Machines beat us at our own game: What can we do?

Feb 17, 2011 By SETH BORENSTEIN and JORDAN ROBERTSON , Associated Press
In this undated publicity image released by Jeopardy Productions, Inc., host Alex Trebek, left, poses with contestants Ken Jennings, center, and Brad Rutter and a computer named Watson in Yorktown Heights, N.Y. On Monday, Feb. 14, 2011, "Jeopardy!" will begin airing two matches spread over three days between Jennings, Rutter and Watson, who was developed by IBM scientists. (AP Photo/Jeopardy Productions, Inc.) NO SALES

(AP) -- Machines first out-calculated us in simple math. Then they replaced us on the assembly lines, explored places we couldn't get to, even beat our champions at chess. Now a computer called Watson has bested our best at "Jeopardy!"

A gigantic computer created by IBM specifically to excel at answers-and-questions left two champs of the TV game show in its silicon dust after a three-day tournament, a feat that experts call a technological breakthrough.

Watson earned $77,147, versus $24,000 for Ken Jennings and $21,600 for Brad Rutter. Jennings took it in stride writing "I for one welcome our new computer overlords" alongside his correct Final Jeopardy answer.

The next step for the IBM machine and its programmers: taking its mastery of the arcane and applying it to help doctors plow through blizzards of medical information. Watson could also help make Internet searches far more like a conversation than the hit-or-miss things they are now.

Watson's victory leads to the question: What can we measly humans do that amazing machines cannot do or will never do?

The answer, like all of "Jeopardy!," comes in the form of a question: Who - not what - dreamed up Watson? While computers can calculate and construct, they cannot decide to create. So far, only humans can.

"The way to think about this is: Can Watson decide to create Watson?" said Pradeep Khosla, dean of engineering at Carnegie Mellon University in Pittsburgh. "We are far from there. Our ability to create is what allows us to discover and create new knowledge and technology."

Experts in the field say it is more than the spark of creation that separates man from his mechanical spawn. It is the pride creators can take, the empathy we can all have with the winners and losers, and that magical mix of adrenaline, fear and ability that kicks in when our backs are against the wall and we are in survival mode.

What humans have that Watson, IBM's earlier chess champion Deep Blue, and all their electronic predecessors and software successors do not have and will not get is the sort of thing that makes song, romance, smiles, sadness and all that jazz. It's something the experts in computers, robotics and artificial intelligence know very well because they can't figure out how it works in people, much less duplicate it. It's that indescribable essence of humanity.

Nevertheless, Watson, which took 25 IBM scientists four years to create, is more than just a trivia whiz, some experts say.

Richard Doherty, a computer industry expert and research director at the Envisioneering Group in Seaford, N.Y., said he has been studying artificial intelligence for decades. He thinks IBM's advances with Watson are changing the way people think about artificial intelligence and how a computer can be programmed to give conversational answers - not merely lists of sometimes not-germane entries.

"This is the most significant breakthrough of this century," he said. "I know the phones are ringing off the hook with interest in Watson systems. The Internet may trump Watson, but for this century, it's the most significant advance in computing."

And yet Watson's creators say this breakthrough gives them an extra appreciation for the magnificent machines we call people.

"I see human intelligence consuming machine intelligence, not the other way around," David Ferrucci, IBM's lead researcher on Watson, said in an interview Wednesday. "Humans are a different sort of intelligence. Our intelligence is so interconnected. The brain is so incredibly interconnected with itself, so interconnected with all the cells in our body, and has co-evolved with language and society and everything around it."

"Humans are learning machines that live and experience the world and take in an enormous amount of information - what they see, what they taste, what they feel, and they're taking that in from the day they're born until the day they die," he said. "And they're learning from all the input all the time. We've never even created something that attempts to do that."

The ability of a machine to learn is the essence of the field of . And there have been great advances in the field, but nothing near human thinking.

"I've been in this field for 25 years and no matter what advances we make, it's not like we feel we're getting to the finish line," said Carnegie Mellon University professor Eric Nyberg, who has worked on Watson with its IBM creators since 2007. "There's always more you can do to bring computers to human intelligence. I'm not sure we'll ever really get there."

