10 ms·

Peter Norvig: The machine age

95 points by pitdesi 16y ago

l0nwlf 16y ago
``One of the biggest issues in AI is determining what the question is really asking i.e. translating a “natural language” query into something machine can understand and find the appropriate answer.'' - so True.
- jbri 16y ago
  That's also one of the biggest issues in software development.
  - celoyd 16y ago
    Surely it’s the issue in software development. Everything else is a case or subproblem of “how do I use a computer to solve this imperfectly defined problem?” I mean, it’s borderline tautological, isn’t it? Software development is the practice of translating natural-language requests into executables. That’s not just a recurring question – it’s definitional. This is why I’m nervous when strongly antisocial people say they want to go into software. The point of this field is to figure out what people need (or want, or mean, or like). Of course you can do this creatively rather than responsively, and with more rather than less respect for your customers. But you don’t escape human tastes and concerns when your job is to explain them to a microprocessor.
    - jbri 16y ago
      Well, there's the second issue of doing it such that the machine can produce results efficiently. Which tends rather towards the mathy computer-science side of things - which is where I think the less social "I just want to write code" people would be much happier.
      - celoyd 16y ago
        Efficiency is implicit in most natual-language queries. Managing efficiency tradeoffs is a good example of something fuzzy yet important that you have to be able to figure out from natural-language queries and human interaction. But yeah, if you really just want to work with numbers, you’ll probably be happier in CS academia.
    - l0nwlf 16y ago
      Even though I like your answer I disagree with "This is why I’m nervous when strongly antisocial people say they want to go into software". Strongly anti-social people are some of the best software engineers around. They devote insane amount of time and pretty much enjoy coding.
- bluekeybox 16y ago
  One would be mistaken to think that this problem is trivial even to us humans.
Luyt 16y ago
A link to the 'print version' of the page: the whole article on one page without the incredible clutter and cruft that the New York Post splashes all over its site: http://www.nypost.com/f/print/news/opinion/opedcolumnists/the_machine_age_tM7xPAv4pI4JslK0M1JtxI http://www.nypost.com/f/print/news/opinion/opedcolumnists/th...
Smerity 16y ago
In this AI/machine learning is compared to the space race and the fight against cancer - I like the comparison considering I work in the field ;) Unfortunately though I think AI and machine learning has the fundamental problem of rolling improvement... Think back a hundred years to a time before the internet, long distance telephones, common air travel and so on. To people from a hundred years ago we'd be the most outlandish science fiction story ever told. I consider the same thing whilst watching science fiction films - in reality they wouldn't be amazed that they can reach Mars in a half hour. They'd instead be complaining about the in-flight entertainment or the turbulence caused by navigating asteroid fields. AI has the same problem. People can now (naively) communicate with each other even though they share no common language. Texts and documents can be re-written in real time to be legible in another language. Credit card fraud and spam emails are handled transparently by systems trained on hundreds or thousands of hours of cumulative knowledge. No-one sees these things. No-one respects these things. Human kind judges itself from the position it currently stands, not from where it began. For this reason alone incremental improvements will never really be considered milestones of achievement.
- zeteo 16y ago
  "To people from a hundred years ago we'd be the most outlandish science fiction story ever told." You'd be surprised: http://en.wikipedia.org/wiki/Paris_in_the_Twentieth_Century#Predictions http://en.wikipedia.org/wiki/Paris_in_the_Twentieth_Century#...
  - derleth 16y ago
    Technology is one thing; society is another: I doubt anyone from the 19th Century could possibly imagine a debate over gay marriage. A crazy promoting it getting beaten down by a righteous world, possibly, but not a debate.
- Cushman 16y ago
  I don't know if the comparison is that inapt; how many people do you think are really aware of the technology that goes into making that little box say "in five hundred yards, turn left"?
mthomas 16y ago
Perhaps offtopic, but I find it odd that there is an opinion piece by Peter Norvig in the NY Post.
- pgbovine 16y ago
  true, but i'm sure that NY Post has more subscribers than NY Times, so if his goal was to reach the largest possible audience, perhaps NY Post was an appropriate venue
  - necubi 16y ago
    This isn't true. According to Wikipedia, the NY Times has 876,638 daily/1,352,358 Sunday subscribers [0] while the NY Post has 525,000/343,361 [1] [0] http://en.wikipedia.org/wiki/NY_Times http://en.wikipedia.org/wiki/NY_Times [1] http://en.wikipedia.org/wiki/NY_Post http://en.wikipedia.org/wiki/NY_Post
  - LostInTheWoods2 16y ago
    Not to be snobbish, but the NY Post is considered a working class paper. Usually bought by those more interested in the funnies, movie times, and local sports, than say international affairs or ramblings about AI.
bmh100 16y ago
This is an excellent article for raising awareness of AI. However, I disagree with its conclusion that we will always relate to AI as tools, pets, advisers, etc. When we ask whether machines can think, I believe it is a question of volition and self-direction. Does the machine have an open-ended goal and is self aware? Can it alter its own programming to change its own methods, its own knowledge, and its own goals? If the answer to these two is "yes," then I believe we will consider the machine to be thinking. What are the implications? If machines gain those capacities, as well as the ability to compute emotions, then what separates them from us? If you agree that humanity is fundamentally a function of our minds, then, if machines can compute in fundamentally equivalent ways, they become "human." What evidence is there that this is forever impossible?
- tgflynn 16y ago
  If you agree that humanity is fundamentally a function of our minds What about the ability to feel pleasure and pain ? I don't believe a deterministic machine will ever experience such subjective feelings (though it may behave as if it does). Computers may well become sapient but they (at least with current technology) will never be sentient.
  - jbri 16y ago
    What is "pleasure" and "pain"? Ultimately, they're just electrical signals triggering the release of certain chemicals in the brain, with subsequent alterations to how the brain processes other stimuli. "It doesn't really feel pain, it just looks like it does" is probably how many people justified atrocities against other humans they viewed as somehow "lesser".
    - tgflynn 16y ago
      Maybe they're just electrical signals in the brain but that's not how we experience them. A CPU cannot experience the electrical signals that flow through it. Unless you believe rocks are sentient. I could just as easily turn your argument around and argue that torturing you is perfectly OK because all I'd be doing is creating electrical signals in your brain.
wooby 16y ago
The episode "The Thinking Machine" of the Nova miniseries "The Machine That Changed the World" is my favorite introduction to the history of AI up until around AI winter. The episode, along with four others, are available here: http://waxy.org/2008/06/the_machine_that_changed_the_world_the_thinking_machine/ http://waxy.org/2008/06/the_machine_that_changed_the_world_t...
Jun8 16y ago
"Learning turned out to be more important than knowing." This seemingly simple sentence, to me, sums up the article, i.e. the data driven approach as opposed to model-driven ones.
mtrn 16y ago
Single page (print) version: http://www.nypost.com/f/print/news/opinion/opedcolumnists/the_machine_age_tM7xPAv4pI4JslK0M1JtxI http://www.nypost.com/f/print/news/opinion/opedcolumnists/th...
TeMPOraL 16y ago
"Spam filtering programs using A.I. learning and classification techniques correctly identify over 99.9% of the 200 billion spam e-mails sent each day." 200 billion spam e-mails each day. Something should be done with spammers just on the basis of how much electricity it wasted on sending, processing and discarding all those e-mails...

