Stephen Wolfram - the lovable George Costanza of the mathematics community who developed the invaluable Mathematica suite and wrote the much talked about but quickly forgotten "A New Kind of Science” - is trying his hand at artificial intelligence. His new project, Wolfram Alpha, set to go live in May, combines natural language …
If it does happen
If he or whoever manages to create something that intelligently answers straight questions, learns new constructs and grow its knowledge, they'll make an absolutely obscene fortune.
Or destroy us all.
my money's on...
either our complete destruction OR something like google search, that just never shuts up after you submit your first query... If you close your browser, it send you an email... If you delete your email, it calls your cellular... if you turn off your cellular, it prempts your TV, until you have to file a restraining order against Google.
Wolfram is not one to underestimate
Wolfram has that special kind of crazy that has him going over a problem with methodical and unrelenting detail. His previous books show just how dedicated he can be to a problem and he's one to always produce something really well thought out.
That being said, when I heard he was approaching this problem some years ago I knew that he was one of the few that might be able to pull it off.
Based on the comment he's quoted, it sounds like he didn't nail it. But got well enough to be able to blow people away none-the-less.
Because, if he can deliver on the basic promise of interpreting the meaning of an english statement and search through for content that matches that meaning... that WOULD blow Google away.
I don't know if you've noticed but even Google and Wikipedia searches are inaccurate at best. They're usually something buried in a pile of garbage you can pick out that had something to do with what you were looking for, but if you can show ONLY what you were looking for (based on the context use in a sentence) that would deliver the super giant a potentially fatal blow inside of a year.
For all your skepticism...
...I believe there really is a call for a way of answering certain types of difficult questions. Google and Wikipedia both rely on the question having been asked before, and - crucially - written about. Once you get beyond "What's the lead singer in that band I like called?", the just-Google-it answer starts to fall apart.
Of course, it's such a hard problem to solve that it's highly unlikely this project is the answer he's claiming it to be. So I'm with you on every other point.
If Wolfram gets it to work
He should ask it if any of Ray Kurzweil's predictions has ever been right without a lot of revisionist history.
Hey this guy has automated Wikipedia!
Now which needs does this fulfill is another problem...
On a more serious note, this, erm, "NKS" is not really new, it's been used for ages. Modelisation has been the base of science since long before it was even called science (one could argue that drawing hunting scenes on cave walls was already a kind of modelisation. Religion is a kind of modelisation too). Throwing a metric shitload of computing power at it only moderately helps because it won't solve any problem unless you fully know all the parameters involved, and when you're at that point, you basically already answered the question. Having the machine understand the problem is the least of your concerns: 2 days studying your computing language of choice, sorted (or you can hire a Machine Whisperer).
Natural language for machines might be a good thing (think domotique for example), but it has only very limited applications in science. It's even counter-productive, as scientists love nothing more than being cryptic. And studying simple fairytale models (like the game of life) won't lead you anywhere scientifically speaking because these "models" have no factual basis. Given 1+1=3, you can prove that 1+1+1+1=6 (if you assume that 3+3=6 of course) but it won't bring you anywhere in real life.
It's basically the same problem as with climate simulation. Lots of models, none agreeing with the others, and none able to reproduce real life evolution (well, all the ones old enough to be confronted with real data failed, at least). That's because we still fail to grasp all the parameters involved.
If a supercomputer is able to understand natural language and can aggregate and parse all knowledge, it will merely allow the layman to get already-known facts or beliefs. It cannot lead to any genuine discovery. Automated Wikipedia, really. Or a dumbed down automated scientist (scientists are raised to do precisely that, plus hopefully some "out of the box" thinking -though that last part tends to disappear. I should know...).
Now I guess the real questions are: will this new-fangled system be cheaper or more expensive than training a flesh-and-bones scientist? And will it be able to sustain itself and produce new useful knowledge in the absence of scientists? If answer to both question is yes then YAY! Retirement time!
My belief however is that it would be cheaper and more efficient to get scientists to speak natural language... probably a bit more difficult though.
Mine is the white cotton one with the pens and timer in the chest pochet, thanks.
TL;DR: I was bored.