Bart Massey, a professor of computer science at Portland State University, quipped: "If you want to build something that thinks like a human, we have a great way to do that. It only takes like nine months and it's really fun."

Working on computer evolution "really makes you appreciate the fact that humans are such unique things and they think such unique ways," Massey said.

Nyberg said it is silly to think that Watson will lead to an end or a lessening of humanity. "Watson does just one task: answer questions," he said. And it gets things wrong, such as saying grasshoppers eat kosher, which Nyberg said is why humans won't turn over launch codes to it or its computer cousins.

Take Tuesday's Final Jeopardy, which Watson flubbed and its human competitors handled with ease. The category was U.S. cities, and the clue was: "Its largest airport is named for a World War II hero; its second largest, for a World War II battle."

The correct response was Chicago, but Watson weirdly wrote, "What is Toronto?????"

A human would have considered Toronto and discarded it because it is a Canadian city, not a U.S. one, but that's not the type of comparative knowledge Watson has, Nyberg said.

"A human working with Watson can get a better answer," said James Hendler, a professor of computer and cognitive science at Rensselaer Polytechnic Institute. "Using what humans are good at and what Watson is good at, together we can build systems that solve problems that neither of us can solve alone."

That's why Paul Saffo, a longtime Silicon Valley forecaster, and others, see better search engines as the ultimate benefit from the "!"-playing machine.

"We are headed toward a world where you are going to have a conversation with a machine," Saffo said. "Within five to10 years, we'll look back and roll our eyes at the idea that search queries were a string of answers and not conversations."

The beneficiaries, IBM's Ferrucci said, could include technical support centers, hospitals, hedge funds or other businesses that need to make lots of decisions that rely on lots of data.

For example, a medical center might use the software to better diagnose disease. Since a patient's symptoms can generate many possibilities, the advantage of a Watson-type program would be its ability to scan the medical literature faster than a human could and suggest the most likely result. A human, of course, would then have to investigate the computer's finding and make the final diagnosis.

isn't saying how much money it spent building Watson. But Doherty said the company told analysts at a recent meeting that the figure was around $30 million. Doherty believes the number is probably higher, in the "high dozens of millions."

In a few years, Carnegie Mellon University robotic whiz Red Whittaker will be launching a robot to the moon as part of Google challenge. When it lands, the robot will make all sorts of key and crucial real-time decisions - like Neil Armstrong and Buzz Aldrin did 42 years ago - but what humans can do that machines can't will already have been done: Create the whole darn thing.

Explore further: Hendersons introduce hoverboard and a future beyond wheels

More information: IBM's Watson: tinyurl.com/4r8w6gr
Jeopardy: jeopardy.com

3.8 /5 (4 votes)
add to favorites email to friend print save as pdf

Related Stories

'Jeopardy!' to pit humans against IBM machine

Dec 14, 2010

(AP) -- The game show "Jeopardy!" will pit man versus machine this winter in a competition that will show how successful scientists are in creating a computer that can mimic human intelligence.

IBM's 'Watson' to take on Jeopardy! champs

Feb 11, 2011

Nearly 15 years after an IBM machine defeated world chess champion Garry Kasparov, the US computer pioneer is rolling out another device to challenge mankind.

IBM puts supercomputer in 'Jeopardy!'

Feb 08, 2011

"Let’s finish, ‘Chicks Dig Me’," intones the somewhat monotone, but not unpleasant, voice of Watson, IBM’s new supercomputer built to compete on the game show Jeopardy!

Computer could make 2 'Jeopardy!' champs deep blue

Jan 13, 2011

(AP) -- The clue: It's the size of 10 refrigerators, has access to the equivalent of 200 million pages of information and knows how to answer in the form of a question. The correct response: "What is the ...

Recommended for you

Skin icons can tap into promise of smartwatch

16 hours ago

You have heard it before: smartwatches are cool wearables but critics remind us of the fact that their small size makes many actions cumbersome and they question how many people will really have them on their ...

Japan firm showcases Bat-Signal of the future

Oct 20, 2014

A free-floating image created by firing lasers into thin air was unveiled in Japan on Monday, offering the possibility one day of projecting messages into a cloudless sky, as seen in Batman.

Do we want an augmented reality or a transformed reality?

