More

Featuring fresh takes and real-time analysis from HuffPost's signature lineup of contributors
John Maeda

John Maeda

 

Watson Is No Match for Humanity

Posted: 02/23/11 02:58 PM ET

The Watson craze last week didn't fully hit me until my cab driver got lost and cheerily exclaimed in thickly-accented English, "Watson! Heeeeelp me!" I find it interesting how the so-called "artificial intelligence" (AI) systems I studied decades ago at MIT are on their way to becoming the Fonzies (Watson can tell you who that is) of our times. There are a few misconceptions about our "new overlord" that I attempted to clarify within the confines of my taxi ride lost in a suburb of DC. Here they are:

1/ The computer is smart as us, and dumb as us. When Watson slipped up with the Oreo/crossword puzzle answer of "19-teens" it was our fault for not teaching Watson what that means. And if you do a Web search for "19-teens" it's brutally clear that the invention of "Oreos" or other innocent games doesn't come first to mind in the darkness of the online world.

2/ The computer never makes mistakes -- or the same mistake over and over -- unless we let it do so. If left alone, like the proverbial broken record, a computer will do the exact same thing it has always done. There is a construct in computer programming called "the infinite loop" which enables a computer to do what no other physical machine can do -- to operate in perpetuity without tiring. In the same way it doesn't know exhaustion, it doesn't know when it's wrong and it can keep doing the wrong thing over and over without tiring.

3/ The computer still needs us to make the right decision. That little exercise you do countless times with the computer on a daily basis of clicking "Yes," "No," or "Cancel" is the important moment when you are able to prevent the computer from doing harm to you or to itself. Were it to decide to, say, show up on Jeopardy unannounced and without asking, that's a completely different story for Watson 5.0 -- a world where Watson can click its own Yes/No/Cancel buttons.

4/ The computer doesn't care -- at best it can act like it cares. In the movie WALL-E we see a trash collector robot that breaks out of its daily routine and discovers consciousness through love. Given that we humans still don't understand how love works (and doesn't work), it's impossible to imagine that we could ever program a computer to truly love the way that we do -- and yes -- in that special case we can't seem to press our own Yes/No/Cancel buttons.

The taxi driver seemed to nod in disinterest until he asked me, "So, where is this place you're looking for?" I solved the problem by pulling out my iPhone and asking my "other overlord", Google, how to get there. S/he delivered the right answer.

PS I suggested the cab driver visit one of the many sites running the original software systems "Eliza" from the 60s, and to tell Eliza, "The first modern crossword is published and Oreo cookies are introduced." When I tried that just now Eliza simply responded, "I see."

 
The Watson craze last week didn't fully hit me until my cab driver got lost and cheerily exclaimed in thickly-accented English, "Watson! Heeeeelp me!" I find it interesting how the so-called "artifici...
The Watson craze last week didn't fully hit me until my cab driver got lost and cheerily exclaimed in thickly-accented English, "Watson! Heeeeelp me!" I find it interesting how the so-called "artifici...
 
 
  • Comments
  • 30
  • Pending Comments
  • 0
  • View FAQ
Comments are closed for this entry
View All
Recency  | 
Popularity
photo
HUFFPOST SUPER USER
Robert Blackburn
12:09 PM on 02/25/2011
The machine can't feel, therefore the pain avoidance, pleasure seeking, and sex programs found
in the human mind can't be programmed in the computer at this time - or in the foreseeable future. The only program in the human mind that can presently be placed in a computer is the mind's "survival" program, and those in a position to program the human mind deny the existence of such a program. Thus, they can't program it.

For more about the only computer program of the human mind that will work, see: RevolutionOfReason.com and YouTube: RobertLBlackburn.
photo
HUFFPOST SUPER USER
David Campbell
09:23 AM on 02/24/2011
I have no problem with Watson playing a TV game show but I do worry that so many people connect show biz with education. These shows are all about memorized information, mostly trivia, have nothing to do with education, although most of our schools function in the same way, memorized answers on tests of trivia where one can guess the "correct" answer. Watson cannot do real education. Ask it a real question: "Watson, discuss the concept of separation of church and state in our nation, how it came about, why and how it has contributed to our success as a nation and society. What would have happened if we had continued to have an official state religion?"
photo
Dave24
Without God, life is everything.
07:47 AM on 02/24/2011
The fact that Watson was able to understand the nuances of human language is a solid step forward. Growth and complexity is gradual, and eventually it will become exponentially so.
08:59 PM on 02/23/2011
I believe that it is only a matter of time for a computer to reach & surpass human intelligence...