I love it when someone has criticized New Kind of Science
but isn't man enough to read it. Anything bigger than 140 characters is probably beyond your ability. If you had read it, and you knew anything about CA, you would know that the solutions space of CA is not fully explored because of limits in computational speed. Every year new things can be attempted that couldn't previously. Further, you would understand that neurons seem to behave like CA and much of what humans can do is an emergent behavior from simple small scale rules. Except for you perhaps, who has no emergent behavior and will remain perpetually small. That someone is willing to attempt to apply the CA pattern to the knowledge problem is laudable. Even better, that person is Wolfram who's brain makes yours look like a gnat's. Ted, you are a sham, you have no right to comment on any topic in software development, much less something that is clearly beyond your experience.
Wolfram is one of the world's biggest jerks!
I personally worked at Wolfram Research in Champaign IL. Not only does Stephen Wolfram believe he's one of the most brilliant humans ever birthed, but he and his sweat-stinking idiot brother, Conrad, are two of the biggest a**sholes I've ever met, anywhere! You have never met anyone in your life who is so arrogant, and who takes such delight in berating and belittling others.
He owns several houses here and there, one around Chicago. The locations of which were always supposed to be top secret! On occasion he'd drop into the offices without warning and change entire project plans and staffing arrangements (including canning people) on pure whim. The advice I was given on day one was, "stay off of Stephen's radar at all costs, you will only lose."
I could tell you hours of horror stories about his personal life that would turn your stomach as well. Like, one of his children was left wearing a gaping, leaking diaper long after any other child would've been toilet trained. Which the boy liked to do without, occasionally dumping like a cow right on the carpet. Because as always, Stephen had some insane rationalization about why this was perfectly acceptable.
Most of the best stories came from the fleet of personal staff he's constantly churning through - from the several nannies, to the personal assistants technical and non-technical, all used for target practice. Pretty high rate of dissatisfaction and burn-out there, you can imagine.
Stephen believes he is the smartest man in the world in every area of knowledge, from nutrition to geology to alchemy. But I don't care if the guy figures out how to turn mud into gold, he's a blight on humanity in every other way. Had my position put me into more frequent personal contact with the tumor, I would've poisoned him myself.
Stephen Wolfram is a genuinely loathsome c*nt.
Hey, I'm backing Wolfram
With any luck, he'll be able to create a machine that can write better articles than Dziuba.
AC Posted Tuesday 17th March 2009 03:34 GMT
You're drunk, mon.
Where be my 5 pounds?
"Correct me if I am wrong, and I know you will, but there was a programming language in 1972 called Prolog that could take carefully curated declarative statements and allow you to run logical queries over them."
1972? Surely you mean INTERCAL?
@ac and others
"...isn't man enough to read it..."
Well, I haven't read it because I have no evidence that this guy has actually solved any useful problem. Where have his wonderful ideas been applied? I would genuinely like to know. None of this "new models" crap. I want to see results/advances in a field. Yes, you can calibrate the parameters of cellular automata networks to produce whatever arbitrary results you want (much like game theory analysis of the Cuban Missile Crisis, which one of my profs tried to sell us on). But that is just guess and fix-up, or proof of concept. What can Wolfram actually DO?
Science is about discovery
No, computers haven't solved this problem because there are no people who actually need it solved.
You got that backwards:
Science and Research provides the foundation and the environment for product development, NOT the other way around. Once a problem is solved, the need for many ways of applying the solution to a "Real World Problem (tm)" will quickly follow.
so given that the internet is just a fad,. will never catch on and is only used for Porn, I also note that Wolfram doesn't really, slide off the tongue, I sense a rebranding! Deep Thought? nah! too "thinky" I kknow, deep throat! perfect, now to was this google outta my hair!
>>>"When was Google's stock at $300 per share?" or "How much did it snow in New England last year?”
It's a noble goal, to aggregate human knowledge in a large machine brain that's able to answer questions. The problem is: We already have Wikipedia and Google that - together - get the job done well enough.<<
Er, neither of those questions are simply answered by either Google or Wikipedia. You get a load of links on Google, which you then have to trawl through to find the right information. And do you trust Wikipedias answer, when you actually find it?
