Machines, teaching each other, could be crucial for artificial intelligence

Machines, teaching each other, could be crucial for artificial intelligence


During a press conference on the announcement function of the autopilot in the Tesla Model S, which took place in October of 2015, Tesla CEO Elon Musk said that every driver will be “coach-expert” for each Model S. Each car would be able to improve their function, autonomy, learning from his driver, but more importantly — when one Tesla will learn from your driver, this knowledge will be distributed among the rest of the cars Tesla.

Very soon the owners of Model S have noticed that the functions of self car gradually improve. In one example, the “Tesla” were done wrong the early exits of the motorway, forcing their owners to manually hold the car on the correct route. After just a few weeks, the owners said that the car is no longer making premature exits.

“It is striking that the improvement happened so soon,” said one of the owners of Tesla.

Intelligent systems, such as those equipped with the latest software for machine learning, not just become smarter, they become smarter faster and faster. Understanding the speed at which these systems are developing, may be particularly difficult part of managing technological progress.

Ray Kurzweil has written extensively on the gaps in the understanding of man, describing the so-called “intuitive linear” view of technological change and “exponential” rate of change occurring now. Nearly two decades after writing an important essay, which he called “the Law of accelerating returns” — the theory of evolutionary change, describing the change of speed improvements, system — connected device began to share knowledge among themselves, accelerating their own improvement.

“I think it’s probably most important exponential trend in AI,” says Hod Lipson, Professor of mechanical engineering and computer science at Columbia University.

“All exponential technology trends have different “exhibitors”, he adds. “But this is probably the biggest one.” In his opinion, this “machine teaching” — when devices transmit knowledge to each other (not to be confused with machine learning) is an important step to accelerate the improvement of such systems.

“Sometimes this cooperation, for example when a machine learns from the other, if they swarm consciousness. Sometimes it’s a mess, like the arms race between the two systems playing chess with each other”.

Lipson believes that this development of the AI it’s powerful stuff, partly because that eliminates the need for training data.

“Data is the fuel of machine learning, but even the machines are hard to obtain some data — this may be risky, slow, expensive, or unattainable. In such cases, machines can share their experience or to create a synthetic experience for each other to complement or replace data. It turns out that it’s not such a weak effect is in fact zamovlennya, and exponential”.

Lipson cites the example of a recent breakthrough DeepMind, the project AlphaGo Zero, as indicative of the AI learning without training data. Many are familiar with AlphaGo, AI machine learning, which has become the world’s best go player, having studied the massive volume of data consisting of millions of games played. AlphaGo Zero was able to beat him even without looking at the training data simply by studying the rules of the game and playing with yourself. Then he beat the best in the world for chess games after an eight-hour training.

Imagine how thousands of these AlphaGo Zero instantly share their acquired knowledge.

And it’s not just toys. We can already see how powerful the impact of the speed at which business can improve the performance of their devices. One example is the industrial technology of digital double software model of a machine that simulates what is happening with the equipment. Imagine if the car were to look inside himself and shows his picture techniques.

For example, a steam turbine with digital double can measure the steam temperature, rotor speed, cold starts and other data to predict failures and alert technicians about preventing costly repairs. Digital twins make these predictions, exploring their own performance, and rely on models developed by other steam turbines.

As soon as machines begin to learn their environment in powerful new ways, their development is accelerated due to the data exchange. Collective intelligence each steam turbines, scattered across the planet, can accelerate the predictive capability of each individual machine. Where will one car without a driver, there will be hundreds of other drivers who will be studying your car, giving knowledge to everyone.

Don’t forget that all this is only the beginning.