IBM Watson neural network now knows what people are afraid of

Twentieth Century Fox Studios, together with IBM Research, the research and development division of IBM Research, ordered the IBW Watson cognitive system to make a trailer for the horror movie Morgan about an artificially improved person. The silicon brain itself identified the brightest fragments of the film and put them in a row.

Determine the scary moment of the film is not so simple. For this you need to clearly understand what a person is afraid of. Weak protein creatures with a sophisticated psyche have a serious vulnerability - emotions.

The most common human fears are fears of death, illness, pain. Such phobias are often used by creators of horror films, because they act on many viewers.
')
There are less common fears, for example, fear of loneliness, the army (men), darkness, clowns, butterflies, insects, porcelain dolls, losing their jobs, losing one earring (girls), etc. Experienced directors study the scientific press and are aware of many human phobias, including them also in the storyline.

Truly talented directors and writers in the “horror” genre even feel these fears themselves, that is, they themselves suffer from them. In this curse of the creator: to create a brilliant work, he must experience the depth of the most terrible suffering and other destructive emotions.

At the same time, horror films use specific techniques designed for a physiological reaction — unexpected sounds, abrupt changes of plans, unclear outlines of objects to stimulate anxiety, etc.

According to Dahl’s dictionary, fear is an internal state due to a threatening real or perceived disaster. That is, the supercomputer must, as it were, identify fragments of a film in which a disaster occurs or is intended, and then highlight the preceding fragments when the viewer has a feeling of this impending disaster. Those moments when hair stand on end and catch your breath.

The easiest way to do this is to remove the electroencephalogram of a person’s brain while watching a movie, also registering sweating and heart rate. But why use coarse methods if IBM Watson is at the disposal of cinematographers?

Using a supercomputer to analyze a horror film about a superman is an excellent opportunity to tell the general public about a new film. Moreover, many viewers will want to watch the trailer, which is generated by "artificial intelligence". Anyway, this is just an interesting scientific experiment. It is interesting to know what the computer considers the most terrible for us. Perhaps this information will be useful in the future.

The printout shows which parts of the Morgan movie the IBM Watson computer allocated for the trailer. These are the most emotional moments in which significant changes in mood are recorded.



During the preliminary training, the IBM Watson neural network trained 100 horror films on trailers, which were manually broken into fragments (“moments”). For each "moment", the system performed an analysis of several characteristics:

1. Video analysis : identifying people, objects and scenery. Each scene was assigned one of 24 emotions and tags from a list of 22,000 thematic tags, including “creepy”, “frightening”, “gentle”, etc.
2. Sound analysis: analysis of ambient sounds, including the intonations of human voices and background music, to understand the mood of the scene.
3. Frame composition analysis : lighting, scene. This is necessary to identify the specific characteristics of horror films and scary scenes.

The analysis was performed separately for each scene and in a combination of scenes with each other, using statistical methods. Thus, the supercomputer has learned to “understand” what types of scenes can fit into the structure of a horror movie.

During the first test on the full-length film “Morgan”, the system identified 10 points that are most suitable for the trailer. Of the 90-minute film, Watson chose the six most emotional minutes.



The developers had no idea which fragments the AI ​​would choose for the trailer. The study of his proposal showed that these moments correspond to the maximum peaks on the emotional characteristics of "tense expectation" and "tenderness." The computer decided that such moments would be most effective.

People also had a job. The computer indicated which points to use, but the editing and sound processing will have to be done manually. I also had to determine in which order to show these fragments, because IBM Watson did not build a logical chain. Nevertheless, his help turned out to be very helpful, because the selection and evaluation of suitable fragments is also a rather time-consuming task. Typically, the creation of a trailer requires 10-30 days of work of a group of experienced professionals, but with the help of Watson this period was reduced to 24 hours.

Interestingly, for the "Morgan" already released trailers, but the computer did not agree with the choice made earlier and included in its selection some points that were not included in the previous trailers.

Previous official trailer for "Morgan"


Having mastered a new area, the IBM Watson neural network can now quickly find material for a trailer for any horror movie. Naturally, he still needs the help of a person to bring the final product to an optimal state. Man and supercomputer - a great team to work together. This form of symbiosis is used in various fields of science and technology, including jurisprudence and medicine. Cinema is another area where AI has shown its abilities to help people.

The premiere of the film "Morgan" in the United States - September 1, 2016, in the Russian Federation - September 8, 2016. IMDb rating: 6.6 .