Original article was published by Roman Romadin on Artificial Intelligence on Medium
Machine Learning Tools For Generation Content: Open AI, DeepAI, D-ID
Machine Learning and Computer Vision have come into our lives. No doubt they open up new opportunities. In particular, the underground security service involves Machine Learning for detecting and searching wanted people. Moreover, Computer Vision applies in the controlling system of the self-driving cars so as to reach the destination and avoid accidents.
Besides obvious advantages these technologies have drawbacks. In fact, the results of Machine Learning work might be used by outside parties to extract personal data and create misinformation and fake news for unfair purposes.
“Only 7% of Gen Z college students found social media to be the most trustworthy news source. And 50% of Gen Z students said they believe online newspapers and news sites to be the most trustworthy.”
A brief overview of Artificial Intelligence (Machine Learning) products
The results of Artificial Intelligence work (Machine Learning) can be well-renowned generated products: photo, video, music, text.
A generation photo
This service thispersondoesnotexist provides pre-generated photos of a person, which has never existed. The situation is more complicated with the generation of the cat’s photos. You are welcome to explore the cat’s photo generation result online by link.
A generation video
All we know about a high-quality video of fake Obama created using Artificial Intelligence tools to make phoney speeches. Synthesized former US President Barack Obama may say any words putting in his mouth. The technique of generating video was offered by researchers at the University of Washington and produces really photorealistic results.
A generation text
The neural network GPT-2/GPT-3 by OpenAI was able to generate a logic text by your own description:
DeepAI has Text Generation API Documentation for your programming language: bash/nodejs/python/ruby/c#.
You can find other interesting API and models by link. Some of them:
A generation music
Quite a fun tool Jukebox is a neural net that generates music from scratch.
The second neural network of music generation is MuseNet by OpenAI . It can generate 4-minute musical compositions with 10 different instruments and knows many different styles and can blend them in.
There is a lot of other interesting projects by OpenAI.
Further development and use cases
The brand-new instruments come via Artificial Intelligence:
- searching objects and patterns of behavior in the picture, photo, video or text — “search by itself”
- misrepresentation and full-deleting objects — “cloaking”
- generation and searching the synthetic objects — “creating and searching fakes”
No wonder, Artificial Intelligence (Machine Learning) is a double-edged sword. On the one hand, new information appears using these tools but on the other hand, they verify content. In short, I have noticed a confrontation between creating and further detection. For instance, the generation of fake news and seeking them among real ones or searching people in the photo(video) and protection of face recognition.
As regards to people searching in photos, the first thing that comes to my mind is service FindFace. It has known from 2016. It currently works only in b2b format. Though the market is flooded with similar companies, bots, services with permitted for personal use.
To protect against face recognition you may use:
Summing it up Artificial Intelligence including Machine Learning — modern tools which bring us new opportunities, new content, and new knowledge to our lives. As said knowledge is power even if they are obtained through synthesized objects.
In addition, fake news can be used not only to manage and control people but also to reveal meaningful opinions and attitudes prior to real events. For example in order to find out any attitudes before adopting the law. Photo “cloaking” may help to compromise between publicity and privacy and protection of personal data.