Oct 14, 2014

It seems we are headed towards a world where augmented reality (AR) systems will be as common as smartphones are today – it's already about to revolutionise medicine, entertainment, the lives of disabled peop ...

Can it be real? Augmented reality melds work, play

Oct 14, 2014

(AP)—Mark Skwarek is surrounded by infiltrating militants in New York's Central Park. He shoots one, then hearing a noise from behind, spins to take down another. All of a sudden, everything flashes red. ...

User comments : 28

Adjust slider to filter visible comments by rank

Display comments: newest first

frajo
4.2 / 5 (5) Feb 17, 2011
What can we measly humans do that amazing machines cannot do or will never do?
Translate the Iliad and the Odyssey while preserving the hexameter structure without torturing the target language.
soulman
5 / 5 (2) Feb 17, 2011
Watson's victory leads to the question: What can we measly humans do that amazing machines cannot do or will never do?

Today, pretty much anything. Tomorrow, less and less.
dogbert
5 / 5 (4) Feb 17, 2011
Machines beat us at our own game: What can we do?


Wrong question. Human beings have always created machines to extend human capabilities. No machine has "beat us at our own game". A car can out run a human, but that is hardly the point. The car was specifically created to out run a human and to be able to run for extended periods. But the car has not "beat us". The car extends our reach. A machine system which can scan large databases quickly extends our reach. It does not "beat us".
Moebius
1 / 5 (2) Feb 17, 2011
Let's hand control of the world over to machines, do you think IT would determine we are overpopulated? And when it does, what do you think it would do to solve the problem?
Corban
not rated yet Feb 17, 2011
Since people complain so much about crappy jobs, why not make it unnecessary for people to work those jobs? Then they can do something better!

This is the beginning of a post-scarcity economy. Better hope assets are distributed or communal, or people will be locked out.
soulman
5 / 5 (1) Feb 17, 2011
Let's hand control of the world over to machines, do you think IT would determine we are overpopulated? And when it does, what do you think it would do to solve the problem?

Make more machines?
Since people complain so much about crappy jobs, why not make it unnecessary for people to work those jobs? Then they can do something better!

The reason they find themselves in crappy jobs is because they can't do anything better.
210
1.5 / 5 (2) Feb 17, 2011
"For example, a medical center might use the software to better diagnose disease. Since a patient's symptoms can generate many possibilities, the advantage of a Watson-type program would be its ability to scan the medical literature faster than a human could and suggest the most likely result."
NO!!! Watson showed an inability to wring out semantic and relativistic/contextual elements!!!! 'What was the patient doing' would be the TORONTO moment for Watson and the grounds for an enormous malpractice suit. As I said before, a few days ago, Watson must be socialized and educated in humanity as humans are in order to work in our world with us! THEN it will have a most valuable trait:IMAGINATION.
jselin
5 / 5 (1) Feb 17, 2011
Let's hand control of the world over to machines, do you think IT would determine we are overpopulated? And when it does, what do you think it would do to solve the problem?


I think it would grow frustrated over conflicting goals, develop an apathetic outlook, and start posting snarky comments on Physorg ;) Welcome to the party Watson!
kencreten
not rated yet Feb 17, 2011
All of your comments are interesting. They all touch on various aspects of the issue.

Someday, sooner or later the machines will pass us in every important category. Eventually they will be able to communicate on levels that we cannot understand (John Von Neumann).

It's coming, and I do not think we can avoid it. People posting in this blog might not see it, but it will happen. Human intelligence, imagination, etc, are still mysteries, but they will continue to be investigated, and eventually known to a useful degree. Some of these discoveries will be applied to machine learning.

If you can't beat them, join them. 1... 2... 3...

Transborgrify!
dogbert
not rated yet Feb 17, 2011
kencreten,

It is certainly possible that thinking machines can be constructed, but stating unequivocally that they will be constructed is a bit premature.

Currently there is no self aware machine and no known method to construct such a machine. We do not know enough about awareness/intelligence to begin to design such a machine. We do not even know if we can ever know how awareness/intelligence is generated in ourselves.
kencreten
not rated yet Feb 17, 2011
I'm not saying we can do this now. Obviously we can't. But all of the pieces are being worked on feverishly, and as fast as they can be assembled, they will.
dogbert
not rated yet Feb 17, 2011
But you presume we will discover the nature of self awareness/intelligence.

