Why is it so difficult to choose which movie to watch (and neural networks will not solve this problem)

Do you know the situation: they decided to spend the evening at home and watch some movie in good company, but, trying to decide which one, they spent so much time choosing that they weren’t left for the movie — or the desire disappeared — or, anyway, they started something to watch, but the mood was not right?

Most people attribute this problem to their lack of awareness about the world of cinema, and try to solve it using different selections and ratings, or asking for advice - and businesses, in turn, try to do the same by offering users selections and ratings, or by developing recommendation system. Nevertheless, the problem does not want to go anywhere - and the development of recommendation systems only repainted it in different tones: now users, instead of asking for advice from friends and strangers on the Internet, endlessly flipping through the ranks of bright posters on Netflix (the problem is global) or some ivi. Businesses, meanwhile, in the absence of better ideas, continue to try to push the cube into the keyhole, hoping that they will nevertheless be able to make a recommendation system that will learn to guess what the user wants, who does not know what he wants; True, the developers’s hopes for a collective mind were no longer justified: it didn’t help to call for help from other users - feedback catalogs and question-answer services only reduce pain without relieving it - so now all bets on the mind are artificial - that's for sure AI must see through this nut!

Do not get it. Firstly, neural networks are not AI. Marketers of companies involved in the development of neural networks and searching for clues to the secret of creating AI blurred this difference, calling one another, but it did not go away from this.

Secondly, they just solve the wrong problem. Each time a user accesses the Internet with the desire to “see something” or “read something”, recommender systems obediently rush to execute it, offering a choice of countless options; whereas the real question that a person needs to answer is not “what do I want to look at?”, but “what do I want?”

Why do I claim that recommender systems do not work

In a way, they certainly work great, helping to increase the time people spend reading Yandex Zen or watching Netflix. But the problem of choice is not its insufficiency - on the contrary, when there is no choice, then there is no problem - but in its presence. When people are looking for advice or recommendations, they are not looking for “here are a hundred excellent options for you to choose”, but for getting rid of this very choice: “this is what you need.” How to learn how to give exactly such advice - something that everyone who has ever had to sell, recommend or advertise is puzzling over doesn’t matter, his or her own. And progress, of course, is, but quantitative - the effectiveness of advertising, advice and recommendations is growing - but a qualitative transition from the situation “here we can offer you” to “here you need” has not happened in the current direction - not expected.

The fact is that the progress of human civilization is very uneven - and, having brilliantly learned to understand how to change the surrounding reality for their needs, people still continue to remain, for the most part, ignorant of what these needs really are. And when the tool to perform the task is chosen incorrectly, its advancement and technological perfection no longer matter.

I will explain by example with the choice of cinema.

The “what to see” problem has been familiar to me for a long time and painfully - and therefore I can immediately dismiss the option with a lack of user knowledge about new movies: movies with experience , over the years I have put together in my head a pretty solid catalog of not only watched, but also unreviewed films - including the “be sure to see!” Saved under the flag - that I didn’t even need the Internet to pick up a bunch of options in different genres on the go - and yet, the inability to decide what I want to see was my rule molecular golovnyakov.

That is, the problem is not in the assortment of options - and a recommendation system cannot solve it. The neural network can be dragged into the incredible accuracy of guessing the taste in cinema, no doubt, but it will become intellect when it learns not to guess the user's wishes, but to clarify first whether the user himself correctly understands what he wants.

In most cases, a successful movie joke does not begin with the desire to “see something,” but with the desire to see something specific.

If the desire to watch a movie is not accompanied in the head by the specific name of the film, then you do not need to obey it, like a slave to a lamp, going dutifully to sort through movie catalogs;
First, let's call the problem:

  1. intention must have a goal;
  2. “Anything” is not a goal;
  3. the intention to do “something” is an aimless intention.

An aimless intention leads to futile decisions: the problem of “seeing something” cannot be solved by searching somewhere .

Intention in itself is also only a tool: a tool to satisfy a need or need.

It is need / need that sets the task, and intention is only a proposal, the notorious recommendation on how to fulfill it. There are no unsolvable problems; there are sometimes incorrectly posed problems. And an obscure intention signals an obscure need.

Intention is just a route, but if the goal, endpoint, is unknown, then you need to turn to the starting point: what triggered this movement? The solution to the problem of vague desire is only one: to step back a step and deal with what kind of need this desire is dictated.

“I don’t understand what I want” is an incorrect statement of the problem, it will be true “I don’t understand why I want to.”

