Slashdot Mirror
Wolfram Promises Computing That Answers Questions
An anonymous reader writes "Computer scientist Stephen Wolfram feels that he has put together at least the initial version of a computer that actually answers factual questions, a la Star Trek's ship computers. His version will be found on their Web-based application, Wolfram Alpha. What does this mean? Well, instead of returning links to pages that may (or may not) contain the answer to your questions, Wolfram will respond with the actual answer. Just imagine typing in 'How many bones are in the human body?' and getting the answer." Right now, though the search entry field is in place, Alpha is not yet generally available -- only "to a few select individuals."
Q: How many bones are in the human body
A: Did you mean cumulatively or at any point in time?
Sig Battery depleted. Reverting to safe mode.
package com.wolfram;
public class Alpha {
public static void main(String[] args) {
System.out.println("42");
}
}
Been there, done that.
All that is old is new again.
God, schmod. I want my monkey man!
Google already does this. Type a question like "What is one plus one?" and you will get an answer. It's artificial intelligence.
I don't think this can be examined without language issues. Lojban attempts to make a parsable constructed language (currently undergoing a few grammar issues, but mostly locked down). As we get closer to the Singularity, with regards to infant-style general AI and perhaps even transhuman implants (thought detector or such), we'll see perhaps a myriad of unambiguous languages.
Your cautiousness and pragmatism in the first two sentences was noted and admired. Then you used the word Singularity in the Vinge sense, and my woo-detector pegged.
I don't think this can be examined without language issues. Lojban attempts to make a parsable constructed language (currently undergoing a few grammar issues, but mostly locked down). As we get closer to the Singularity, with regards to infant-style general AI and perhaps even transhuman implants (thought detector or such), we'll see perhaps a myriad of unambiguous languages.
Any language that is truly unambiguous is uninteresting. Firstly, you've got Goedel incompleteness to worry about (which stems from statements that are fundamentally ambiguous as to their interpretation, such as "this statement is false"). Secondly, languages are there for people to communicate with, and people seem to prefer ambiguity. Ask a poet if you need proof of that.
"Little does he know, but there is no 'I' in 'Idiot'!"
Either:
1. Windows version of program crashes without answering
2. Mac version of program says "after your next question, smartass"
3. Linux version of program says never, 'cos it can't even drive a car
Those who can make you believe absurdities can make you commit atrocities. - Voltaire
...they only give you answers.
is the answer to this question "no"?
If you want to answer a question without understanding the question then how do you know when the question can be answered?
Some drink at the fountain of knowledge. Others just gargle.
Actually, the original source, TechCrunch, not the dumbed down linked article, discusses in much better detail what Alpha is about.
Good points, but this is still just a different (better perhaps?) implementation of the same concept. The big issue with the implementation is that it will only "know" what you tell it, the same as any other computer. Further it will only be able to tell you about what you want to know based on the system's ability to parse your question and return what it "thinks" you want to know.
Look, I'm not saying it isn't a cool idea, I'm just saying that it isn't as shiny and new as the creator would lead you to believe. I'm also not inclined to be impressed considering that it isn't even available to try yet. It hasn't even been released yet.
God, schmod. I want my monkey man!
As long as they are not showing the tool to the public, I do not believe they build a system which promises that. However, there have been lots of research in this area and there are methods to convert queries into horn-clauses so you can query knowledge bases. I designed a method in my master thesis which does similar things, however it was laid out to be performed by humans.
As ingredients for such a system you need
- a knowledge base filled with facts (you can use OWL for it if you want or a rule based approach)
- a reasoner (e.g. something like pellet)
- a rule engine (e.g. something like Jess)
- a method which understands simple English query sentences.
The really hard part is the knowledge base, because it is lots of work. And an automated approach which can understand written documents and classify them correctly would be great, but I doubt that they found a solution for this problem.
This problem includes:
- How to handle uncertainty?
- What to do with contradicting knowledge?
- What to do with temporal aspects in that knowledge?
However, if they built a tool which can answer question of one single domain of knowledge, this is nothing new. Such machines exist now for a long time. They can be helpful, but there is nothing exciting about them.
You look for an answer until you find it or give up.
Oh, so (i) you don't understand the question, then (ii) you look for an answer, (iii, A) you find it (how? how will you know you found it?), or (iii, B) you give up.
Just wanted to make sure that this thread was really about this. Here's a new low, even for slashdot.
Are you limited to yes/no answers?
Why are people presuming that the program will be limited to yes/no answers?
Q: Will you answer no to this question?
A: It's rather unlikely.
(Or, "I doubt it" or any of several different answers.)
There are enough legitimate paradoxes that you don't need to construct such obvious losers.
How about:
Is "This statement is false." false?
It's still easy enough to handle (in several different ways), but at least it's a valid challenge.
I think we've pushed this "anyone can grow up to be president" thing too far.
To my mind, any reasonable definition of understanding a subject includes the ability to reason based on information about the subject. In the case of a question, this would include the ability to say, at the very least, whether a given answer is a correct answer for the question.