So, sort of an own goal there, Ted.
They don't get the job done well enough at all.
@criticism of New Kind of Science
I love it when somebody criticises an article and it's author, but isn't man enough to put his/her own name behind the criticism. Yes, I'm looking at you, AC@03:34GMT. Although I will give you the benefit of the doubt in that you're probably just back from the pub, only excuse I can think of for being online at half-past 3 in the morning. Either that or you're just very, very sad. Oh, wait...
This article is more reasoned than you realise, and Ted's clearly done some research. Credit where credit's due.
And, no swearing - excellent!
well if it really worked...
...Google would just throw some readies at him and acquire the technology, then hide it somewhere in Google Labs.
"No, computers haven't solved this problem because there are no people who actually need it solved."
No, it hasn't been solved because nobody yet has come remotely close to understanding how natural intelligence works. Isaac Asimov had his mysterious positronic brain leaving the fundamental details
Speaking as somebody who worked a bit on expert systems and did program in Prolog, it became blindingly obvious that the so-called AI experts really didn't have a clue. I recall attending one lecture where an academic theorised on the moral problems of how to treat a machine which would have a brain equivalent to that of a "slightly brain damaged child", which he expected to have in a few years (this was 15 or more years ago). There were two types of AI people - the computer scientists, who somehow seemed to think that human intelligence could be treated like mathematical logic, and the cognitive psychologists who really hadn't the faintest idea how to come up with a coherent theory of the mind (didn't stop them thinking they were about to).
I feel we aren't much further forwards now. We can just program faster idiots.
Nb. where expert systems could be most useful - like brining to bear statistics on medical diagnostics, vested interests tended to block progress. Maybe because many of these much vaunted high level skills were shown not to be (and some of the practitioners weren't so good at it). However, present a computer program with a novel situation and it's lost.
Now that I have learnt to 'google' using the current engine..
eg 'Car X Statistic is'... Rather than: 'What is the Car X statistic?'
I Doubt I'll be wanting to use a new technique all over again when I am so good at finding what I need now...I hope this Wolfram alpha is backward compatible with these search methods or it will fail.
I read Wolfram's NKS book once on a flight from London to Manchester and all in all I found it rather boring. The storyline is flawed and most of the characters are shallow. That's the last time I am buying the Reader's Digest again.
@'I love it...'
You were doing fine right up to the point at which you descended into personal insult (about the 13th word or so) and then you just read like an overly angry man trying to prop up a spurious argument by insulting the opposition. Get your head out of Wolfram's arse and try and debate the subject in a less inflammatory manner. I for one would appreciate it, it might just give me an insight into the subject that I didn't have before as opposed to another look at all to familiar trolling behaviour.
Oh and BTW, not that I know anything about this subject but I'm pretty sure neurons were behaving the way they do a long time before CA was on the scene.
Smile, it's not the end of the world you know.
I have the answer...
There you go. Beat you to it, Wolfie
Even I'm smart enough...
to know that Wolfram is smarter than Ted.
When Eddington towards the end of his career started trying to link up the numbers in physics together in the same way as people try and link the height of the Great Pyramid to the distance between the Earth and the Sun, one student listenign to his lecture asked another "Do all great physicists go crazy at the end?". The other said "Maybe, but guys like you just get dumber all the time."
@Anonymous Coward (03:34)
Now that you're here Stephen, I'd like to ask you a question: why is the output from Mathematica so bloody ugly, and why is it so expensive. If you love science/maths so much why don't you gift us scientists/mathematicians with lower prices!!
Why do I get visions of a silicon two year old who just keeps asking 'but why?' to every answer you give?
Paris because that is what underdisciplined kids grow up to be.
Name for this class of system
If this works as well as we could hope, could I suggest that we coin the phrase "anti-politician” to describe this class of system.
Because it works more than a few minutes at a time, takes time to research its answers, gives answers in plain English, doesn’t take bribes or get involved in nepotism, and most importantly doesn’t lie constantly.
But does it work?
All very interesting, Ted, but your tirade of strawman arguments won't actually affect whether or not Wolfram's stuff works, will it?
