In fact, everything is simple.

The human ego prevents us from accepting the fact that we, in a certain sense, are quite primitively arranged.

We made our cars “in the image and likeness”, but they differ from us in their architecture. And radically.

How to add Intuition to a computer?

image

Has anyone ever wondered where this list came from?

Note that this has been working for many years and saves to humanity 600,000 man-hours per day (this is only Yandex).

But these are clues. Sequence of words.

What is a word? Phrase? Offer?

This is an image


A certain set of meanings, parameters, a multidimensional vector or tensor, speaking the language of mathematics and modern programming.

That is, in general form, “mom washed the frame” is a sequence of images. But there are nouns and a verb (although the action is the same image).
For a better understanding, there is a more verse appropriate to the context.
Forest, clearing, knoll, pit.

Smiled Did you feel? Now it is clear? Intuition (memory) suggested what will happen next? Did you see pictures before your eyes?

Yes, not from childhood. Nature!
Images can be very abstract. At the level of sensations.

I feel that it’s necessary, but I can’t explain why

This is our intuition that gives out a very complex image.

By the way... not always right, there are mistakes. The so-called "complexes" can mislead us, or they can save lives.

To comprehend is to match images


If it’s very primitive, then we will operate on a set of images that are usually divisible by modality for ease of perception:

  • image
  • audio
  • smell
  • taste
  • kinesthetics.

The following are more abstract combinations of these images developed by human culture and science. In general... technology.

All (almost) this can be described with words: semantics, meanings, relationships. In principle, with mathematical precision.

For description, we use text, numbers, formulas, schemes. And as part of programming, it all comes down to a sequence of numbers - vectors or classes.

To reduce the amount of information processed, coding is used. A unique number (ID) is given for each image.

Encoder


1 - Forest
2 - clearing
3 - mound
4 - fossa

As a result, there will be a sequence of numbers: 1, 2, 3, 4. When there are more images, there will be more numbers, and the sequences will be different.

We also peeped at nature. There is an area in our brain that performs this function - the hippocampus.

If we submit such sequences (patterns) to our e-brain, then we get statistics for each transition from image to image.

Complex one-dimensional vector can also be encoded with one number:
{temperature: 20, humidity: 60, wind: 14, pressure: 120/60}
{after the rain, Thursday} {after the hail, Tuesday}
{dirty, fearless, tank}
{warm, light, do not bite flies}
{I, in the sun, lie with my ears moving}.

All these images reflect the state of a particular system in the Universe, visible by our senses or IoT sensors. This process is ongoing - we isolate certain images from the stream of events.

And everyone sees in his own way, according to his system of values, sets of situations in which the individual participated, and his internal structure.

Therefore, the classification of objects is the topic of more than one doctoral dissertation, although it is well studied and described algorithmically.

Algorithm


Let's say we heard:
“Mom washed the frame” - 100 times
“Mother washed Mila with soap” - 20 times
“Mommy Soap Calculator” - 5 times
“Mom washed the store” - 1 time.
A total of 100 + 20 + 5 + 1=126 phrases.

The result is this picture:

image

And if we hear “soap mom,” an option with a higher rating/frequency will come to mind. In this case, the "frame".

Other options also pop up in memory, but since their repeatability was lower, they are lower in the list for issuing.

The simplest implementation of a similar algorithm for experimenting on my github.

In search engines, “everything is complicated” - giant recurring and even convolutional neural networks.
Networks are formed as a result of processing a huge amount of information that we, today, generate in abundance in electronic form.

As you already understood, intuition in a car is not so difficult.

It is much more difficult to build an encoder to feed numerical patterns into an “intuition block”. To do this, you need to use a system of pattern recognition and classification. Visual, sound or even logs of the company website.

Get to the point


What patterns can be?
Business process: receiving goods, checking the configuration, entering the invoice in 1C, transferring to the warehouse.
Mechanics: bolt, washer, grover, nut.
Driving a car: clutch, transmission, gas.
Time of day: morning, day, evening, night.
Season: winter, spring, summer, autumn.
Sales: fishing rod, set of baits, tent, jeep.

As you see, everything is predictable when you see everything many times. After the summer, winter will not start. Unlikely.

Recognition and prediction alone are not enough.
What is important is what we can do with this information.
And if we can.

Everything is very simple


Therefore, it is difficult to understand and forgive to apply.

My intuition tells me indignant exclamations like:
- Yah! Even all the servers on the planet will not replace one human brain!
- Nonsense! A car cannot have intuition!

I will dispel your fears and doubts:
Intuition is not Enlightenment.
.

Source