You are, of course, welcome to any opinion you may have, but I always find it presumptuous to predict with certainty processes we cannot describe and which we have failed to understand.

It is much like the enthusiastic proclamations of certain people that we will soon discover life/intelligences not of this world when we have failed to find life anywhere but here. We certainly may discover life elsewhere, but there is nothing with which we might project a probability since we lack any indication that life exists anywhere else.
soulman
5 / 5 (2) Feb 18, 2011
But you presume we will discover the nature of self awareness/intelligence.

It's a self emergent property borne from highly complex, interconnected networks with an ability to learn and process sensory inputs.

If we could build such a network, approaching human scale complexity in hardware (we can't) or software (still can't, not even Blue Brain), then I think it would just 'pop out'.

Another words, you wouldn't need to code for it specifically or know how it is achieved in humans.
dogbert
not rated yet Feb 18, 2011
soulman,

...then I think it would just 'pop out'.


Me too. My comment was that it is unwise to make predictions of certainty in matters where we have no knowledge.

I like the term "emergent property" too. A very useful way of restating "we don't know how".
soulman
5 / 5 (1) Feb 18, 2011
I like the term "emergent property" too. A very useful way of restating "we don't know how".

No, that's not what is generally meant. For example, cellular automata are very simple units with very simple rules of behavior that can be specified exactly. But these very simple units, interacting in very simple ways, can give rise to extraordinarily complex patterns and structures which could not be predicted a priori. You have to run the process to see the outcome. That's what is generally meant by emergent behavior.
dogbert
not rated yet Feb 18, 2011
Yes, I know what "emergent property" means and how the term is used. It still implies that we do not know how it happens. As you said, it could not be predicted a priori.

As I said, I like the term. It has a valid and otherwise difficult to express meaning.
frajo
5 / 5 (1) Feb 18, 2011
There is an old adage for the phenomenon of emergence:
"It's more than the sum of its parts".
Quantum_Conundrum
not rated yet Feb 20, 2011
A multi-threaded algorithm which is sufficiently complex and run on multi-core processors may be indistinguishable from an truly "emergent" machine.

After all, recursion, iteration, and container classes organizing other objects in software can be viewed as a form of "emergence".

A person can build "virtual machines" even if they know nothing about electronics, and they can even "cheat" and make virtual machines which are not directly bound to physicality through the use of "global" variables and other tactics. Now in the real machine there is an underlying physicality, but the virtual machine does not necessarily need to "understand" the physical machine to benefit from it.

Once you have virtualized A.I. you can then use the A.I. to work on the deal of making a physical machine optimized to the algorithm, or which may have little or no software at all, and represent true emergent intelligence at the hardware level, as our own brains do.
TheGhostofOtto1923
1 / 5 (3) Feb 20, 2011
We do not even know if we can ever know how awareness/intelligence is generated in ourselves.
Human awareness and intelligence are inextricably bound to our desires to protect ourselves and to procreate. An inordinate amount of time and effort goes into creating personas which might ensure these things. They determine our self-perception and our interpretation of the world.

Human thought is also tainted by the flaws we are born with and accumulate as we age. Endemic flaws become irrational memes which damage and disrupt the social fabric in significant ways.

Machines will have no desires to survive or replicate unless we program them to. And their thinking will be (is) something unrecognizable to us. They can be hobbled by making them simulate our behaviors and reasoning, to where we could not tell the difference. But why bother, except to make us a little more comfortable when interfacing with them? Watson got a hand to press the buzzer to make US feel better.
Quantum_Conundrum
not rated yet Feb 20, 2011
Consider this:

The starcraft 2 Terran a.i. is not very good script, however, it is able to beat MOST human players based on brute force of the computer's ability to multi-task 4200 to 4800 game actions per minute, vs 120 for what is considered the "average" among VETERAN Starcraft players. True beginners are around 30 actions per minute or even less.

I am atuallly "below average" for a veteran player in terms of actions per minute, but I can beat 2 insane computers allied against me, becuse I'm just that smart to make up for my low APM.

But the point is, MOST humans cannot defeat the highest NON-cheating A.I. in 1v1 due to the brute force. The computer's script is pretty bad, but the fact that it executs up to 4800 actions per minute means it can overwhelm a "smarter" player through better micromanagement and better multi-tasking.

