search: JSON-LD, Mobile-First, "Palekh"
- Neural networks recognize images
- Neural networks understand speech
- Neural networks translate text
- Neural networks are talking
- Neural networks rate images
- Neural networks recognize emotions
- Neural networks determine whether you are drunk
- Neural networks draw pictures
- Neural networks make images
- Neural networks make music
- Neural networks create encryption protocols
- Neural networks tweet
- Neural networks write text
- Neural networks communicate
- Prospects for the near future
- Rewriting and copywriting
- Illustration and design
- Customer support and social networks
Neural networks are looking for information
In early November, Yandex announced a new algorithm "Palekh" based on neural networks. It allows you to more accurately determine what the user wants to find when entering a non-standard query.
Tell me, dear Yandex, a movie in which potatoes were grown on another planet. Yes, I exaggerate. But there are similar queries, and they need to be somehow processed.
Palekh indexes the headings of all pages and tries to understand their meaning. He then writes the meaning as numbers and stores them in a multidimensional matrix. When you enter a query, the algorithm also converts it into a numeric format and searches for the most appropriate header in the database. This allows you to make the search more useful and responsive.
The next goal is to capture the meaning of the whole text, not just the title.
Such things change the approach to writing texts. We need to think less about keys and less about the meaning, usefulness and, probably, interestingness of the content.
Neural networks recognize images
By the way, neural networks have been working in Google and Yandex for a long time already - in search by pictures. The robot determines what is in the image and saves this information. When you enter a request, the network compares it with its notes and gives the most appropriate pictures and photos.
The same algorithm uses the Clarifai service. You specify a photo or picture, and the neural network determines what is depicted on it and creates tags. Even the color gamut shows.
Google Photo has a similar thing. You upload your photos for storage, and the service analyzes them and helps you find the ones you need. For example, you can view pictures taken in the mountains, or those on which your friend Maxim is.
Another interesting service is FindFace. You upload a photo of a person, and the neural network finds his profile Vkontakte. Expansion for maniacs!
Profit: Search for contacts in Central Asia from photographs from events to show them current advertising. Targetologists cheer.See also: How to create visual content using the convenient and cheap tool Canva
Neural networks understand speech
Okay, Google, do you understand what I'm saying? Remind me tomorrow to finish this article. My smartphone understands me well. Not perfect, but good. Clearly better than sleepy saleswomen in stores.
Do not forget about voice input. English-speaking Google Docs writes well from dictation. He also begins to understand when "Delete" is a command, and when it is just a word that must be entered. Russian localization is lagging behind in development by several years. As always.
And here is an article about the LipNet neural network that reads lips. Under laboratory conditions, speech recognition accuracy was 93.4%. This is 93.4% more than mine. It is clear that in real life it is more difficult to grasp the meaning, but LipNet will develop further.
The Tribe video messenger not only recognizes your speech, but also understands it. It can automatically create and enable subtitles or find information about the subject of the conversation. The interlocutor mentioned that he was reading “Texterra”, and the Tribe will immediately find and show you the link. Cool, right?
Profit: automatic creation of normal subtitles for video and transcription of interviews.
Neural networks translate text
Well, yes, machine translation is no surprise. He appeared long ago. True, he still has not managed to become adequate. But Google created a new NMTS algorithm based on neural networks and connected it to its Google Translate,
So far the NMTS works with only two pairs of languages: English-Spanish and English-Chinese. The results still do not reach the level of professional live translators, but they are much closer to them.
Profit: translation of foreign content, adequate communication with colleagues and clients from other countries.See also: Deep Learning and the future of search engine marketing
Neural networks are talking
In September, Google revealed the new WaveNet neural network, which simulates the human voice. Well models. Call it on the phone, I would not have guessed that this is not a man. Listen to examples on the site.
WaveNet can speak English and Chinese, and also models various female and male voices.
Profit: Call center automation, video dubbing.
Neural networks rate images
The MemNet algorithm, created in the Computer Science and Artificial Intelligence Laboratory at the Massachusetts Institute of Technology, determines the index of image memorization.
