The Election Autopsy

[userque]

I would love to see the current Watson, given infinite amount of training, to articulate how many things it sees in this picture. And this is an easy one.

Regarding the captcha, I still maintain that ML, given enough training data, can solve 2 second problems.

Regarding the old/young lady problem...it would all depend upon whether the ML algo was trained with similar images of the young lady, and separately, the old lady.

But we are far off topic.

http://scholarworks.sjsu.edu/cgi/viewcontent.cgi?article=1425&context=etd_projects

 

[nitro]

Regarding the captcha, I still maintain that ML, given enough training data, can solve 2 second problems.

Regarding the old/young lady problem...it would all depend upon whether the ML algo was trained with similar images of the young lady, and separately, the old lady.

But we are far off topic.

http://scholarworks.sjsu.edu/cgi/viewcontent.cgi?article=1425&context=etd_projects

I am talking about __any__ image that a human being could easily solve in seconds. Train Watson on any number of finite images you can think of, and I can create one that a human being would be able to solve that Watson couldn't given its past training. I am not restricted to the Chinese Room Experiment.

This is an inductive argument. If you now trained Watson on my new image, I would come up with another one it couldn't solve that a human being would have no problems with. No matter what you do, I could always tell it was a computer, unless it has achieved "consciousness" [whatever that means].

Q.E.D.

 

[userque]

I am talking about __any__ image that a human being could easily solve in seconds. Train Watson on any number of finite images you can think of, and I can create one that a human being would be able to solve that Watson couldn't given its past training. I am not restricted to the Chinese Room Experiment.

This is an inductive argument. If you now trained Watson on my new image, I would come up with another one it couldn't solve that a human being would have no problems with.

Q.E.D.

How can you be sure Watson hadn't already been trained on the image you pick?

 

[nitro]

How can you be sure Watson hadn't already been trained on the image you pick?

Because it would be a creative act done by a [human] artist trained in optical illusions. It wouldn't exist until I created it. Since there are an infinite number of possible such illusions, Q.E.D.

For example, I would ask a modern Escher to do an image for me.

 

[userque]

Because it would be a creative act done by a [human] artist trained in optical illusions. It wouldn't exist until I created it. Since there are an infinite number of possible such illusions, Q.E.D.

For example, I would ask a modern Escher to do an image for me.

Got any examples of such 2 second images with the question?

 

[nitro]

Escher.Ask is this possible.

 

[userque]

Escher.Ask is this possible.

Certainly, a human would need more than two seconds to know if it's possible. These images could also be given to the ML algo for training. Here is the "Impossible" part of the gallery:

http://www.mcescher.com/gallery/impossible-constructions/

 

[nitro]

Certainly, a human would need more than two seconds to know if it's possible. These images could also be given to the ML algo for training. Here is the "Impossible" part of the gallery:

http://www.mcescher.com/gallery/impossible-constructions/

It wouldn't take me more than two seconds. But fine, "two seconds". "New York Minute".

 

[AAAintheBeltway]

I am talking about __any__ image that a human being could easily solve in seconds. Train Watson on any number of finite images you can think of, and I can create one that a human being would be able to solve that Watson couldn't given its past training. I am not restricted to the Chinese Room Experiment.

This is an inductive argument. If you now trained Watson on my new image, I would come up with another one it couldn't solve that a human being would have no problems with. No matter what you do, I could always tell it was a computer, unless it has achieved "consciousness" [whatever that means].

Q.E.D.

Isn't this the obvious Achilles heel of autonomous cars? They cannot react predictably to unanticipated scenarios.

we will need a gargantuan investment in road-based sensors etc to make them remotely feasible. Until then, each one is a loaded gun primed to go off at any moment. The freakin' Tesla in flordia could not discern a semi trailer that turned in front of it. What if it had been a child? What happens if ice and snow block the camera?

 

[Scataphagos]

Isn't this the obvious Achilles heel of autonomous cars? They cannot react predictably to unanticipated scenarios.

we will need a gargantuan investment in road-based sensors etc to make them remotely feasible. Until then, each one is a loaded gun primed to go off at any moment. The freakin' Tesla in flordia could not discern a semi trailer that turned in front of it. What if it had been a child? What happens if ice and snow block the camera?

What's the real appeal of autonomous cars anyway?

Do the young and stupid actually resent the attention that driving a car takes away from time they could spend with their beak buried in their iPhone?

Good grief!