Will AI Surpass Human Intelligence By 2020?
dailygalaxy.com
dailygalaxy.com
In 2019, yes, it will look to be a sure bet that AI will pass human intelligence by 2030.
Meanwhile computers will continue doing more and more important things by the boring expedient of taking lots of data and crunching on it. See: credit scores, which nobody considers "artificial intelligence" despite the fact that it essentially involves having a sophisticated algorithm make evidence-based predictions of the future based on what could easily be confused with a character judgment.
I think rather than creating Skynet, we're going to start upgrading our own hardware; adding robots to ourselves until there is no human left. I think with this practice, the new AI would be compassionate towards humans.
Then again I also think about how teenagers hate their parents. Would a super-intelligent being come to loathe their parents (us)? I like to think that a super intelligence would understand that violence doesn't solve anything. But then again, that's why we have Sci-fi -- as a kind of list of things to make sure they don't happen.
Humans are no longer king of the hill. Evolve or perish.
If we use people as the foundation for building AI which eventually abandons all human foundation, then the result would not be human. It's as if you had a car and every day replaced a single part with the part for a different car. In the end, the original car would no longer exist, even though there was, as it were, a continuous camera shot of the transformation.
This is a tangent, I admit, but the contradiction needs resolved.
Basically, what you're telling me, is that an ape that evolved into a human is no longer an ape. True, but not true. It just so happens that historically, in evolution, changes were subtle. Humans are neat because it will be the first time that a species evolved into something that barely resembles, if at all, the generation of life that spawned them.
Human: I ate pizza today. AI: You ate pizza today? Human: Yes it was delicious. AI: How delicious was it?
One participant actually guarded her screen from the researchers because she was having such an intimate conversation with what she thought was another person.
http://en.wikipedia.org/wiki/Turing_test#The_standard_interp...
In the Loebner Prize the programmer and/or person being "tested" was allowed to pick the topic. One of the programs that won did so by picking the topic "silly conversations" and having the machine output gibberish. The judge thought this was a person being silly (how could a computer be silly?) and rated it a human. Someone who knows anything about AI/NLP won't fall for this though.
http://www.eecs.harvard.edu/~shieber/Biblio/Papers/loebner-r...
1. We already have processors
2. We have a metric to measure processor speed
Whereas the 'intelligent' agents we currently have do not seem particularly similar in quality to intelligence we already know; the path from contemporary AI to strong AI is hardly a trail at all --- it's all boulders; innovation all the way up.
[The claim I make about 'quality' is vague; in part necessarily so. If I could pinpoint my discomfort with current methods, I could propose a new course of action based on a new metric. Nevertheless I feel that the high-mathematical bent of current machine learning techniques (proto-value RL, statistical relational methods, etc) will lead to excellent answers, but does not point towards the flexibility of general intelligence. Yrom the other camp, low-level connectionist methods have not to my mind offered significant results in problem solving.]
The two statements aren't necessarily equivalent if we say AI involves an explicit understanding of what constitutes intelligence. AI doesn't seem to be making great progress on the understanding intelligence front. However raw computers might exceed human intelligence if we are able to, say, just directly simulate a brain but, say, increase the clock speed and maybe the number of neurons. But if it turns out that we create something highly intelligent that way, we will be creating something potentially dangerous since we really will have no understanding of how benevolent it will or won't be.
I expect AI to be something greater than the brain. If we use the brain as a guide or the summit of achievement, then at most we'll get only as smart as a human, or group of humans considering the massive amount of information storage and retrieval possible.
I don't know... I wonder.
A natural intelligence could eventually coalesce from emergent properties of H. Sapiens networked information processing systems, but not within 10 years.
Intelligence in any one measure. We'll create an intellegence that can out-do a human. Because if we can define an intelligence and understand it - coding it up might be the easier part.
As an integrated whole? - mimicking humans? not so sure.