In fact, the robot has learned to anticipate the work of human memory. He indicates which photos people will remember for a long time, and which ones will be forgotten in 5 minutes. It also highlights the most memorable areas in the photo.
Profit: selection of memorable images for advertising, social networks and blog articles.
Neural networks recognize emotions
Service from Microsoft is able to recognize the emotions of people depicted in the photo. You upload a photo, and the neural network finds faces on it and analyzes what emotions they express. The list has:
- Anger (anger);
- Joy (happiness);
- Neutrality (neutral);
- Sadness (sadness);
A similar thing was done at the Massachusetts Institute of Technology (MIT). Service Affectiva recognizes the emotions of people on video broadcasts on the webcam.
At the time of this writing, the site of the service lay, so I leave a link to Wikipedia. Just in case.
And neural networks are able to recognize sarcasm in messages. The robot analyzes the text of the post, the image, the previous posts of the user and determines whether it really is "Wow, cool!" or still sarcasm.
Profit: search for negative comments and reviews on the Internet. The study of user reactions to different content and advertising.
Neural networks determine whether you are drunk
Do you like to write tweets after going to the bar? The robot will find you! In Rochester Research University created a neural network that recognizes posts that people wrote while intoxicated.
So far, the network can only work with Twitter.
Neural networks draw pictures
So we got to the most interesting - to create content. In the summer, neural networks boomed on the Internet - they began to process images. You show the robot a photo and some picture. And the robot redraws the photo in the style of the picture.
In some services, the style can be selected only from a given list. In others, download your own and provide complete freedom of machine creativity.
Look at the results of my collaboration with the neural network. Ordinary boring photos turn into something incredibly funny.
The author of the article, his wife, their cat and several places where they visited
That is, robots are already able to create content. No not like this. They are able to create cool content. I could not draw this.
If you want to play with the neural network, here are some services to choose from:
- Deepart.io - cool standard styles and you can upload your own. I even made my portrait of buckwheat. Minus - the neural network draws pictures for 10-20 minutes.
- Deepart Turbo - a faster version of the previous service. Draws fast and cool, does not put watermark. But you can’t upload your styles - only 39 standard ones. Also available in the form of Android and iOS applications.
- Ostagram - allows you to upload your styles and adjust the intensity of their use. But the quality of paintings is worse than that of Deepart.
- Vinci - you can download the application for any mobile platform or process photos using a Vkontakte and Telegram bot. Only 25 built-in styles, you can not upload. Compared with the previous ones, it is rather boring, but it works several times faster.
- Prisma is a Russian-language application for Android and iOS. It works quickly, but does not allow loading its filters.
Here is another cool neural network - Neutral Doodle. She draws on your exizam and outline. They glowed something in the Point, indicated the style in which you want to get a picture - the robot drew. There are no web services yet, but you can download the code from Github and compile the neural network on your computer.
Profit: creating interesting unique content. For example, you can process all images for an article in the same style.Read also: How to express ideas and make publications attractive with simple drawings
Neural networks make images
The British company Magic Pony Technology has created a neural network that improves the image: it increases the image resolution, sharpens the video and draws textures.
Neural networks make music
WaveNet, about which I spoke above, knows how to write music. Examples are in the same article. Ringtones, honestly, so-so. As if the cat on the piano trampled. But not for these purposes the neural network was created.
But the Yandex robot, inspired by the works of Yegor Letov, wrote a musical album. True, he came up with only lyrics to songs. Music made its creators. They played and sang.
Seriously, I would never have thought that it was not Letov who wrote this. I even listen to these songs sometimes. Another case where artificial intelligence made cool content. We are waiting for the neural network to make friends with WaveNet and will be able to independently perform songs.
If you are not an ardent fan of Letov and do not know all his songs by heart, try to pass the test. Try to guess where the work of the musician, and where - the neural network.
Neural networks create encryption protocols
In addition to works of art neural networks create more serious things. For example, new ways to encrypt data. The network developed by Google Brain has come up with its own SSL protocol.
The essence of the experiment is quite entertaining. Researchers have created 3 neural networks. Two of them had to learn to communicate with each other so that the third could not understand them. When the third one managed to decipher something, they quickly changed the algorithm. The result was a protocol by which robots transmitted messages. And only they could understand each other.