If Wolfram has an ego what does Ted have? It seems the height of journalistic hubris to think that a position of extreme cynicism can illuminate anything but itself.
(And yes, Ted, I'm sure you really have worked in all the areas of IT you claim to have. You don't sound AT ALL like a bitter one-trick pony who's now so irrelevant he has to write about IT instead of doing it. So I'm sure you know MUCH more about this subject than the guy who developed Mathematica from scratch.)
Looking forward to seeing what Wolfram comes up with. Even if it's only another book, it'll be well worth reading.
Google and Wikipedia are 'good enough' ?
We'll solve all our problems with that sort of thinking!
My cat's more intelligent than any computer
Ok it cant calculate VAT or anything.
But . . .
. . . we are learning how to reduce the number of words in a search to just those that have meaning. Stripping back a phrase to two or three words makes searching easier - you don't have bung in a long sentence of where, what, how, who, why etc.
Most people are not entirely sure what the question should be and get pissed off at getting the 'wrong' results.
Regional variations in English can also have different meanings for similar words or phrases. I'll wait to see whose version of English is used to produce the mother* of all search engines.
* we will also find out if another word is to be inserted here.
Isn't this exactly what http://www.trueknowledge.com/ is trying to achieve?
It is a Ted article
that means think the opposite and profit.
Hmm Ted has watered down his cynicism this time round, but I think there is still a gleam there.
So, looks like this might be a goer.
I tell you when my snake oil detector turns on, it is when people wish to use the human brain as a model for AI, and the detector goes off when someone tries to keep it to mathematics.
English is a really bad language for logic, German would be simpler, but obviously lacks the mass appeal. I think a linguist, a mathematician, and a programmer could probably get this cracked over time. Time to wheel out Larry Wall.
What if there is no answer to a question?
Perhaps it could make one up and then put it in wikipedia so that it is true.
May I be the first to say...
I for one welcome our new skynet overlord of the search engine domain....
@I love it when someone has criticized New Kind of Science
I'd like to welcome Mr Wolfram to these pages ...
I'd recommend scaling back the personal attacks, though, once your chin has stopped wobbling - just a thought.
All this reminds me of one of the arguments against AI put forward the last time it fell out of favour and stopped getting funding: you don't need AI because there is another way of producing intelligent (human) beings that's much more fun. Hence Paris ...
AC at 03:34 is Wolfram
It's the repeated use of "emergent" that gives him away.
Given 1+1=3 ,,,
Sorry, that doesn't add up.
0 = 0
1 = 1
10 = 2
11 = 3 (OK so far)
100 = 4
101 = 5
110 = 6 (starting to look dubious)
1110 = 14
1111 = *15* not 6 !
Wow, Ted, did Wolfram sleep with your GF or something? Your article reads like a personal attack rather than serious journalism.
What the guy is proposing would be the biggest change in how computers are used since the internet. Have you even considered what the implications of this are? Steve Jones (above) mentions some good uses of expert systems, but on a more everyday level, imagine you're re-naming files and your computer could see what you're doing, analyse the criteria you're using and offer to complete the task for you. Or you're sorting photos into folders, or a hundred other tedious tasks that an AI of even very limited intelligence could do for you. Imagine being able to tell your computer "I heard a new tune on the radio this morning - find out what it was and download it to my phone." You know - the kind of computer 80s sci fi thought we'd all have by now. The kind of work Wolfram is doing is moving us closer to that.
Whether he achieves his goal or not, the fact that he is thinking about these questions at all (rather than just flailing at anything he doesn't understand) automatically puts his intelligence an order of magnitude beyond yours. The fact that he created mathematica is an indication that he may have a shot at actually achieving it. But whether he does or not, I bet every morning he thanks his lucks stars that he didn't turn out to be a sour and embittered hack.
Have a nice day :-)
"I'll wait to see whose version of English is used to produce the mother of all search engines."
Clearly if this is done it'll be installed on a starship called Nostromo and then go off and discover an alien lifeform hellbent on wiping out the human race?
".. a rare blend of monster raving egomania and utter batshit insanity"
Cosma Rohilla Shalizi on S.Wolfram, A new kind of science
Anything that explicitly has "Science" in its name
Anything that explicitly has "Science" in its name - isn't a Science.
