Actually the term strong AI only refers to AI in which can feed itself and improve its judgement through that feedback. In that case, we might have similar opinion towards this.
"http://en.wikipedia....wiki/Strong_AI"
"Strong AI is hypothetical artificial intelligence that matches or exceeds human intelligence — the intelligence of a machine that could successfully perform any intellectual task that a human being can. [1]"
With that aside, learning is a strong word to be used right now, as today's AI often relies on predetermined algorithms to learn. Even if it's a recursive algorithm, in which it's nothing more than a feedback system that has been developed since the 60s, it's nothing spectacular in terms of intelligence. I should have asked you that your definition of intelligence because you confuse it with knowledge. Again google's machine learning algorithm, with all due respect, can also be learnt brefly in a undergraduate level machine learning course. They have brilliant and sophisticated approach towards this, however it's still a predetermined algorithm. The reason we know is that there are many web developers who can still cheat on the algorithm and attracting viewers to their websites. Myself included.
Every time that AI does something remarkable, such as beating the world chess champion, understanding natural language enough to crush world Jeopardy champions, or learning to recognize salient features from fuzzy data (pictures/videos), critics remark about how that achievement really isn't "spectacular." It's so tiring to hear these people constantly downplay these incredible achievements in emulating a part of human capability just because they have not achieved full fledged human intelligence AI yet.
AI is becoming much broader very quickly. We're not simply building systems that can recognize a human face, but systems that can recognize an abundance of features from the data they're being fed.
Learning to recognize features from fuzzy data, learning to walk upright like a human being (see my original post for the Petman videos), and achieving natural language understanding to recognize subtle puns and jokes is incredibly hard for computers to do. Yet here we are in 2013, capable of doing all those things and more.
These systems do learn, and they do it in a remarkable way: namely, by taking vast amounts of unlabeled, unsorted data, much as humans learn about the world, and recognizing patterns and features within that data.
It's not perfect, yet, by any means, but that doesn't mean that these capabilities will won't quickly improve as the technology evolves over the next decade.
Alright you are correct about the fact that we don't need strong AI to replace many human menial jobs. In fact, that's what we are doing right now, replacing many jobs by using industrial machinery. This is old news my friend. However if you try to replace jobs that involved uncertainty, then it really takes a long way before getting there. Namely we know that there are more than 20k vending machines in US, however none of them has replaced convenience store, which basically sales the same thing. Automatic machinery still requires supervision of human to make it work. That's what we have right now. And even the google car argument fails because they haven't been put into massive trials.
Far from old news. The landscape of trade is changing at a vicious pace. The internet has changed everything about shopping.
Everything is going online. Amazon.com and online shopping in general has revolutionized the way we do business and shop for goods. Everyone I know shops for goods online. Convenience stores, and stores in general, are getting wiped out because of online shopping and online entertainment. Look at Blockbuster's recent closure and you will see the fate of most physical stores.
Let's not forget that 3D printing could be a huge game changer. Although it is in its infancy, 3D printing may on day have a disruptive effect on the world when we can print out physical goods at home. While materials, speed, and resolution remain key limitations to 3D printing's success, we can expect these to be overcome in the foreseeable future.
Driverless cars are another example that you dismiss just because they haven't been put into trials. You can look at the data so far. 500,000 miles (more by now) with zero accidents that are the fault of the computer and only 2 or 3 that were a result of human error. Sergei Brin says they'll be commercially available by 2017. A huge number of other car companies, including Nissan, Mercedes, and Tesla, have already built prototypes. These cars are, without a doubt, coming to a road near you over the next decade.
Given a simpler example, many programming courses, as well as literature courses use auto marking software to mark. And guess what, students immediately search for methods to beat the system. In fact I have to file a remarking petition every time after the auto marking results. That's the limitation of weak AI, it doesn't adapt to the environment but it rather relies on existing algorithm and the intelligence of the developers, who at least have their jobs secured because they are required. However if strong AI exists, my collegues and I all believe that every working class will be out of their jobs forever from that point. There's no way a human can compete with a 24/7 machinery which can learn from its mistakes.
You can make fun of the limitations of certain types of AI that exist at the moment, but it's clear that its capacity is getting far better in many key domains, including machine vision and natural language understanding. Don't forget, we're also changing the hardware that these neural networks can run on, and one of the key efforts is being led by IBM in building neurosynaptic chips.The potential gains in efficiency and effectiveness in building cognitive, adaptive systems, is incredible.
Edited by Elus, 24 November 2013 - 06:10 AM.