Neural networks tweet
Music, pictures, encryption protocols. This everyone can! What about tweets? This is where real brains are needed.
Microsoft created a self-learning Tau bot, which (yes, this bot-girl) had its own Twitter feed. In theory, Tau had to communicate with people, learn something new and form as a person.
It was formed - grew up in an ardent misanthrope and neo-Nazi. Bot admired Hitler, despised blacks and Jews, and also wrote about her hatred of people. Because of this, the channel had to close.
The other Char-rnn neural network tweets, parodying US President Donald Trump. The creators of the Deepdrumpf project trained the network on policy speeches.
In Russia, there is also a similar project. Only instead of Trapma we have Ramzan Kadyrov. The network trains on politician tweets and his Instagram posts. The result (rather dubious) can be seen here.See also: Twitter Cards - how to go beyond 140 characters and get more clicks
Neural networks write text
Copywriters, admit, you were waiting for this item. When do you think robots will become our competitors?
The Russian company Meanotek is developing neural networks that write unique texts for websites. In Russian. But the online store, the description of the goods which made the robot. Of course, Ilyakhov and Kaplunov are far away, but this is only the beginning.
And at the London Film Festival, the neural network wrote the script for a short film. And they took it off. Solid art house turned out. Nothing is clear, but cool.
Profit: even no need to explain. Robots themselves write seo-content for sites. A great alternative to stock rewriters.
Neural networks communicate
For me, this is the most interesting part. Pictures and descriptions of goods are, of course, entertaining. But full communication - that's what's really cool.
More recently, chatbots were stupid as logs. Yes, they learned something, but they still talked too clumsy. Neural networks have corrected this business.
In the Google Inbox application there is a Smart Reply function - smart answers. The robot reads the message you received and offers 3 short answer choices. You choose the right one and send it. Nothing to write. Such semiautomatic.
But I found examples and more abruptly.
Digitalgenius service allows you to automate customer support in messengers, social networks and email. First, the bot monitors the managers and learns. Then he tries to answer himself. Employees can adjust the answers, and the neural network will remember the edits and will take them into account next time.
As a result, you can replace the technical support department and sales managers with one robot. So far, of course not. But you can soon. By the way, the service is already used by BMW and Panasonic. Perhaps choosing a camera, you have already communicated with chatbotom.
The Luka company created the Replica neural network, which makes a “digital copy” of a person, communicating with him and studying his messages. The developers promise that in the future, Replica will be able to search for information itself, fully communicate with people and even become a coach.
So far, the service is unavailable. But you can already register and reserve a name for the bot. As soon as the application appears - start growing your copy.
A similar service - Ghostbot - was created in order to otshivat bothersome partners. Exactly. First you teach the bot to communicate like you. And then he talks to your wife while you work.
Okay, just kidding. In fact, he is chatting with your admirers and ex, whom you feel sorry to send. Seriously, the bot was created just for this. Read on the site.
Profit: just great! Automate customer support, replies to comments in social networks, communication with customers, quick replies to emails and dozens more ways to make your life easier.
The list is impressive, isn't it? Notice that I chose only those things that at least overlap with Internet marketing. But robots can also play in Lego. And drive cars. And troublemakers with a stun gun to beat. And cats on the lawns scare.
Prospects for the near future
I warn you right away, now there will be my unscientific forecasts. I do not guarantee anything. Just guessing what technologies are coming soon, and how they will change the life of internet marketers.
Rewriting and copywriting
Rewriting will die. I think it will not take 10 years, as rewriters and seo-copywriters will not be needed. Robots will write cheap content.
Indicated the topic and the number of characters - the neural network in 2 minutes wrote the text, which is guaranteed to be received by search engines. And thematic pictures will select why not. In principle, even on the site will be able to place. Pointed text - the robot otreraytil. Beauty!
Instead of going to the copywriting exchanges, webmasters will buy a one-year SkyNet subscription and stamp sites with hundreds.
