Robots are advancing: what is artificial intelligence capable of?

Every day, machine learning technologies are penetrating deeper into different areas of our life and mastering a new arsenal of unexpected abilities. In this article, we will tell you what neural networks are already able to do and whether they can replace a person.

To drive a car

The cars of the future will learn to understand their position on the road, see their surroundings and even communicate with other vehicles while their owners relax in the back seat. Volvo has already shown the world a completely autonomous concept car: without steering wheel, pedals and driver. It has an electric propulsion system that drives all four wheels. There are sound signals and special lighting to alert other road users. And the salon can be used as an office, recreation room or cinema.

Ford keeps up with the Swedish automaker and has developed the unique All-wheel Drive Disconnect technology: its electronic "brain" receives information from a dozen sensors, makes calculations and makes a decision in just a few milliseconds.

Paint pictures and compose poetry

This fall, an art auction will take place in New York, which will sell a portrait of a fictional member of the count's family, created by artificial intelligence (AI) based on a dataset of 15,000 real paintings from the 14th to 20th centuries. And this is not the first robotic work of art. So, for example, Yandex invented a "neuropoet" who can write in the style of any chosen author. And Japanese technology, capable of coming up with a fictional story, generally made it to the finals of the famous literary competition. “Everything that can be imagined is real” - it is not without reason that Pablo Picasso bequeathed it.

Lead everyday life

Today artificial intelligence has found its application in the system of "smart houses", where in the morning they will automatically make coffee for you, turn off the iron and even steam your clothes. For example , the Hexa robot spider will take care of your plants, the virtual assistant Einstein will do the lessons with the children, and the multifunctional AEOLUS robot will be able to control all smart devices in the apartment, bring food, remove trash, sweep, mop the floors, and, most importantly, communicate with family members and build knowledge about your home.

Work with text

Yes, it is very likely that in the near future, translators and writers will suffer the fate of the dinosaurs. Already, the Narrative Science agency has a Quill software platform that converts a lot of data into coherent text. The system generates news stories almost every 30 seconds, collecting the most interesting facts and conclusions.

Understand emotions

Imagine: the program receives an image of your face and determines your emotional state by the shape of your lips, eyebrows and facial expressions. For this, a special project Feelix Growing was launched, the goal of which is to teach computers to respond to human emotions. And this is perhaps the main step towards a dialogue between robots and humans.

Today, the creation of a system close to human emotional intelligence is a huge task for the world's IT giants. Recently, Microsoft has taught neural networks to distinguish the emotional coloring of text, and Google intends to incorporate AI-based face and emotion recognition into Raspberry Pi smart boards for robotics. And further - more ...

Predict the future

How do you like this turn of events? The system developed at MIT is truly foresight. The technology receives an initial image, on the basis of which deep learning algorithms create a video with possible scenes of the near future. For example, AI sees a picture of a railway station and, as it were, animates it: there are passengers, a train is leaving, it is raining, etc. In the future, such a system can help predict accidents, but for this it still needs the development of intuition.

Where are we going?

According to Gödel's theorem, any closed system is not complete, which tells us about the impossibility of the existence of AI without a person. However, according to researchers-futurists, the theory of consciousness will soon allow robots to acquire human thinking. How can this happen?

English mathematician Alan Turing came up with a test that tests whether machines can think. If the computer manages to deceive 30% of live interlocutors, the test is considered passed. In 2014, the Eugene bot recreated the intelligence of a thirteen-year-old teenager and, in the process of communication, convinced experts that he was a real person.

By about 4 years of age, the child begins to become aware of himself and to compare with others, learns to understand their feelings and predict the behavior of other people. It is this ability that still fundamentally distinguishes a living person from a machine.

Recently, OpenAI, of which Elon Musk is one of the founders, conducted an experiment: digital players competed with real ones in the computer game Dota 2. First of all, this is a team game, and to win it you need to have intuition and be able to predict what to expect from an opponent. As it is obvious, people won the tournament. But what happens if the computer does develop strategic thinking?

To this end, the English professor of robotics Alan Winfield proposes to program machines with internal models of themselves and each other, and then teach them to design various situations of interaction with their environment.

DeepMind's artificial intelligence ToMnet recently passed a series of tests and showed the ability to read the minds of other AIs and predict their actions. Thus, it is likely that soon they will be able to finally understand people, and therefore think like people.