The desire to see something does not mean the desire to see something or in general the desire to watch something in principle. This is just a recommendation - a recommendation of an internal recommendation system built into every person. This system is called the unconscious, and works just like any recommendatory neural network, because it is a neural network - the original neural network, the prototype of all neural networks.

What happens when a signal of need is heard in the field of the unconscious?

  1. Having sensed the need signal, the unconscious goes into memory and begins to look for a suitable solution - that is, it looks at what actions last time satisfied the need, which is as close as possible to the received signal in signature.
  2. Having found the most suitable option from the existing experience , the unconscious transmits a signal to the prefrontal cortex, the prefrontal cortex of the brain aka decision-making area: the desire <ref> came in, maybe the solution “to watch a movie” will suit you.

In short, neural networks will not help us, because our subconscious is also a neural network. In short, here it is time to connect higher nervous activity and deal with desire as a signal of need, and we need to find out not what to do, but where the signal comes from.

The difference between a neural network, artificial or biological, and intelligence, artificial or biological, is the difference between a recommendation system and a decision system. The unconscious of a person is the same recommendation system, and any recommendation is always a game in association:

  1. judging by the problem you described
  2. based on my previous experience
  3. I can recommend you the following solution

Of course, exact hits also happen - for example, if the unconscious began to solve the 3 + 3 problem:

  1. hmm, we don’t have previous experience in solving the 3 + 3 problem, but in our library of patterns successful cases of solving the 2 + 2 and 3 * 3 problems are stored;
  2. 3 * 3 differs from 3 + 3 by only one character, and 2 + 2 - by two characters out of three;
  3. Recommended answer: 9.

Moreover, a similar method for solving problems is quite capable of producing an effective result: for example, if the task were 2 + 2, and there were examples of successful solution of tasks 3 + 3 and 2 * 2 in memory, then the unconscious (and neural network) would recommend answer 4. And the richer the experience (the library of images) - the greater will be the probability of a good choice, but in the associative paradigm the right decision will always remain a matter from the category of probabilities.

So what about choosing which movie to watch?

I will tell you by the example of such a phenomenon as stress eating:

  1. the unconscious receives a stress signal;
  2. begins to rummage through the expanses of memory, until it stumbles upon memories of peace and goodness in the process of eating a bath of ice cream;
  3. recommends eating something to calm down.

In case of success - that is, a person obeyed, ate sweets and, indeed, felt better for him - the stress-food pattern will be fixed. The more repetitions, the stronger the fixation: one-time fluctuations that do not fit into the pattern, recommender systems pessimize. So, if one day the ice cream tray does not work, but there are still more examples when it was possible to calm the stress signal in this way, then when it is repeated it will still be the first in the list of solutions recommended by the unconscious.

So: the solution of the "what to see"

  1. the desire to “see something” is an impulse caused by an unrecognized need;
  2. the task that needs to be solved is not what to see, but what kind of need causes the urge to “see something.

The real signal of biological need, both physical and psychological, is not intention, but the emotions attached to it. Intention is just rationalization; need is hidden behind emotions.

Therefore, to solve it, it is necessary to reflect which particular emotions are anticipated in connection with a specific intention. In other words, the method of working with unclear intentions implies movement not towards the search for the next step, but reverse engineering to the source of this intention - needs. In the process of analysis, you should move in small steps, because the most complex nervous activity is divided, as a result, into the simplest binary operations: the synapse in the neural system is either lit up or not.

Here's what the (simplified) reverse engineering of stress eating looks like:

  1. the intention is to eat something;
  2. the emotion associated with it - peace and goodness;
  3. need for rest is caused by stress;
  4. the solution is to deal with stress, and not choose what to eat.

Those who wish can do the same reverse engineering of the desire to “see something in the evening” on their own.

PS Examples “And I found a movie in the same situation and watched normally” are not relevant objections, because there are too many unknowns - the very problem of searches could, in the end, drown out the desired signal - like bread drowns out the feeling of hunger, but in both cases the desired need remains unsatisfied. But if the movie looked half-eye, without interest, intermittently, etc. - it means that it was definitely not a desire to watch a movie.

I read the comments, but I will respond to them in the following publications. If you want to supplement or correct me in something - write to me in PM or by mail. If you have your own thoughts on this subject - do not exchange for comments, it is better to write a response post. If you want to express support or wish the following articles appear as soon as possible, the best way to do this is to use the "Support the author" function below (Yandex money) or through Qiwi (Qiwi, card, cash).

Source: https://habr.com/ru/post/462615/

All Articles