Jeopardy, Watson, and Why Artificial Intelligence is Still Pointless

I'm generally considered a "cynic" when it comes to the value of Artificial Intelligence in general (and the Semantic Web nonsense in particular). This tends to get me into heated disagreements with people who have careers in the field... or those who cling to Jetsons-type fantasies of having Rosie the Robot doing their dishes.

I frequently respond, "you already have a robot that washes dishes. It's named Maytag."

The most recent breakthrough in the field is the highly impressive computer named "Watson," which after several years, and millions of dollars of R&D, was able to beat Ken Jennings at Jeopardy... or to be more accurate, the computer was faster than a human at clicking the buzzer when it came to answering questions that were from previous episodes of Jeopardy.

Interesting? Of course! Practical in any measurable way? No...

Let me be clear: the human language "fact mining" technology in Watson is impressive. And many have discussed the possibility that it could be used by doctors or lawyers or hedge-fund managers who need to react quickly to find the "right" answer. Well, I got a couple of monkey wrenches for ya:

  1. There is rarely "one right answer" in medicine for why you have a tummy-ache.
  2. Other than patient history, there is no study that says greater access to information would yield to better patient health... and a great number that have found the exact opposite. Like the explosion of unnecessary surgery since the invention of MRIs.
  3. If a lawyer or hedge fund manager ever used Watson to uncover "the truth," it wouldn't take long for somebody else to make software to game the system, making it practically worthless without massive human intervention (not unlike Google's losing battle against SEO Spammers)

I could go on and on... or I could just point out that after IBM's Deep Blue 2 defeated Kasparov at chess, IBM shut down the program, denying Kasparov the rematch he wanted. Why? Well, Deep Blue 2 was incredibly expensive to build and maintain... and other than good PR for IBM's hardware, the machine had practically no value.

Why does it have no value? Because with a laptop and $50 worth of software, just about any experienced chess player could beat Deep Blue, or it's rival Hydra. These machines just did brute-force analysis, and didn't have a "hint" about what strategy would be optimal. Somebody who knew what he was doing -- aided with a very small shell script -- could beat any supercomputer on the planet. Which is kind of the point I've been trying to make all along:

Artificial Intelligence isn't impossible, its just impractical. Deeply, deeply, deeply impractical...

The point is not that human minds are better than robots. It's just that until my Roomba has some petty cash to spend -- and a desire for something other than electricity -- the economy will be driven by human needs. Whereas it would be cool to have a general-purpose supercomputer that looks like a human and does your dishes, or beats the pants off of you at Jeopardy, it's always going to be more cost-effective to create single-purpose machines that increase human ability to meet human needs...

So... what kinds of tools would these be? Let's talk medicine: what kinds of software would actually improve human health? It seems pretty obvious: keep patient history in a secure, portable repository... and have your doctors read it!! Or perhaps a big database of drugs, and how they interact with other drugs, to ensure no side-effects? Or, perhaps a database of standard medical processes and procedures that have scientifically proven their efficacy? Or maybe slap an RFID chip in the leg to be amputated to avoid hospital error? Any of those would cost a heck of a lot less money than the Watson project and yield immediate benefits...

Watson is -- quite frankly -- a solution in search of a problem...

Good observation

You are definitely correct from a practical point of view, and Deep Blue is a total marketing trick. Although from a theoretical perspective we need to keep exploring that area. Or do we? I myself have never fully wrapped my mind around a possibility of a program that would mimic/surpass human intelligence and i certainly would not want to be the guy debugging it...Here's a concern i have though: currently rulers of the world (governments, corporations, etc.) all depend on humans to do the major part of the everyday work - this in turn gives us the power to protest and strike when our rights are taken. What happens when robots replace majority of the activities needed to run this world? Anywho...back to WebCenter! Yay!

"I've seen things you people wouldn't believe. Attack ships on fire off the shoulder of Orion. I watched C-beams glitter in the dark near the Tannhauser gate. All those moments will be lost in time... like tears in rain... Time to die." - Blade Runner

cyborgs beat AI every day of the week...

The point is to remember the socioeconomic drivers that make innovation possible in the first place... and then look to the future.

All people want to be better / smarter / stronger / faster... so we build machines to help them. As Scott Adams says, we're already cyborgs: we just have easy-to-remove modules. Eyeglasses, cell phones, automobiles, fork lifts, these all make us meta-cyborgs already.

The ultimate next-step is not AI: it's the neural interface. Once we can connect safely to Google with our minds, there's no longer any purpose to AI. Even if we could create AI, there's pretty much no way it would ever be of any use. A general purpose guessing engine? I already got one: my brain. What I could really use is a faster memory bank...

game perspective

Being an avid gamer this issue is usually the make or break of a good game versus a bad one.

The problem is not developing AI for easy to learn games such as chess, its for more complex ones.
For example, anyone can make a Texas Hold Em Poker AI. The rules of poker are actually quite simple when broken down into raw mathematics. The problem is, simulating human emotional responses. A strict mathematical poker simulator might be effective for a Casino for calculating slot machines but it would be incredibly boring to play against. Without the human element the computer would become incredibly easy to predict, much like a robot.

This issue is compounded with more complex games like Starcraft or other strategy games. The human player wants to be challenged but not overwhelmed. You can't make the game too easy or it gets boring very quickly. Likewise you can't make the game too hard or it becomes too frustrating.

A good game must feel like you could lose at any moment but allow you the opportunity to win.
There are drastic differences when playing against a human opponent versus the computer AI. The real genius is the one that creates the game where you can't tell if you are playing a human or the computer.

You need to add emotional elements.
"I'm tired so I'm going to take this big risk and do something stupid hoping it might pay off"
"I'm bored so I'm only going to make all BLUE troops this time"
"I'm going to just sit in this corner build up a massive defense and do nothing while I surf youtube videos"

The point being, there are Thousands of really bad games out there due to the bad AI and only a handful of good ones. The good ones become "Classics" ie Civilization

very true...

The best games make it almost FUN to fail... or, at the very least, not feel so challenged that you become helpless.

Hey, I have an issue

hey there, I'm having problems viewing your site in the Opera browser (the font is extremely tiny). I've tried increasing the font size from the settings option but that doesn't seem to work. Do you have any tips on what I should do? (By the way, I'm using Windows 7) - strep throat symptoms

Is your Opera updated?

Another user experienced issues with Opera, which went away after updating the browser:

Artificial intelligence

is no match for the real thing - but is often better than real stupidity.
As Douglas Adams remarked:
"I am rarely happier than when spending an entire day programming my computer to perform automatically a task that it would otherwise take me a good ten seconds to do by hand."


My general opinion is that a machine can never replace somebody who:

  1. knows what they're doing, and
  2. gives a damn!

For somebody like that, machines simply make them more efficient at their job... of course, that's no guarantee that you wont be fired by some short-sighted manager... just that you aren't so easily replaced.

Recent comments