With these techniques, machines are also learning to read and write. For years, experts questioned whether neural networks could crack the code of natural language. But the tide has shifted in recent months.
Organizations such as Google and OpenAI, an independent lab in San Francisco, have built systems that learn the vagaries of language at the broadest scales — analyzing everything from Wikipedia articles to self-published romance novels — before applying the knowledge to specific tasks. The systems can read a paragraph and answer questions about it. They can judge whether a movie review is positive or negative.
This technology could improve phishing bots such as SNAP_R. Today, most Twitter bots seem like bots, especially when you start replying to them. In the future, they will respond in kind.
The technology also could lead to the creation of voice bots that can carry on a decent conversation — and, no doubt one day, will call and persuade you to divulge your credit-card information.
These new language systems are driven by a new wave of computing power. Google engineers have designed computer chips specifically for training neural networks. Other companies are building similar chips, and as these arrive, they will accelerate A.I. research even further.
Jack Clark, head of policy at OpenAI, can see a not-too-distant future in which governments create machine-learning systems that attempt to radicalize populations in other countries, or force views onto their own people.
“This is a new kind of societal control or propaganda,” he said. “Governments can start to create campaigns that target individuals, but at the same time operate across many people in parallel, with a larger objective.”
Article source: https://www.nytimes.com/2018/11/19/science/artificial-intelligence-deepfakes-fake-news.html?partner=rss&emc=rss
Speak Your Mind
You must be logged in to post a comment.