Some fans have written better A.I. scripts which are almost impossible to defeat, even extremely hard for pro gamers to defeat.
TheGhostofOtto1923
1 / 5 (3) Feb 20, 2011
Once you have virtualized A.I. you can then use the A.I. to work on the deal of making a physical machine optimized to the algorithm, or which may have little or no software at all, and represent true emergent intelligence at the hardware level, as our own brains do.
QC wishes he had a holodeck.
Quantum_Conundrum
not rated yet Feb 20, 2011
So the improved "fan written" Starcraft 2 A.I., which does not cheat, would be an example of an "expert machine".

It has a very solid build order and scripts, and because it is doing 4800 actions per minute, vs around 400 for the best korean pro gamer, it is virtually impossible for a human to beat it.

By comparison, I can defeat TWO "insane" A.I. which CHEAT and get 25% free money, even when they are allied against me, on any map, and I can defeat FOUR of such A.I. allied against me on an island map.

But a "slight" improvement by fans to fix a few bugs in the scripts, and to give it a better build order causes the non-cheating A.I. to be better than the cheating A.I.

Modern "Expert machine" chess A.I. are almost unbeatable. In a few more years it will be possible to simply have the computer calculate all possible board positions at any stage of a game, and all possible game outcomes, and choose the best move that way.
Quantum_Conundrum
not rated yet Feb 20, 2011
Once you have virtualized A.I. you can then use the A.I. to work on the deal of making a physical machine optimized to the algorithm, or which may have little or no software at all, and represent true emergent intelligence at the hardware level, as our own brains do.
QC wishes he had a holodeck.


I'm not talking about a holodeck or "virtual reality".

I'm talking about a "Virtual Machine," just like an the virtua machine that runs server side scripting and video game A.I., etc. Examples are the Zend Engine for PHP, and the Starcraft 2 Galaxy editor, which can damn near do anything you imagine.
TheGhostofOtto1923
1 / 5 (3) Feb 20, 2011
which can damn near do anything you imagine.
Again with the superlatives. Can it create for me a palpable version of Seven of Nine? Deanna Troy perhaps? I didn't think so. The holodeck could create a version of the Enterprise so realistic that it could fool Professor James Moriarty. Who was himself a holodeck entity (originally). AND it can recreate a copy of the universe for Moriarty and his girlfriend to explore for a virtual lifetime, without them knowing the difference. Can your little widget do that?

"Picard mentions the possibility that the crew's reality may actually be a fabrication generated by "a little device sitting on someone's table." This unnerves Barclay enough for him to test the nature of his own reality one more time, by giving an audible command to 'end program' to any suspected computer. This has no effect."
nuge
not rated yet Feb 20, 2011
OF COURSE computers will eventually be able to do everything that humans can do, even creativity, empathy and "all that jazz". We are, after all, only matter. Whatever it is that happens in a human brain could in theory be perfectly mimicked by a synthetic structure. It will take time and effort, but there is absolutely no reason why machines will never have "that special something", whatever it is.

Whether they should or not is a separate matter.
Quantum_Conundrum
not rated yet Feb 20, 2011
Otto:

Why are you obsessed with holodecks?

I am not talking about "virtual reality" or holodecks. I'm talking about making the leap from "Expert Machines" to General purpose A.I.

Holodecks and "virtual reality" are irrelevant.

You clearly have never heard the term "virtual machine".

A "virtual machine" is a simulated machine in software which does some useful function.

PHP runs on a "virtual machine".
TheGhostofOtto1923
1 / 5 (3) Feb 20, 2011
A "virtual machine" is a simulated machine in software which does some useful function.
The holodeck could create virtual machines as when it recreated the enterprise and a shuttlecraft. They did a useful function in convincing the virtual villain Moriarty that he and his squeeze could transport off the holodeck, which they could not.
Quantum_Conundrum
not rated yet Feb 21, 2011
The holodeck could create virtual machines as when it recreated the enterprise and a shuttlecraft. They did a useful function in convincing the virtual villain Moriarty that he and his squeeze could transport off the holodeck, which they could not.


I'm pretty sure most have seen that episode.

What's your point?