Artificial Intelligence vs. Humans: Who will take over?

Dec 22, 2016


Image recognition, diagnosing diseases, winning games: neural networks already surpass human brain in various areas.

Algorithms, intelligent software, and robots are now able to undertake more and more tasks that before could be carried out just by humans. Neural networks and Deep Learning make it possible.

Areas being conquered by AI:

1). Traffic signs recognition

Deep Learning networks are perfect for capturing image content. In fact, it is the right technology for autonomous cars as driverless vehicles have to correctly recognize traffic signs, trees, pedestrians or cyclists. Car manufacturers are working on the development of automatic scene analysis in real-time. Thus, cars will be able to recognize the intention of the pedestrian to cross the road, e.g. by analyzing his/her body posture.

2). Reading house numbers

The researchers at Google have used Deep Learning network to detect house numbers, no matter if they are turned or tilted, on millions of images made for Google Street View. It was done in order to precisely localize the houses in Google Maps. The computer managed the task just in a few seconds. In the future, this technology might be used for automatic detection of suspect persons in public areas.

3). Reading facial expression

Basic emotions like joy, astonishment, disgust or sadness always reflect the same way in our faces, doesn’t matter where we come from or how old we are. There are technologies based on Deep Learning, that are quicker and more accurate in face recognition than humans. Such systems can even distinguish a true smile from a fake one.

4). Winning games

Computers master a lot of games better than humans. As an example, the win of AlphaGo (a combination of Deep Learning network, analytical calculations, and random generator), a software developed by Google DeepMind, against human world champion Lee Sedol. Currently, researchers try to teach computers to play poker. It is highly difficult, as it is a game of imperfect knowledge. Anyway, there are a few advances in the field. Some programs have already learned to bluff.

5). Diagnosing diseases

The US researchers have already developed learning algorithms that can detect cancer cells by identifying suspect characteristics. It is impressing that algorithms are already able to detect molecules which can be suitable for new medicines.

Doctors and pharmaceutical companies expect that Deep Learning networks or other processes of AI will help to make an accurate diagnosis, find better treatments and create new active substances.

6). Maintaining machines

Big wind turbines have a lot of sensors, which daily produce hundreds of gigabytes of data. Neural networks learn to analyze this data flow better than a human. They recognize unusual vibrations and unsmooth run and arrange a maintenance team for reparations before it results in damages and the breakdown of a turbine.

However, neural networks are not all-rounders. They are truly perfect for pattern, image, text or audio data recognition, but they don’t have an everyday knowledge.

In order to build smart machines, having such common knowledge, Deep Learning process is not enough. It requires a combination of other methods for knowledge acquisition and processing, which is very challenging for the research.