“AI is not as good as some think it is”, says developer in charge of beating humans in a card game

In 2019 the DeepStack algorithm was the first artificial intelligence to beat humans in Texas Hold'em Poker. But according to one of the people who created it, AI is not as good as some think it is. "They're biased towards problems where the answers are in the training data," says Michael Bowling, a professor of computer science at the University of Alberta, in Canada.<br><br>Bowling has spent his career working with AI and games. His project beat the best professional poker players in the world. But despite his experience, he affirms that his major worry is the overconfidence in technology. “I’m not too worried about AI taking over the world, I’m not too worried about AI creating conscious suffering. I’m worried about us relying on it when we shouldn’t be."<br><br>His concerns aren't due to fear of AI becoming superhuman. "[AI] is very much not as dangerous as we think it is," he said. By that he means that AI is not as good as the public thinks it is. “They're biased towards problems where the answers are in the training data," he said.<br><br>"They're very narrow. They're not these super general AI that everyone's worried about, or at least we're not there yet. As evidence, I would say look at all the applications of AI right now, which it’s been a decade since the big deep learning revolution, and what’s come out of it? What are the new applications that we have?"<br><br>Bowling doesn't believe AI is bringing us revolutionary benefits. According to him, most AI applications are based on what he calls the “modest benefit” it provides. "Media companies are using AI to push their algorithms to keep you watching videos on YouTube. And I don't think that's a net positive. Digital media companies are using AI to get you to spend more time on their platforms. I don't think that's a net positive."<br><br>He's worried about people believing in AI more than they should. He gives another example. "We all know that AI has actually made the self driving cars worse now, not better. And when it fails, it fails spectacularly."