Skip to content

What the history of AI tells us about its future

But traditionally, what was worse was the strategy of the computers – the ability to think of the shape of the game in the future in many, many moves. That’s where the men still had the edge.

Or Casparov thought, until the Deep Blue move in Game 2 knocks him out. He seemed so civilized that Kasparov began to worry: maybe the machine was a lot better than him! Convinced he had no way of winning, he resigned from another game.

But he shouldn’t have. Deep Blue, it turns out, wasn’t really that good. Kasparov failed to see a move that could have ended in a draw. He was cycling himself: worrying that the machine might be more powerful than it really was, he began to see human-like logic where no one existed.

Breaking his rhythm, Kasparov kept playing worse and worse. He repeatedly pulled himself out. At the beginning of the sixth, winner-take-all game, he made such a light move that the chess observers shouted in shock. “I was not in the mood to play at all,” he told a news conference later.

IBM benefited from its moonshot. Following the success of Deep Blue, the company’s market cap rose to $ 11.4 billion in a single week. More importantly, IBM’s win seemed to melt over the long AI winter. If chess can be won, what was next? The minds of the masses turned again.

“That,” Campbell tells me, “is the one who got people’s attention.”

The truth is, it was not surprising that the computer beat Kasparov. Most people who focus on AI – and on chess – expect that to happen.

Chess may seem like a unit of human thought, but it is not. Indeed, it is a mental function that is perfectly suited to brute force computation: the rules are clear, there is no hidden information, and the computer does not even need to keep track of what happened in the previous move. It just evaluates the condition of the pieces.

“There are very few problems where, like chess, you have all the information you need to make the right decision.”

Everyone knew that once the computer was fast, it would drown the man. That was the only question. By the mid-90s, “in a sense the text was already on the wall,” says Demis Hasabis, head of the AI ​​company Deepmind, part of Alphabet.

Deep Blue’s victory was the moment that showed how limited hand-coded systems can be. IBM spent years and millions of dollars developing computers to play chess. But he could do nothing else.

“It simply came to our notice then [Deep Blue] Campbell says AI will have a huge impact on the world. They couldn’t really find any principles of intelligence, because the real world doesn’t look like chess. Campbell adds, “There are very few problems where, like chess, you have all the information you need to make the right decision.” “Mostly there are strangers. There is chaos. “

But even as Deep Blue was cleaning the floor with Kasparov, a handful of scrapie upstarts were tinkering with the radically more promising form of AI: Neural Net.

With Neural Net, like the expert system, the idea was not to patiently write rules for every decision made by AI. Instead, training and reinforcement reinforces the internal connections (as is the theory) in the rough emulation of how the human brain learns.

1997: After Gary Kasparov defeated Deep Blue in 1996, IBM asked the World Chess Champion for a rematch, which was held in New York City with an upgraded machine.

AP Photo / Adam Nadel

The idea has been around since the 50’s. But training a large neural net in a useful way requires lightning-fast computers, a lot of memory and a lot of data. None of that was readily available then. Even in the 90’s, neural net was considered a waste of time.

“At the time, most people in AI believed that the neural net was just rubbish,” says Geoff Hinton, an emeritus computer science professor at the University of Toronto and a pioneer in the field. “I was called a ‘true believer'” – not a compliment.

But by the 2000s, the computer industry was evolving to enable the neural net. Video game players’ longing for ever-better graphics created a huge industry in ultrafast graphic-processing units, which turned out to be perfectly suited to neural-net math. Meanwhile, the Internet was exploding, producing streams of pictures and text that could be used to train the system.

In the early 2010’s, this technological leap allowed Hinton and his crew of true believers to take the neural net to new heights. They can now network with multiple layers of neurons (which means “deep” in “deep learning”). In 2012, his team won the annual Emanate competition by hand, where AIs compete to identify elements in pictures. It stunned the world of computer science: self-learning machines were finally viable.