The same is in social networks. Instead of admins who copy-paste each other button accordions and spread them into their communities of 50 pieces a day, there will be robots. I connected the service - he will find thematic content himself, remove the watermarks himself, otrayrightit, post it himself.
Can you imagine what kind of content crap flows to users? Marketers will have to try to get people interested. For example, it is very cool and interesting to write. And do not post accordions, which will be a hundred times more.
Along with the rewriting, such a profession as a proofreader will go to oblivion. If robots write correctly, they will be able to correct mistakes. And embed this feature in some Google Docs. Everyone will be literate.
Illustration and design
In my opinion, neural networks very well show themselves in working with images. You give a picture - you get a cool picture. And these are still raw algorithms, they don't even have a year.
And now let's connect the imagination. Robots can draw a portrait or landscape in the style of the selected picture. They can draw a picture of a simple sketch. And what prevents them from starting to draw designs for banners and landings?
We make a prototype, we find a cool landing and we feed both images of a neural network. She takes the design of the landing and clings it to our prototype. Well, the color scheme, you can choose why not. Everything, the model is ready, it is possible to impose.
Although stop. What for? The neural network takes a layout and imposes a page on it. Now everything is exactly, you can pour traffic.
Sounds like science fiction? А вот я взял и попробовал с помощью нейросети сделать иллюстрацию к промопосту Вконтакте для этой статьи. Нарисовал в Иллюстраторе прототип и отдал его сервису Deepart.io.
Вот что из этого вышло.
Первый в мире дизайн от нейросети
Некоторые баннеры разнесло до неузнаваемости. Другие остались вполне читабельными. The service, which was not intended to create a design, made the design of the Vkontakte promotional post. Yes, poor. Yes, the curve. But did.
I am sure that in the next 5 years we will see the service described by me.
Another cool thing that may appear is a service for evaluating designs and illustrations. Found a picture - the neural network said how good it is. Is it suitable for the topic of the post, what kind of reaction people will have, how to edit it so that it is cool.
Oh yeah, neural networks can look for pictures themselves. Then it's still easier. You wrote a post - the robot found illustrations for it, adjusted them to the same style, removed the watermarks. And he offered several options. Like, take it if you want to make people laugh and collect a lot of comments. And your boss will like this - he will give a bonus.
By the way, almost all the tools described in my example already exist. But separately. Artificial intelligence is already able:
- Search for images;
- Remove watermarks and labels;
- Increase resolution;
- Edit, change the style and color range;
- Anticipate the reaction of people;
- Evaluate memorability.
It remains only to bring them together and teach the robot to understand the meaning of the text.
And some more predictions. Remember the neural network that draws pictures from sketches? So, sooner or later there will be an application that will do this.
You write an article - and you desperately need a picture where a robot Bender walks along the beach in an embrace with Monica Bellucci. Explain the task of the neural network - download the illustration. Given the speed of development of computing power of computers, after 5 years you will receive FullHD-images with awesome detail.
Customer support and social networks
There is no need to doubt - there will be automation of customer support. First in English, then in Russian will catch up. There is already a Russian project iPavlov, which they promise to launch by 2019.See also: How to resolve conflicts with customers: instructions for technical support
First, robots will solve the problems of users around the clock. And in chat rooms, and by mail, and by phone. Telling how this monitor is better is to explain how to use a VR helmet and call a repair team when the Internet suddenly disappeared. Orders to take, again.
"We will answer you within 24 seconds" will cease to be something cool. Neural networks will respond instantly.
Secondly, robots will replace admins in social networks. Respond to comments, look for brand references on the Internet, advise doubters. Well, post content at the right time, of course. And if the bot fails to work out the subscriber’s negative, he will call for the help of the SMM specialist - and he will tell how to behave in such cases.
The world will inevitably change. Some professions will become extinct, because robots will do the same thing faster, cheaper and better. Others will get cool tools to solve their problems and improve work efficiency. Yes, and the tasks themselves will be different. After all, neural networks will be used not only by marketers, but also by everyone else.
PS: If you are interested in this topic, I advise you to read the book Martin Martin Ford "Robots come."
Are you ready for a change? Write in the comments what you think about this.