Then, a computer will be our ideal friend...

Always loving us,never misbehaving...

BTW,does anyone know the "market price "Watson might have,if offered to consumers ?
photo
HUFFPOST SUPER USER
RedDogBear
09:38 PM on 02/23/2011
From what I know the next step envisioned by IBM is to make a version of Watson tailored to the healthcare domain.

But its seems to me a sort of super-google would also be a logical next step. Something that instead of giving you links to pages that you have to scour for your answer actually gave you specific answers to questions (as well as the links to tell you where it got the answer and where to look for more info). It could use the Google model of just having ads or a small subscription fee. That's all just speculation though, I have no idea if they are actually planning something like that.
07:28 AM on 02/24/2011
Thanx.

Kurzweil said that AI will reach HI in..2029...

It will definitely have surpassed HI by Aug 4,2097...
09:45 PM on 02/23/2011
August 4th, 1997. Didn't anyone watch Terminator 2?!?!
08:42 PM on 02/23/2011
Yep - The moral of the story being that computers are not "intelligent" and never will be.

They store massive amounts of information and execute software programs (WRITTEN BY HUMAN BEINGS) at extremely fast speeds.

And that is ALL computers are.

My apologies to all the sci-fi fans who know nothing about the actual nature of computer programming and have been hoodwinked by shows like BSG and Caprica, etc. (And note that there is no more rabid sci-fi fan than I am, having read Rocketship Galileo when I was 7 years old, it making me into an instant, lifelong sci-fi nerd.)
07:32 AM on 02/24/2011
I think u r negative...

The computer receives inputs from humans...

The more inputs it receives,the more "human " it becomes...

If humans teach the machine some...100k variations of human behaviour,the computer
will function accordingly...

Getting nervous,being diplomatic,commenting on Huffpost as conserrvative/independent/leftist
& so on...
photo
HUFFPOST SUPER USER
RedDogBear
09:30 AM on 02/24/2011
"moral of the story being that computers are not "intellige­nt" and never will be."

Quotes about how some scientific achievement will never happen have a way of turning out to be wrong.

When I first started programming there were many serious philosophers who claimed that a computer would never be able to play chess. Then they claimed it would never be able to play at the Grand Master level which of course it has.

I'm not sure how you define "intelligence" but it seems to me it should include things like playing jeopardy or chess. Or perhaps flying an airplance. Or making a medical diagnosis. Computers have been doing all those things for years.

I do strongly agree with you that all the talk about Watson as Skynet or HAL are nonsense. While Watson is an amazing AI achievement and IMO clearly "intelligent" its not even close to being conscious or self aware.

However, I don't think any responsible scientist should say we know for a fact that a computer can NEVER be conscious for one simple reason: we don't have an accepted scientific definition for what it means for HUMANS to be conscious yet. Until we do we can hardly say for sure whether computers can or can't be. And that is why I find work like Watson ultimately so fascinating, it gives us insight into what intelligence and thought are that may help us understand how those things work for humans.
photo
HUFFPOST SUPER USER
GregCoyote
07:33 PM on 02/23/2011
Watson is just the evolution of the new search engine technology companies like Google will inest heavily in. Watson has limited abilities in any cognitive area, he is just dorsal good (most of the time) at putting vast amounts of data together. This has obvious ramifications in defense. Good for IBM.
photo
HUFFPOST SUPER USER
RedDogBear
09:35 PM on 02/23/2011
No, that's wrong. For one thing Watson doesn't have access to the Internet. For another try typing a Jeopardy question into Google and then just use the first hundred words or so of the most probable link as an answer. If you get anything that would be considered a correct answer out of the first hundred questions I would be surprised.

What Watson does is similar but much more difficult than Google. It parses natural language. It deals with puns and other word play and colloquialisms. It then searches through its knowledge base (that is somewhat similar to Google) and finds the best answer and again uses natural language processing to present the answer in a form acceptable in Jeopardy.

It has ramifications everywhere but actually I think there are many more promising than defense. As an intellgent assistant to physicians or stock traders for example.
07:09 PM on 02/23/2011
The jepeordy computer bit was kind of entertaining but ultamitely pointless. Retreiving data fast is what computers do. It's no suprise that a computer can analyize a question and retreive the answer quickly, that's not the same as thinking.

