These days there is no shortage of strict warnings about the dangers of artificial intelligence.
Modern forecasters, such as physicist Stephen Hawking and investor Elon Musk, Fortal Imminent decline of humanity. with the advent of Artificial common sense And self-designed intelligent programs, new and more intelligent AI will appear, increasingly creating ever-smarter machines that will, eventually, surpass us.
Is this really what we have to look forward to?
AI’s checkered past
not really no.
AI, rooted in computer science, mathematics, psychology, and neuroscience, aims to create machines that mimic human cognitive tasks such as learning and problem solving.
Since the 1950s, It has captured the imagination of the public. But, historically, the successes of AI have often been regarded as a cause of disappointment – due, in large part, to the perceived predictions of technological visionaries.
In the 1960s, one of the founders of the AI field, Herbert Simon, Predicted “Machines will be able within twenty years to do any work that a man can do.” (He did not say anything about women.)
A neural network pioneer, Marvin Minsky was more direct, “Within a generation,” he said, “… the problem of creating ‘artificial intelligence’ will be solved to a great extent”.
But it turns out that Niels Bohr, early 20th century Danish physicist, Was right when he (reportedly) quipped, “Prediction is very difficult, especially about the future.”
These talents hardly made humans irrelevant.
New neuron euphoria
But AI is moving forward. The most recent AI Euphoria of 2009 was sparked by very fast learning Deep neural network.
Artificial intelligence consists of large collections of connected computational units called artificial neurons, which correspond to neurons in our brain. To train this network to “think”, scientists provide it with several solved examples of a given problem.
Suppose we have a collection of medical-tissue images, each combined with a diagnosis of cancer or no-cancer. We will pass each image through the network, from which the associated “neurons” calculate the probability of cancer.
We then compare the network’s responses with the correct answer, adjusting the connections between the “neurons” with each failed match. We repeat this process, fine-tuning until most of the responses match the correct answers.
Eventually, this neural network will set out to do what a pathologist normally does: examine images of tissue to predict cancer.
This is not unlike how a child learns to play an instrument: he practices and repeats a tune until perfection. Knowledge is stored in neural networks, but mechanics are not easy to interpret.
Networks with multiple layers of “neurons” (hence the name “deep” neural network) became practical only when researchers started using multiple parallel processors on graphical chips for their training.
Another condition for deep learning success are large sets of solved examples. Mining on the Internet, social networks and Wikipedia, researchers have created large collections of images and text, enabling machines to classify images, recognize speech, and translate language.
Already, deep neural networks are performing these tasks almost simultaneously with humans.
AI doesn’t laugh
But his good performance is limited to a few works.
Scientists have not seen any improvement in AI’s understanding of what images and text actually mean. If we show a snowpoppy cartoon in a trained deep network, it can recognize shapes and objects – a dog here, a boy there – but will not understand its importance (or see humor).
We also use neural networks to suggest better writing styles to children. Our tools suggest an improvement in form, spelling, and grammar as much as possible, but are helpless when it comes to logical structure, logic, and the flow of ideas.
Current models also do not understand the simple creations of 11-year-old schoolchildren.
The performance of AI is also restricted by the amount of data available. My own AI research, For example, I apply deep neural networks in medical diagnostics, sometimes resulting in a slightly better prognosis than in the past, but nothing dramatic.
In part, this is because we do not have a large collection of patients’ data to feed the machine. But the data the hospitals currently collect cannot capture the complex psychoses that lead to diseases such as coronary heart disease, migraine, or cancer.
Robots stealing your jobs
The capabilities of AI drive science fiction novels and films and fuel Interesting philosophical debateBut we still have to build the same Self improvement program Capable of general artificial intelligence, and there is no indication that intelligence can be infinite.
However, deep neural networks will be indefinitely Automate many jobs. AI will take our jobs, endangering the existence of our laborers, medical diagnostists, and perhaps someday, to my regret, computer science professors.
Robots are already conquering Wall Street. Research suggests that the “artificial intelligence agent” could make some 230,000 finance jobs disappear by 2025.
In the wrong hands, artificial intelligence can also cause serious danger. New computer virus Can find unspecified voters And bombard them with news adapted for swing elections.
Already, the United States, China, and Russia are investing in autonomous weapons, fighting drones, war vehicles, and robots in AI, Dangerous arms race.
Now we should probably be worried about this.