There's an old joke: a typist applies for a job, and when asked how fast she types, answers confidently — two thousand characters a minute — then adds quietly, under her breath, but it comes out complete garbage. In some ways, the joke was ahead of its time. Today's algorithms process data and make decisions at a far greater speed, but sometimes what comes out is genuine garbage too. Without getting into complicated mathematical details like Big O notation, we can take as a starting point the claim that data-processing speed is limited only by the power of the available computing infrastructure — which means every new technological breakthrough pushes it closer to infinity.t For those who never experienced the era of the typing pool: a good typist, reading straight off a page and typing on an electric "Yatran" typewriter, could manage a maximum of 400 characters a minute. Machine data processing — which has traveled all the way from supercomputers (one of which, according to old Moscow legend, is hidden somewhere deep beneath the physics building of Moscow State University) to distributed cloud networks — isn't really such a new technology. In fact, the very first weather forecast was produced using the ENIAC computer in the United States back in 1950. Seventy-six years later, specialists at ETH Zurich loaded 100 petabytes of NASA weather-satellite data into their own data center to train an AI system specializing in long-term climate forecasting. Weather forecasting is, in fact, the first example — leaving aside military applications — of using algorithms to make decisions based on analysis of a large body of data. Starting in the 1990s, a breakthrough period for the global microelectronics industry, the algorithmization of processes seized new ground as rapidly as the Red Army did during the 1945 Vistula–Oder Offensive: medicine, the stock market, logistics — there's hardly a field left where a specialist's judgment hasn't come under fire, or at least under the helping hand of an algorithm. If, at first, the machine only had an advisory role — merely suggesting the right decision to a human being — over time the roles reversed, and now it's the human who has become an observer of the machine, albeit one with the right to occasionally make adjustments to its work. Yes, the machine sometimes made mistakes, but no more often than a human would, and, on the whole, within a reasonable margin of error. The first serious breakdown came in May 2010, when algorithmic trading systems around the world failed to recognize a maneuver by the broker Navinder Singh Sarao, who had decided to bet on a decline in the S&P 500 — and in a matter of minutes wiped a trillion dollars off the value of the US stock market. Whether that particular order really was to blame for the crash is still disputed, but the fact remains: automated trading came within a hair of destroying dozens of companies without any human involvement at all. The question of whether decision-making can be entrusted to a machine is, in my view, a good deal thornier than the classic "trolley problem." On one hand, a machine is built to flawlessly carry out the rules programmed into it (though we increasingly hear about AI models attempting to write their own rules for how they operate). On the other hand, unconditional obedience to instructions matters for people carrying out orders — not for those in charge, whose whole job is knowing when to depart from the rulebook. This clearly holds true across every field where process automation has already taken hold. What does a robot do when it comes across a defective part or a broken tool? The whole production line stops. A human, by contrast, sizes up the situation instantly: "Ah, there's a burr here — I'll file it down and keep going," or "This bolt won't fit — I'll swap the head." A human can improvise and find workarounds in exactly the spots where a robot's algorithm hits a dead end. Incidentally, the new trend in factory automation isn't building fully automated lines, but populating the assembly line with collaborative robots ("cobots") capable of working literally hand in hand with people. Another field where automated decision-making is advancing at breakneck speed is medicine. And, paradoxically, the push isn't coming only from doctors. Physicians describe the emergence of a new type of patient — people who, before ever seeing a specialist, have already had all the tests they deem necessary, fed the results into one of the public LLM models, and shown up for their appointment with a ready-made diagnosis and treatment protocol in hand. All that's left for the doctor to do, it seems, is take on the legal responsibility for that decision by writing out the corresponding prescriptions.Standing back and looking at all these processes together, it seems fair to say that the period of romantic infatuation with algorithms, and the sincere faith in their infallibility, is giving way to a sensible, pragmatic attitude toward them. The illusion of the machine's total replacement of the human being is gone. An autopilot may be able to get a plane into the air, but it's still the pilot who will bring it down.