Call me when the computer can understand what the host is saying and buzz in for itself.
10:30 PM on 02/23/2011
You've missed the whole point of the exercise. Retrieving answers quickly wasn't the issue, it's understanding the question that's hard. Jeopardy questions rely on complex contexts, wordplay, and ambiguous phrases. Try to answer them just "retrieving data fast" and you're guaranteed to get the WRONG answer fast. Watson is as much beyond a keyword search program such as Google as Hamlet's soliloquy is beyond "Me Tarzan, you Jane".

Watson also does a very good job quantifying how well it knows what it knows and recognizing when it doesn't know anything about a subject -- something most humans have trouble doing well.

BTW: Watson did buzz in for itself. Did you actually watch the show?
photo
HUFFPOST COMMUNITY MODERATOR
LeftLeanWing
RightKickFoot
12:10 AM on 02/24/2011
I think they put a delay in the buzzing mechanism.... such that it gave the humans somewhat of a chance.
photo
HUFFPOST SUPER USER
RedDogBear
10:30 PM on 02/23/2011
No it was not pointless at all. Yes "Retreiving data fast is what computers do" but understanding natural language is something computer usually do very, very badly. That was the significance of Watson, that it could handle all the nuance, word play, every day language in Jeopardy questions, retrieve the correct information and then present it back in a way that would be considered appropriate as a Jeopardy answer. It was actually quite an accomplishment.
photo
HUFFPOST SUPER USER
jabailo
(Participant) Texeme.Construct()
06:13 PM on 02/23/2011
Well, remember, there's a lot of us and one of Watson.

However, Watson gets his buddies to work in a social network (direct cable I imagine) and demands quality education...
HUFFPOST SUPER USER
Visionary Excellence
05:47 PM on 02/23/2011
Eliza is awesome! I cant wait for computers to fluidly understand natural language. Carpel tunnel sux!
photo
HUFFPOST SUPER USER
sporty1
being me
04:29 PM on 02/23/2011
One thing about computers that will never equal that of humans is to understand the concept of infinity. It has been shown by logical and mathematical proof that a computer can never "understand" whether a mathematical operation is finite, whether a series of operations ends, it will just keep calculating and never be able to conclude that an operation is infinite and has no end solution. That is what is so amazing about human learning, knowledge and imagination, we understand the concept of inifinity, that numbers are infinite and go on forever, that space is conceptually infiinitely big or that spaces can be divided into infinitely small pieces, etc etc. In this sense, computers can never "match" or become "like" humans.
photo
HUFFPOST SUPER USER
RedDogBear
09:27 PM on 02/23/2011
"It has been shown by logical and mathematic­al proof that a computer can never "understan­d" whether a mathematic­al operation is finite, "

So what exactly is this proof? I would like to see it. I'm pretty familiar with computer science and logic and I've never heard that.
10:11 PM on 02/23/2011
Haha, yeah it's been awhile since I took that course, I was wondering the same. There was one proof that even a computer with infinite resources and infinite time still couldn't answer in a determinate amount of time if an algorithm would ever complete itself though due to the nature of algorithms themselves, right? Can't quite remember how that one went.
10:49 PM on 02/23/2011
He's probably garbling the halting problem. (http://en.wikipedia.org/wiki/Halting_problem) However, the halting problem isn't about "the concept of infinity". It's about the limits of analytical knowledge.

Now, if sporty1 could demonstrate that humans *can* solve the halting problem for all algorithms but computers can't, that would be pretty impressive. But there's no reason to believe that any human can do it, either.
09:51 PM on 02/23/2011
'a computer can never "understan­d" whether a mathematic­al operation is finite'...a computer can never "understand" anything really, it's all just 1's and 0's without us to interpret the message.

You could very easily teach a computer "practical" infinity that would suit all but 1 in a google of it purposes, though. Dear computer, when you reach a billion to the power of a billion, stop and try something else, that's one line of code.

Also, the author posits an infinite loop as 'to operate in perpetuity without tiring' which also isn't true. The world's most powerful supercomputer consumes the equivalent of 50,000 BLT sandwiches in energy every day.
photo
HUFFPOST SUPER USER
sporty1
being me
09:04 AM on 02/24/2011
Wow that is a lot of lettuce and pork. Yeah, I mean a computer can't gaze into the sky at night and contemplate the infinity of space, or the possibility thereof. And on and on. That is what makes us little humans so special, we can think the sublime, apprehend the unknown even if only haltingly, accept the possibility of or insist on the necessity of the existence of God.