95 points by pitdesi 16y ago

``One of the biggest issues in AI is determining what the question is really asking i.e. translating a “natural language” query into something machine can understand and find the appropriate answer.'' - so True.

jbri 16y ago
That's also one of the biggest issues in software development.
- celoyd 16y ago
  Surely it’s the issue in software development. Everything else is a case or subproblem of “how do I use a computer to solve this imperfectly defined problem?” I mean, it’s borderline tautological, isn’t it? Software development is the practice of translating natural-language requests into executables. That’s not just a recurring question – it’s definitional. This is why I’m nervous when strongly antisocial people say they want to go into software. The point of this field is to figure out what people need (or want, or mean, or like). Of course you can do this creatively rather than responsively, and with more rather than less respect for your customers. But you don’t escape human tastes and concerns when your job is to explain them to a microprocessor.
  - jbri 16y ago
    Well, there's the second issue of doing it such that the machine can produce results efficiently. Which tends rather towards the mathy computer-science side of things - which is where I think the less social "I just want to write code" people would be much happier.
    - celoyd 16y ago
      Efficiency is implicit in most natual-language queries. Managing efficiency tradeoffs is a good example of something fuzzy yet important that you have to be able to figure out from natural-language queries and human interaction. But yeah, if you really just want to work with numbers, you’ll probably be happier in CS academia.
  - l0nwlf 16y ago
    Even though I like your answer I disagree with "This is why I’m nervous when strongly antisocial people say they want to go into software". Strongly anti-social people are some of the best software engineers around. They devote insane amount of time and pretty much enjoy coding.
bluekeybox 16y ago
One would be mistaken to think that this problem is trivial even to us humans.

Luyt 16y ago

A link to the 'print version' of the page: the whole article on one page without the incredible clutter and cruft that the New York Post splashes all over its site: http://www.nypost.com/f/print/news/opinion/opedcolumnists/the_machine_age_tM7xPAv4pI4JslK0M1JtxI http://www.nypost.com/f/print/news/opinion/opedcolumnists/th...

