The Holy Grail to the Knowledge of Everything: Wolfram/Alpha
The Holy Grail of the Knowledge of Everything
(Here is the original TED Video: Stephen Wolfram: Computing a theory of everything)
There is a rule, maybe in life but especially in physics (I was taught this in physics at University of Toronto):
if it is too good to be true, it probably is.
Those who took physics will understand the underlying meaning of this: a solution to a physics problem is likely more complex than it reveals itself to be.
What does any of this have to do with enterprise IT, IT solutions, convergence, knowledge management, cloud computing, or any of the hot topics for this industry?
Knowledge, and the pursuit of it, has everything to do with it.
Further, the idea of making systematic knowledge computable is no longer a concept. It is here in a tool called Wolfram | Alpha.
The creator behind this is Stephen Wolfram. He is the author of “A new Kind of Science,” a Ph.D, the creator of Mathematica (1991), and last year released (as described by CNN).
Wolfram|Alpha, “a knowledge engine that answers users’ questions on the Web by computing answers in real time with the help of a vast collection of databases.”
Some facts on Wolfram | Alpha:
- Some of the abstract intellectual ideas were turned into millions of lines of code
- The system represents terabytes of data
- 10,000 servers were assembled
- At the time of launch last year, the site had networking and load-balancing challenges
- At 9:33:50 p.m. central time, May 15, 2009, site went live
What makes this system so great?
Ask WolframAlpha “how much wood could a woodchuck chuck?”
It gives the right answer.
How, and why are any of the IT systems we interact with (internal, vendor-created, or out of the box) not able to understand questions in plain language? Will this mean that SQL developers and Report developers will be out of a job? Will the role of a business analyst in translating technical requirements to business requirements change? These questions become too basic. The reason is the power for Wolfram/Alpha could very well be to predict things.
Wolfram explains that plain language queries were possible because:
1) A bunch of new ideas on linguistics set about by studying the computational universe
2) Having actual computable knowledge changes how one can set about to seek knowledge
Is there any system out there with information (as opposed to knowledge) that is capable of making predictions and forecasts?
Wolfram’s view for the system is to give a tool to users for providing answers for things that have not been asked before. This is different from getting answers that have already been written.
By having Wolfram/Alpha on top of Mathematica, it is possible to get information based on real-world data (in the computational universe).
The dual application effectively democratises computing and computation.
Wolfram then asks about the rules of the universe. What if the universe is not as complex as suggested by physics, and that it operated under low-level abstract ideas (lower than space-time)?
This product (or rather, project) represents 30 years of Wolfram’s work. It is truly phenomenal. It is a project, because it is still under development. Wolfram hopes that within 10 years, Wolfram/Alpha will be able to answer and challenge our fundamental understanding of the universe.
April 28, 2010 - 9:14 pm
Answer and challenge our fundamental understanding of the universe? I don’t think so. Whatever Wolfram/Alpha does, it does not create new knowledge; it just applies previously existing knowledge, algorithmized and turned into code by its creators. In other words, it does not know anything that its creator would not already know. (Incidentally, I was trained in computer science with specialization in artificial intelligence, so I am quite sceptical about all the nice-sounding promises). See also “chinese room argument” http://en.wikipedia.org/wiki/Chinese_room
April 28, 2010 - 9:37 pm
I did read up on chinese room argument before bringing up that outrageous point made by Wolfram. After all, the writing begins with the idea that anything too good to be true probably is. But this project brings together linguistics, math, and knowledge.
Note that the fundamental understanding of the universe does not fall under the idea of spirtual or to artificial intelligence. Wolfram is only referring to describing the universe in the computational universe.
This goal has the sound of being a vision. A company could really appreciate the power of having a vision (statement), just as the excitement expressed by Mr. Mathematica for a project having such a grand vision.
April 29, 2010 - 1:17 am
But this project brings together linguistics, math, and knowledge
—
So, what’s new in all that? The first attempts to formalize linguistics date back probably to Chomsky, if not earlier, and formalizing knowledge is even older. What exactly did Wolfram invent? A faster access to existing data and an easier way to match them together, I agree. That’s great. But, as far as I can see, NO NEW KNOWLEDGE IS CREATED BY THE SYSTEM. Buzzwords like “computational universe” do not explain anything, they are only convincing to people who know nothing about computer science.
“fundamental understanding of the universe does not fall under the idea of spiritual or to artificial intelligence” – I profoundly disagree. Understanding means discovery of new laws if there are any, and this requires intelligence, but the machine cannot discover – it can only apply anything that was not put in it beforehand by its creator. Therefore, for a totally new problem, this system would be helpless.
