Much to the chagrin of vacationers who are planning a summer picnic, the weather is incredibly Moody and unpredictable. Small changes in precipitation, temperature, humidity, wind speed or direction can change to outdoor conditions for some hours or days. Why weather forecasts are usually not made more than seven days into the future — and, therefore, the picnics require replacement plans.
But what if we could understand chaotic system well enough to predict how it will behave far into the future?
Is it possible to predict the weather for the year?
In January 2018, scientists have succeeded. They used machine learning to accurately predict the outcome of a chaotic system over a much longer period of time than thought possible. And the car made it just by observing the dynamics of the system, not having any idea about the equations behind it.
The thrill, fear and excitement
We have already started to get used to the incredible manifestations of artificial intelligence.
Last year a program called AlphaZero has learned the rules of chess from scratch in just a day, and then defeated the world’s best programs for playing chess. She also learned the game of go, and surpassed the former champion of silicon, the algorithm AlphaGo Zero, who improved his game in the process of trial and error after he fed rules.
Many of these algorithms begin with a pure state of blissful ignorance and quickly gain knowledge by watching or playing against ourselves, improving ourselves at every step thousands of times per second. Their ability to inspire fear, awe, excitement. Often we hear about the chaos into which they can plunge humanity once.
But more interesting, which will make artificial intelligence with science in the future, with its “understanding”.
Perfect prediction means understanding?
The majority of scientists will probably agree that prediction and understanding are not one and the same. The reason lies in the myth about the origin of physics — and, one might say, modern science in General.
The fact that more than a thousand years, people have used the techniques proposed by the Greco-Roman mathematician Ptolemy to predict the movement of the planets across the sky.
Ptolemy didn’t know anything about the theory of gravity or that the sun was the center of the Solar system. His methods included the ritual calculations using circles within circles within circles. And although they predicted planetary motion pretty well, nobody understood why it works and why the planets obey this seemingly complicated rules.
Then there was Copernicus, Galileo, Kepler and Newton.
Newton discovered the fundamental differential equations which govern the motion of each planet. With their help it was possible to describe each planet in the Solar system. And it was perfect, because we understand why planets move.
The solution of the differential equation turned out to be a more effective method of predicting planetary motion in comparison with the algorithm of Ptolemy. More importantly, however, is that our belief in this method has allowed us to open a new invisible planet, because of the law of universal gravitation. He explained why rockets fly and the apples are falling, and why are the moon and galaxies.
This basic pattern — find the set of equations describing the unifying principle has been used successfully in physics over and over again. So we have identified the Standard model, the culmination of half a century of studies of particle physics, which accurately describes the structure of each atom, the nucleus or particles. So we are trying to understand high-temperature superconductivity, dark matter and quantum computers. (The unreasonable effectiveness of this method even raised questions about why the universe is so perfectly amenable to mathematical description).
In the whole of science the concept of understanding something means returning to the original pattern: if you can reduce a complex phenomenon to a simple set of principles, you have understood it.
Exceptions to the rule
And yet, there are regrettable exceptions that spoil this beautiful story. Turbulence is one of the reasons why predicting the weather is difficult — a vivid example from physics. The vast majority of problems in biology, from the intricate structures in other structures also can not be explained with a simple principle of unification and simplification.
Although there is no doubt that atoms and chemistry, and hence the simple principles underlying these systems are described by a universally effective uravnenii, it’s a pretty inefficient way of generating useful predictions.
At the same time, it becomes obvious that these problems are easily amenable to machine learning methods.
Just as the ancient Greeks sought answers from the mystical Delphi Oracle, we will look for answers to difficult questions of science the all-knowing oracles with artificial intelligence.
Such oracles are already driving Autonomous cars and choose objects for investment in the stock market, and very soon will be to predict which drugs will be effective against bacteria — and what will be the weather in two weeks.
They will make these predictions with high precision, which we never dreamed of, not using any mathematical models and equations.
It is possible that, armed with data on billions of collisions at the Large hadron Collider, they will cope better with the predictions of the outcome of the experiment with particles, than even favorite physicists Standard model.
Like unexplained sources of revelation of the priestesses of Delphi, our prophets of artificial intelligence are also unlikely to be able to explain why they predict so, and not otherwise. Their findings will be based on many microseconds that can be called “experience.” They will be like the uneducated farmer who is able to accurately predict how weather will change, “because the bones are aching” or other premonitions.
Science without understanding?
The consequences of machine intelligence in science and philosophy of science can be stunning.
For example, in the face of more accurate predictions, though obtained by the unknown man will we deny that machines have better knowledge than we?
If the prediction actually is the main goal of science, as we need to modify scientific method, an algorithm, which for centuries allowed us to identify mistakes and correct them?
If we abandon the understanding, whether it makes sense to do science, which we were doing?
Nobody knows. But if we can’t articulate why science is more than the ability to make good predictions, scientists will soon discover that the “trained artificial intelligence makes them better than their own”.