From this, we can see that if we can build a reasoning engine that can determine if a given answer is correct for a question, hypothetically we can iterate over a large set of answers and apply our filter to each one. This provides us with a machine to answer questions (although depending on the size of the set of answers, "I don't know" might be a frequent response) which (by my definition, at least) 'understands' the question.
Rampant carbon sequestration destroyed the Dinosaurs' tropical paradise. I'm here to help repair the damage.
three, sir
Check out my sysadmin blog!
I'm not sure I really want to trust a product by Wolfram and Heart. Seems like there is a possibility of some soul loss.
"You should always go to other people's funerals; otherwise, they won't come to yours." -- Yogi Berra
I'd be impressed if it could answer "Could you explain your previous answer using a car analogy?"
The Internet is generally stupid
Don't sweat it ... these are the same people who believe that a computer can "think" about chess, instead of just searching through N number of plies in T time, then offering the best solution it has found in those constraints, without ever having to "understand" chess on any level.
This can be applied to ANY problem, provided you want to invest the design and testing time.
This whole question was answered decades ago (1970s) with the "foreigner in a sealed room" turing thought experiment. It showed that the person in the sealed room doesn't have to understand english, or even know the answer to questions, provided they are given some simple rules to link words together in a response depending on what words are in the original statement.
Who says that this is insufficient for "thinking"?
I think understanding the Chinese room paradox as having provided a solution to this question is a misinterpretation. The best thing to take away is that "thinking" is not well defined.
You look for an answer until you find it or give up.
Oh, so (i) you don't understand the question, then (ii) you look for an answer, (iii, A) you find it (how? how will you know you found it?)
Once you have found the answer only then will you understand the question, grasshooper.
Also, until you can claim to solve the halting problem in real life (as opposed to a "theoretical device"), don't go around claiming that the brain is turing-complete. It isn't, and cannot be - not in this universe, anyway.
Of course the brain is turing complete. You can prove it the same way you prove any other machine is turing complete: it has the ability to simulate a turing machine. I can simulate a tape driven turing machine pretty damn easily with a sheet of paper and a pencil. I think you're confused as to what "turing-complete" means. Solving the halting problem is not a requirement. In fact, you can prove that a turing machine cannot solve the halting problem. So the brain's inability to do so doesn't have any bearing on whether it's turing complete.
The Chinese Room is misdirection, pure and simple. We're supposed to conclude that because the person in the room doesn't have the subjective experience of understanding Chinese, the system as a whole (the person, the data tables, the rules) doesn't "really" understand Chinese.
But there's no logical reason to assume a specific part of the system should have a subjective experience of understanding something that the system as a whole understands. This becomes obvious if you follow the logic a few more steps. Do you believe each specific part of your brain subjectively experiences understanding? How about individual neurons? How about the atoms that comprise the neurons in your brain? If you don't believe these things have the subjective experience of understanding the things that your brain as a whole understands, then your brain is incapable of "really" understanding anything, according to the logic of the Chinese Room.
This space unintentionally left unblank.
I have always hated Searle's Chinese room "paradox", since it is just playing a semantic game with the definition of the system. The claim that the person in the room doesn't understand things is no different from saying that a neuron doesn't understand things, or that 1/4 of my brain alone doesn't understand things. The "rules" in the box are part of the system, and I would claim that if it passes the test, the person+rules do demonstrate understanding. We have no evidence that human thought somehow transcends the model of executed rules anyway; at some level it is all chemistry and physics.
A modern example would be that my CPU (::person in box) doesn't know how to behave as a web browser. While true, my computer does know how to be a web browser when you add the software (::rules), and an input and output system (::box interface). The Chinese room paradox is just yanking out the CPU and saying that it doesn't know how to be a web browser. Nice trick.
The other thing the "paradox" does it to try to evoke imagery of a very simple ruleset because it is a person executing rules on paper, which would be very slow. The person executing paper rules is slow enough to have the computational power of a few neurons at best, while the brain has ~100 billion. So the equivalent rules in the paradox's imagined transformation would never fit in a room and could not be executed to completion by a person before their death. While it is supposed to be a thought experiment, the relative scale is so incredibly different that it makes imagining it difficult, and I wonder if it was chosen for that purpose. I will cut Searle some slack though, since Turing's guess about how much computing power needed to pass the Turing test was ridiculously low (~50 MB of storage), when compared to what we now know of human brain capabilities.
I think the appeal of the paradox is that deep down many people want to believe that we are qualitatively different from computers, rather than quantitatively so. As for me, I'm happy enough knowing that atop my shoulders sits a computer with more raw processing power than the largest supercomputer, with rules/programming far beyond anything we can create now, or perhaps for hundreds of years.
> Also, until you can claim to solve the halting problem in real life (as opposed to a "theoretical device"), don't go around claiming that the brain is turing-complete. It isn't, and cannot be - not in this universe, anyway.
The halting problem is undecidable over Turing machines. Claiming 'the brain is not turing-complete because it cannot solve the halting problem' makes no sense.