Ten years into the deep-learning revolution, neural nets and their pattern-recognizing capabilities have colonized every corner of everyday life. They help Gmail automate your sentences, help banks detect fraud, allow the photo app to automatically recognize faces, and – in the case of OpenAI’s GPT-3 and DeepMind’s Gopher – write long, humane voice essays and summarize text. . They are also changing how science is done; In 2020, DeepMind debuted AlphaFold2, an AI that could predict how proteins would fold — a superhuman skill that could help guide researchers in developing new drugs and treatments.

Deep Blue, meanwhile, disappeared, leaving no trace. Playing chess, it turns out, was not the computer skills that were needed in everyday life. Hasabis, founder of Deepmind, says, “Everything that Deep Blue showed in the end was a flaw in trying to craft.

IBM tried to resolve the situation with Watson, another specialized system, designed to deal with this one more practical problem: getting a machine to answer questions. It uses statistical analysis of a large amount of text to gain an understanding of the language that was advanced for its time. It was more than a simple if-then system. But Watson faced an unfortunate time: it was only a few years later that it was eclipsed by a revolution in deep learning that brought a generation of much more subtle language-crunching models than Watson’s statistical techniques.

Deep learning runs precisely on old-school AI because “pattern recognition is incredibly powerful,” says Daphne Caller, a former Stanford professor who founded and operated Insitro, a novel that uses neural net and other forms of machine learning. To investigate drug treatment. The flexibility of the neural net વિવિધ various ways of identifying patterns ઉપયોગ can be used જ for the same reason that AI winter has not yet arrived. “Machine learning has really paid off,” she says, “never did the previous waves of enthusiasm” in AI.

The inverted fate of Deep Blue and Neural Net shows how bad we were at deciding what is difficult and what is valuable in AI.

For decades, people have believed that mastering chess is important because, well, chess is difficult for humans to play at a high level. But mastering chess became quite easy for computers, as it is very logical.

The hardest thing for a computer to learn was the casual, unconscious mental work that humans do – such as communicating live, driving through traffic, or reading a friend’s emotional state. We do these things so easily that we rarely realize how difficult it is and how vague, grayscale judgment they need. The great utility of deep learning comes from being able to capture these tiny bits of subtle, unheard human intelligence.

However, there is no ultimate victory in artificial intelligence. Deep studies may be moving too far now – but it is also gathering sharp criticism.

“For a long time, this was the techno-chauvinist zeal that, well, AI is going to solve every problem!” Says Meredith Brosard, professor of journalism and author who became a programmer at New York University Artificial intelligence, But as she and other critics have shown, deep-learning systems are often trained on biased data અને and they absorb biases. Computer scientists Joy Buolamvini and Timnit Gabru discovered that three commercially available visual AI systems are terrifying for analyzing the faces of dark-skinned women. Amazon trained AI for VAT resumes, just to find those women down rankings.

Although computer scientists and many AI engineers are now aware of these bias problems, they are not always sure how to deal with it. On top of that, the neural net is also a “big black box,” says Daniela Rouss, an AI veteran who currently runs MIT’s computer science and artificial intelligence laboratory. It is not clear how he came to this conclusion – or how he will fail.

“For a long time, this was the techno-chauvinist zeal that, well, AI is going to solve every problem!”

Relying on the black box for work that is not “important to safety” is not a problem, according to Russian statistics. But what about high-end jobs like autonomous driving? “It’s really remarkable that we can put so much trust and confidence in them,” she says.

This is where Deep Blue had an advantage. The old school style of craft rules may have been brittle, but it was understandable. The machine was complex – but it was no secret.

Ironically, that old style of programming could make a comeback as engineers and computer scientists grapple with the limitations of pattern matching.

Language generators like OpenAI’s GPT-3 or DeepMind’s Gopher can pick up a few sentences you’ve written and move on, writing pages and pages of intelligible-sounding prose. But despite some impressive mimicry, Gopher “still doesn’t really understand what he’s saying,” says Hasabis. “Not really.”

Likewise, Visual AI can make terrible mistakes when it encounters edge cases. The self-driving car collided with a fire truck parked on the highway, as they were trained in millions of hours of video, they had never encountered such a situation. Neural Net has, in its own way, a version of the “brittleness” problem.

Leave a Reply

Your email address will not be published.