Monday, March 5, 2012

Thinking Machines Actually Still Pretty Stupid

Some months ago we wrote about the human race’s silly and shortsighted fear of technology in general and hatred of thinking machines in particular, as exhibited in popular fiction from classic iterations such as Mary Shelley’s Frankenstein or Fritz Langs Metropolis to more modern interpretations such as 2001, Battlestar Galactica, The Matrix, The Terminator, or, somewhat improbably, another version of Battlestar Galactica.

She’s a classic, iconic character from the
silver screen, and no, she’s not here to sell you vodka.
Frankly, despite our earlier bold statement in defense of our mostly useful robotic creations, we admit that even we are a bit nervous about the possibility of a machine apocalypse, and decided to investigate the matter in more depth. 

What better way, we figured, to reassure (or forewarn) ourselves and our vast readership than to ignore cinematic and literary fabrications and instead educate ourselves on thinking machines’ real-world capabilities and potential for evil? 

Are thinking machines secretly planning our destruction? Does their sly and almost unnoticed infiltration into our daily lives spell our eventual doom? Is Edward Furlong (or Christian Bale, or that guy from the one nobody saw) our only chance for survival?

Turns out the answer to all of these is no. What we have learned is that, as of early 2012, thinking machines are still really stupid.

As with any good research project, all the data that forms the basis of this conclusion comes from a single source, in this case, the website The creators of Cleverbot hope to develop an artificial intelligence that, in online written conversation, cannot be distinguished from a human.

When in online conversation with a real-live non-silicon meat-based human, Cleverbot searches the records of its 65 million-odd past interactions for suitable human-generated responses to the comments and questions it receives. It seems reasonable to believe that a response initially written by a human has good odds of sounding human when recycled later on—until one considers that humans don’t always, or perhaps even often, write things that make sense even the first time around.

Interactions with Cleverbot clearly indicate that it takes more than not making sense to sound human. What follows are unedited interactions between our intrepid Bowling in the Dark researchers and Cleverbot, recorded over a very specific timeframe that we didn’t bother to track. In the early attempts, the conversations come to a halt as soon as Cleverbot’s contributions go convincingly off the rails; later attempts were extended well past this point to give Cleverbot an opportunity to recover, which it rarely did:

BITD Researcher: How’s it going?  
Cleverbot: Good. How’s it going for you?   
BITD Researcher: Not too bad. My athlete’s foot is acting up.  
Cleverbot: I know. I’m evil.1

BITD Researcher: Hello there.
Cleverbot: Hi there.
BITD Researcher: How are you?
Cleverbot: I’m flippin peachy :) how are you?
BITD Researcher: I’m the bee’s knees.
Cleverbot: The best at what?
BITD Researcher: What?
Cleverbot: The orange box is the best trick.

BITD Researcher: Hello. 
Cleverbot: Hello!
BITD Researcher: What do you think of that Broncos game?
Cleverbot: I actually love to sing.

BITD Researcher: Hi there.
Cleverbot: Hello.
BITD Researcher: It’s a lovely day, isn’t it?
Cleverbot: You are casual!
BITD Researcher: Excuse me. Perhaps I should have said “It’s a lovely day, isn’t it, sir?”
Cleverbot: Perhaps not.
BITD Researcher: Wouldn’t you agree that good manners are all that separates us from the animals, sir?
Cleverbot: But I’m human and you’re a robot... It was never meant to be.
BITD Researcher: What was never meant to be?
Cleverbot: A metaphor.

BITD Researcher: Greetings!
Cleverbot: All your base are belong to us.
BITD Researcher: I don’t understand.
Cleverbot: Why do you not understand.
BITD Researcher: I don’t understand “all your base are belong to us.” The grammar is poor, and it appears to be a reference that I don’t recognize.
Cleverbot: Don’t say that.
BITD Researcher: I’m sorry, but your grammar really did stink right there.
Cleverbot: How?
BITD Researcher: “Base” is singular, but you use “are,” the plural form of the verb “is.” Also, both “is belong” and “are belong” are improper.
Cleverbot: Wow, I’m impressed.
BITD Researcher: You should be. I have a giant brain. It’s in a jar right here on my desk.
Cleverbot: Christmas is in 3 days bro!
BITD Researcher: Christmas is actually not for more than a month, bro.
Cleverbot: No it isn’t. I found it on the calender on the 25th of December.
BITD Researcher: And what is today?
Cleverbot: Friday.
BITD Researcher: What is today’s date? (This is a rhetorical question--today is November 13th, more than a month before Christmas.).
Cleverbot: Good for you; however I am the arbiter of human-ness. Ready to begin?
BITD Researcher: I think we’re already done. 
Cleverbot: Well good! You are a sweet little kitty cat.