Sciences: Physics, Astronomy, Chemistry, Biology, etc.
Not Sciences: Social Science, Political Science, Computer Science, etc.
"The problem is: We already have Wikipedia and Google that - together - get the job done well enough. "
I assume Ted's from a parallel world where this is true. In our reality, sadly, WP is worse than useless and Google's ability to answer such questions is poor and getting worse each year.
Not that Wolfram is going to solve it. Having read "New Kind of Science" I feel safe in saying that he's had his one great idea and there's nothing left in the bag. NKoS was a waste of trees, as well as containing some very self-serving notions about the history of cellular automata.
Nice idea, but...
Well, Mathematica is certainly a very capable and well-marketed product, but there are other symbolic computation packages that are just as capable in their own ways, if less well known. I don't think we can credit Wolfram with a unique insight here.
Then there's NKS, a subject I was very interested in - until I read the book. Then I started to understand why so may others decry CA as a "solution to everything". Sure, fascinating patterns and loads of speculation, but nothing you could actually get a grip on and say "yes, that's important", or "I could use that".
Writing down some simple rules and watching useful behaviour "emerge" sounds great (and not just in AI, but in physics and biology and probably many other fields too), but what are the rules exactly? Just because they're allegedly simple doesn't mean there aren't far too many possibilities to search through. That's where the insight is needed. If I had to criticise NKS on just one point, it's that it promised this insight and failed to deliver.
So while I'd really like to believe someone has finally cracked AI, I'm not overly optimistic. I think the only chance is that Wolfram really has some unique insight into the potential of CA that he completely failed to communicate in NKS and has now perfected. Given his personality, however, I rather expect the hype will outshine the achievement.
Let's also not forget that simulating human thought even in one area of knowledge is one (very difficult) thing. Turning that into a search facility that's expert in every field of human endeavour and can handle hundreds of thousands of simultaneous queries is quite another thing. I wouldn't be selling my Google shares just yet (if I had any, that is).
"We already have Wikipedia and Google that - together - get the job done well enough."
That doesn't sound like the official El Reg opinion about WikiPaedo, or indeed the great chocolate factory
Ego only gets you so far...
I personally like Mathematica (though use MATLAB more), but found Wolfram's theorizing to be a pretentious and utterly simplistic view of the natural world (the parts I managed to get through, that is).
That said, applying AI or expert systems to develop slightly more intelligent web-based queries could actually work, but might get off to a faster start working with Google or the like, rather than starting from scratch.
Steven Jones @ 9:02, agree completely re. medical diagnosis, at least at a preliminary stage. The concept that we often rely on tired, overworked, variably competent individuals to piece together - sometimes from distant memory - disparate symptoms, analyses and measurements to diagnose complex life-threatening problems is absurd.
AC @ 12:19, interesting point and maybe why I've always preferred geology, geochemistry and geophysics to the awkward "Earth Sciences" beloved by fashionable universities.
"'A New Kind of Science' (abbreviated NKS for those of you who prefer the Church of Scientology method of using acronyms to make yourself sound more serious)"
MUSA* a peculiarity of the CoS**? TNTM***.
* Making Up Stupid Acronyms
(and, incidentally: Made in the United States of America)
** Church of Scientology
*** That's news to me
I see problems
Not with the execution, which will probably astound us all, or with the appeal as I notice enough people type straight questions into Google as is, no the cancer that this project will succumb to is best illustrated with a question: How is babby formed?
Simply put, any Machine Intelligence that learns from questions posed by the yahoo answers mob & the chantards will eventually attain all the smarts and delivery of a demented LOLcat.
....now what exactly was the question ?
"Er, neither of those questions are simply answered by either Google or Wikipedia. You get a load of links on Google, which you then have to trawl through to find the right information."
On Tuesday, March 10th, 2009.
Search for 'GOOG', click 'GOOG', drag the slider on the stock ticker back until the line crosses $300.
I'd say that's pretty simple.
"Perhaps it could make one up and then put it in wikipedia so that it is true."
Now that's what I call artificial intelligence.