DailyDirt: Teaching New Computers Old Tricks
from the urls-we-dig-up dept
Despite the staggering growth in computing power and capabilities over the history of the technology, there remains a line in the sand between what computers can do and what we think of as "true" artificial intelligence. This line has gotten blurrier as computers have succeeded in performing certain tasks that were formerly human-only, but even these instances often feel like a brute-force approach to simulating something our own brains seem to accomplish more genuinely and abstractly. On the flipside, the success of these simulations raises questions about what's really happening inside our own heads. Here are a few of the latest developments in artificial intelligence that try to approach that line:
- Google has developed software that can visually solve CAPTCHA puzzles with 99.8% accuracy. The project grew out of technology for reading house numbers in Street View — something it's actually less successful at. [url]
- Researchers are experimenting with robots that get taught their task — in this case, playing Pac-Man — by other robots. Though the test has been successful, the researchers openly admit that the (virtual) bots are "very dumb." [url]
- A startup called Vicarious is chasing the holy grail of AI: true imagination and abstraction. The company is keeping mum about the details of its tech, but it has some serious resources from high-profile backers including Jeff Bezos, Mark Zuckerburg and (according to a rumour) possibly Elon Musk. [url]
If you'd like to read more awesome and interesting stuff, check out this unrelated (but not entirely random!) Techdirt post via StumbleUpon.