April 29, 2010 - 1:18 am
Sorry, “it can only apply anything that was not put in it beforehand by its creator” should read “it cannot apply anything that was not put in it beforehand by its creator”
April 29, 2010 - 2:29 pm
Re: Searles Chinese Room Argument – I wrote and presented a paper at Worldcomp06 (International Conference on Artificial Intelligence (ICAI-06)]- Las Vegas 2006. I tried assuming that Searles assertions were true – then working backward to see how I might construct the room to keep the assertions true – it turns out to be quite difficult:
http://cyberspace-industries-2000.com/pub/Worldcomp06/Understanding%20Instruction%20Books%20and%20Programs.pdf
Re: Can a computational system create new knowledge? Hmmm – might depend on what we mean by new knowledge. Einstein took the experimental fact of the non-existence of “the ether” – i.e. the speed of light seems to be a constant regardless of the speed of the light source or the receiver – and managed to create a “theory of gravity”. Was this new knowledge or merely something hiding in the mass of data we already had waiting for someone to see it? Curious that Einstein couldn’t even believe his own equations and introduced a cosmological constant to counter the “expansion of the universe” his equations predicted – which everyone thought was not true at the time – but was proved to be so at a later time.
The universe appears to be a computational engine – and so far we humans appear to be the only entities in that universe that have the property of sufficient intelligence to process that information. Buckminister Fuller believed our role in the universe was as information harvestors: See my Facebook group http://www.facebook.com/topic.php?uid=8334477439&topic=4306
Quantum theory seems to point to a strange connection between our material universe and it’s “observation” – something John Wheeler called “It from Bit”.
So far tools like Wolfram Alpha do seem to be only able to regurgitate the information fed in…. but if you look a little more closely, you find it pops up stuff that you could not begin to imagine are related – again based only on previous knowledge – which might spark in a human a new way of looking at something and coming up with something completely new – again always based on what we already know.
Now if you believe Ray Kurweil’s prediction in “The Singularity is Near” that by 2045 there will be a second intelligence along side us humans – and that will be a non-biological intelligence a billion times more powerful than all human intelligence today – then what?
It seems clear to me that even such a piece of intelligence as that could only utilize existing known facts to deduce new patterns and information – just like we can – only millions of times faster, processing data at speeds we can’t come close to matching.
Still – we have “rule 30″ – same algorithm as all the rest – different parameters – but a human recognized it as something different and special. Will that new intelligence have that same capability as part of it’s “intelligence”? By then, I will either be 103 or dead so I will probably never find out the answer to that.
Very interesting topic.
April 30, 2010 - 10:13 am
42
Suppose we can build a machine that can answer any question. How do we know we are asking the right question? How do we know the data being used is correct? How do we know the computational algorithm applies in all cases? The selection of the algorithm is based on human intelligence which is limited. We cannot build a comprehensive QA test set for programs of medium complexity, how can we know that increasingly complex programs are actually correct? We can’t, hence the desire to find simple ‘rules’ that can generate apparently complex systems (emphasis on apparently). The system seems non-deterministic and seems to be evolving. The TED chair pointed out the comparison with fractals which create beautiful, apparently random patterns from simple algorithms. I’m not a mathematician, but when we needed to use polynomials encoded into to logic to create random patterns for structural test coverage of semiconductors (because you could never create enough functional test, like QA tests), the mathematicians reminded us that they were pseudo-random, not truly random, they only appeared to be random to a human because of our limited processing ability, even though we have the best pattern recognition machine in the universe between our ears.
Recognizing that Wolfram’s machine is still Alpha (does anyone know when the beta is expected, is that in 10 years or 2045?) the answer to the question “how much wood …” asked of the “knowledge engine” was not computed, but simply quoted from another book, doing no more than a Google search.
We all use Wikipedia. We all know that the data is not always accurate, but most often it is good enough for information that we quickly need at the moment. So what is good enough?
So why not experiment with Wolfram alpha yourself, and see if it can answer simple questions of interest to IT. I tried it, it did not know what a teraflop was. That is the measure used by the Top500 supercomputer list that is published every 6 months. It could tell me that a teraflop was a 1000 gigaflops (I’m sure Homer Simpson knows that). As I poked around for a while, it eventual found the answer of what a flop was (that was curious, why did it not recognize it at the first question). I’d like to know if there is a correlation with a nation’s GDP and the amount of teraflops a country has with supercomputers on the Top500 list. An interesting IT question – Wolfram Alpha did not “understand” that question, I got a chart, but it did not make any sense.
All this to say that there is a simple rule for the “computational universe” ; garbage in = garbage out
May 5, 2010 - 8:22 pm
Hugh, Tania, Rob -
your detailed perspective was much appreciated.
One thing I didn’t get into about Wolfram results is data presentation. Assuming the data is reliable and correct, the display and presentation of data can also offer insight. If other statistical viewpoints were offered by Wolfram (correlation, trends, for example), that may be of value.
But a system that offers a statistical viewpoint? Is that possible in the future?
Knowledge is not “created” but fairly important decisions can be made based on the way data is presented.
May 6, 2010 - 12:14 am
fairly important decisions can be made based on the …
—
Made by whom? If by people, with the aid of a computer, that’s nothing new (decision support systems do exist already). If by a computer, you will have to prove to me first that a computer can create new rules for these decisions, such that were not previously put into it by its creator. Formal logic had been invented a long time ago, and computer systems that use it to make decisions are nothing new, they are called expert systems and they have been around since 1960s. So far, I don’t have a clear explanation of what exactly is new in Wolfram/Alpha.