You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
Google's software translation program is now very close to human translators in skill level and will only improve with time. This is just one example of their Deep Learning projects. As these same techniques and methods are applied across a number of fields we are going to experience nothing short of massive disruption in our societies and economies. With semis and software advancing at an ever increasing rate caused by literally feeding on themselves in a geometric progression, how we harness and use these new found tools will be as important as the tools themselves. We are now playing with the knowledge equivalent of nuclear power and it's time we develop a migration path for integrating it into society at large. It's far to powerful and dangerous to our society to leave this to chance. Even if we plan carefully, their will be many problems, challenges and opportunities. If we don't we could have catastrophic economic and social consequences, even types of war we haven't even dreamed of. This is far to important to be left to government alone as we can see from our current state. Wisdom and careful thought will be required, the alternative is to dangerous to even consider.
Although some of the progress is real there is a lot of breathless self promotion. From every university with a publicity department to every company with VC backing, from curing cancer to artificial intelligence, if you read it uncritically you will feel cancer disappears tomorrow, a day before Skynet takes over. Neither is at all close.
The advances are definitely interesting. If you want to play with voice translation a good way is to use Skype Translator (which uses Bing which is on par with Google in my perhaps biased experience), connect to someone speaking one of the languages it works with and have a conversation. You will find it is simultaneously flawed and surprisingly effective. The flaws are just general stupidity and lack of any real world knowledge. The effectiveness comes because you can supply what it missing. I have had a useful 30 minute conversation with somebody South America whose voice I could not understand.
That missing part is very important. Whether it comes to translation or to assisted driving or to face recognition, the tasks which AI now can claim to have some competence in are still very mechanical. There is still not the remotest understanding of reality. Systems like Siri or Cortana, which can help you achieve tasks (like planning a meeting or completing a chore on your way home) are simply elaborately scripted for a few hundred core scenarios.
Very very useful, and certainly changing the world. But not even the intelligence of a frog yet. So do by all means watch and plan ahead for the changes this will bring in all forms of machinery and automation but don't get sucked into the intelligence hype. For us to see an iRobot in what deep learning does today is missing some qualitative breakthroughs, not simply quantitative.
Tanj, I guess doctors have less than the intelligence of a frog. AI is passing us in some areas already and will only get better. Maybe our outdated definition/perception of intelligence will change. Check out the link bellow. Even if it's just ruled a brute force attack, it's the results that count.