Smerity 16y ago

In this AI/machine learning is compared to the space race and the fight against cancer - I like the comparison considering I work in the field ;) Unfortunately though I think AI and machine learning has the fundamental problem of rolling improvement... Think back a hundred years to a time before the internet, long distance telephones, common air travel and so on. To people from a hundred years ago we'd be the most outlandish science fiction story ever told. I consider the same thing whilst watching science fiction films - in reality they wouldn't be amazed that they can reach Mars in a half hour. They'd instead be complaining about the in-flight entertainment or the turbulence caused by navigating asteroid fields. AI has the same problem. People can now (naively) communicate with each other even though they share no common language. Texts and documents can be re-written in real time to be legible in another language. Credit card fraud and spam emails are handled transparently by systems trained on hundreds or thousands of hours of cumulative knowledge. No-one sees these things. No-one respects these things. Human kind judges itself from the position it currently stands, not from where it began. For this reason alone incremental improvements will never really be considered milestones of achievement.

zeteo 16y ago
"To people from a hundred years ago we'd be the most outlandish science fiction story ever told." You'd be surprised: http://en.wikipedia.org/wiki/Paris_in_the_Twentieth_Century#Predictions http://en.wikipedia.org/wiki/Paris_in_the_Twentieth_Century#...
- derleth 16y ago
  Technology is one thing; society is another: I doubt anyone from the 19th Century could possibly imagine a debate over gay marriage. A crazy promoting it getting beaten down by a righteous world, possibly, but not a debate.
Cushman 16y ago
I don't know if the comparison is that inapt; how many people do you think are really aware of the technology that goes into making that little box say "in five hundred yards, turn left"?

mthomas 16y ago

Perhaps offtopic, but I find it odd that there is an opinion piece by Peter Norvig in the NY Post.

pgbovine 16y ago
true, but i'm sure that NY Post has more subscribers than NY Times, so if his goal was to reach the largest possible audience, perhaps NY Post was an appropriate venue
- necubi 16y ago
  This isn't true. According to Wikipedia, the NY Times has 876,638 daily/1,352,358 Sunday subscribers [0] while the NY Post has 525,000/343,361 [1] [0] http://en.wikipedia.org/wiki/NY_Times http://en.wikipedia.org/wiki/NY_Times [1] http://en.wikipedia.org/wiki/NY_Post http://en.wikipedia.org/wiki/NY_Post
- LostInTheWoods2 16y ago
  Not to be snobbish, but the NY Post is considered a working class paper. Usually bought by those more interested in the funnies, movie times, and local sports, than say international affairs or ramblings about AI.

bmh100 16y ago

This is an excellent article for raising awareness of AI. However, I disagree with its conclusion that we will always relate to AI as tools, pets, advisers, etc. When we ask whether machines can think, I believe it is a question of volition and self-direction. Does the machine have an open-ended goal and is self aware? Can it alter its own programming to change its own methods, its own knowledge, and its own goals? If the answer to these two is "yes," then I believe we will consider the machine to be thinking. What are the implications? If machines gain those capacities, as well as the ability to compute emotions, then what separates them from us? If you agree that humanity is fundamentally a function of our minds, then, if machines can compute in fundamentally equivalent ways, they become "human." What evidence is there that this is forever impossible?

tgflynn 16y ago
If you agree that humanity is fundamentally a function of our minds What about the ability to feel pleasure and pain ? I don't believe a deterministic machine will ever experience such subjective feelings (though it may behave as if it does). Computers may well become sapient but they (at least with current technology) will never be sentient.
- jbri 16y ago
  What is "pleasure" and "pain"? Ultimately, they're just electrical signals triggering the release of certain chemicals in the brain, with subsequent alterations to how the brain processes other stimuli. "It doesn't really feel pain, it just looks like it does" is probably how many people justified atrocities against other humans they viewed as somehow "lesser".
  - tgflynn 16y ago
    Maybe they're just electrical signals in the brain but that's not how we experience them. A CPU cannot experience the electrical signals that flow through it. Unless you believe rocks are sentient. I could just as easily turn your argument around and argue that torturing you is perfectly OK because all I'd be doing is creating electrical signals in your brain.

wooby 16y ago

The episode "The Thinking Machine" of the Nova miniseries "The Machine That Changed the World" is my favorite introduction to the history of AI up until around AI winter. The episode, along with four others, are available here: http://waxy.org/2008/06/the_machine_that_changed_the_world_the_thinking_machine/ http://waxy.org/2008/06/the_machine_that_changed_the_world_t...

Jun8 16y ago

"Learning turned out to be more important than knowing." This seemingly simple sentence, to me, sums up the article, i.e. the data driven approach as opposed to model-driven ones.

mtrn 16y ago

Single page (print) version: http://www.nypost.com/f/print/news/opinion/opedcolumnists/the_machine_age_tM7xPAv4pI4JslK0M1JtxI http://www.nypost.com/f/print/news/opinion/opedcolumnists/th...

TeMPOraL 16y ago

"Spam filtering programs using A.I. learning and classification techniques correctly identify over 99.9% of the 200 billion spam e-mails sent each day." 200 billion spam e-mails each day. Something should be done with spammers just on the basis of how much electricity it wasted on sending, processing and discarding all those e-mails...