How AI solves problems: unusual cases

Researchers of artificial intelligence published an interesting selection. In it, they reflected how the artificial algorithms solve the tasks assigned to them. Often AI use non-traditional and very extravagant methods.

---

The era of artificial intelligence has already come. Every year, programs invent ever more sophisticated ways of accomplishing the tasks that demanding scientists put before them. Machine logic is often different from human: where people assume one way to solve a problem, AI invents a few new ones.

The article, recently published in ArXiv, highlights several such incredible and slightly frightening examples of thinking in the digital mind. Some AIs were designed to simulate the process of "evolutionary" learning on trial and error. While others fought each other in speed and ingenuity, honing skills and improving results with every attempt.

Each of the 27 cases presented in the article is an almost anecdotal story. I will not quote them all, but I will give some of the most striking and interesting examples:

The study, whose goal was to simulate various systems of movement, gave rise to real "hackers" who violated the rules of simulation and, burying themselves in the junctions between the collisions of models, broke beyond the boundaries of the 3D scene to quickly reach the final goal.

AI, who played five games in "Tik Tak Toi" (similar to "tic-tac-toe" game) against each other, came up with a very effective winning strategy: using extremely long queries during addressing, they simply overloaded the opponent's system, after which he was automatically counted defeat.

The program, designed to simulate effective methods of braking an aircraft when landing on an aircraft carrier, has experimentally found out that if a certain value of the variable denoting the force of resistance is exceeded, it becomes zero. The result was a catastrophic situation, which was also a technically perfect solution to the task at hand.

The robot, imitating a living being, received a simple task: to pass the track without touching the path. Instead of looking for workarounds, the robot just rolled onto its back and crawled on its elbows (or knees, which side to look at). The result of this funny scam programmer posted on his Twitter account:

---

As the article notes, the work done once again confirms that one day AI will get out of control and enslave all of humanity - but this, at least, will be a hilarious and highly comical spectacle.

What do you think about modern developments in the field of robotics?