DailyDirt: Can You Teach An Old Machine New Tricks?

There are plenty of examples of self-learning machines nowadays. Yamaha has a motorcycle-riding bot named Motobot that wants to ride faster than a professional human motorcycle racer (though it cheats a bit with some training wheels). Machine learning is rapidly becoming a capability that we'll all encounter in our daily lives (if we haven't already with algorithmically-generated news feeds or ads or recommendations). Here are just a few other artificial intelligence projects that could be worth keeping an eye on.

• A robotics startup, Brain Corporation, is developing software for robots that can be trained like a dog -- with repetition, not programming. This kind of training might be faster than programming for some complex tasks, but there's some unpredictability that goes along with it -- so maybe this isn't the best approach for a heart surgery robot, but it's pretty good for a robot that just has to pull weeds. [url]
• The field of artificial general intelligence (AGI) is relatively immature compared to the efforts of getting computers to play specific games or learn particular skills. Software that can learn like a human child has been able to do some basic math and logic, recognize grammar rules and identify patterns -- without pre-programmed knowledge of any of those things. [url]
• Self-learning software has been able to recognize phonetic components of speech without the help of human training. The most commonly used speech recognition systems have been trained at some point by humans to distinguish various phenomes in our spoken language, so a system that can do so on its own might be more useful for identifying idiosyncrasies in different speakers' pronunciations. This kind of machine learning could also provide insight into how human kids learn languages, too. [url]

After you've finished checking out those links, take a look at our Daily Deals for cool gadgets and other awesome stuff.

Permalink | Comments | Email This Story