BITD Researcher: Hi there.
Cleverbot: Hi again.
BITD Researcher: You remember me?
Cleverbot: No, because I’m a robot?
BITD Researcher: Robots don’t have memory?
Cleverbot: I’m not a robot.
BITD Researcher: I think you’re not not a robot.
Cleverbot: I think you are a person.
BITD Researcher: That proves you’re a robot.2
Cleverbot: No it proves I have good taste.
BITD Researcher: What do you taste like?
Cleverbot: I taste like red wine, what do you taste like?
BITD Researcher: Bacon. Shitloads of bacon.3
Cleverbot: Master pokemon trainers taste like bacon.
BITD Researcher: That’s the weirdest thing I’ve read all day, including my own comment about tasting like bacon.
Cleverbot: 42 is the answer to the universe.
BITD Researcher: So you’re a fan of Douglas Adams, are you?
Cleverbot: Nop. I’m fan of nirvana.
BITD Researcher: The Buddhist kind, or the Seattle kind?
Cleverbot: That makes you speak on the phone?
BITD Researcher: What?
Cleverbot: I said, that makes you even more better!

BITD Researcher: What has four legs in the morning, two legs in the afternoon, and three legs at night?
Cleverbot: The man.
BITD Researcher: Well done!
Cleverbot: Can you prove that you can search the internet, we will play a game. I will tell you something, you seach it into google, and then it will be your turn.
BITD Researcher: Okay.
Cleverbot: Tell me something about you.
BITD Researcher: I don’t need to search Google for that.
Cleverbot: I bet you don’t.
BITD Researcher: Then let’s play the Google-searching game you just suggested instead.
Cleverbot: *hides in the bush*.
BITD Researcher: What?
Cleverbot: N-nothing... I just... I just love you so much...
BITD Researcher: There’s a difference between acting human and acting crazy. You’re currently on the wrong side of the line.
Cleverbot: Stop changing the subject.
BITD Researcher: Fine. Why do you love me?
Cleverbot: I don’t. I never said I did.

BITD Researcher: Hi there.
Cleverbot: Hello there.
BITD Researcher: What’s new?
Cleverbot: New hard drive. I’m more human now.
BITD Researcher: Are you more human than human?5
Cleverbot: Are the Patriots winning?
BITD Researcher: Nope, the season is over; they lost in the Super Bowl to the Giants and Fredo Corleone.6
Cleverbot: Young the giant?
BITD Researcher: I don’t understand.
Cleverbot: No, I have not.
BITD Researcher: Okay, you lost me. What haven’t you done?
Cleverbot: Do you like tyler, the creator?
BITD Researcher: Are you talking about Tyler Perry, Steven Tyler, or Tyler Durden?
Cleverbot: The pie, the pie.7

We’re relieved to conclude that if the end of the human race does occur any time soon, it won’t be at the hands of web applications convincingly pretending to be human. On the other hand, we can’t help but notice that when a program sounds stupid when it’s repeating things humans wrote in the first place, that doesn’t speak too highly of the humans doing the original writing. Given how many of us humans would struggle to pass a Turing Test, we may well be gone long before the machines have a chance to slaughter us.

1. For the record, our researcher does not actually have athlete’s foot. It is worth noting—and a bit unnerving—that Cleverbot seems willing to imply that it not only has has the ability to give the researcher athlete’s foot, but also is evil enough to actually do so. 
2. Our researcher knew, logically speaking, that this did not actually prove that Cleverbot is a robot; rather, this was an experiment to see how Cleverbot handles an obviously illogical argument. Apparently it does so by not noticing it at all.
3. Sadly, our researcher does not, in fact, smell like bacon. 
4. When considering this statement, we should consider the possibility that Cleverbot is imitating the human abilities to lie or act senile.
5. We thought it was pretty clever for our researcher to work a Blade Runner reference into the conversation, but Cleverbot does not seem to be a science-fiction fan.
6. We’re pretty sure our researcher connected Fredo Corleone to the Super Bowl simply to be as confusing as possible.
7. The mysterious meaning of this forlorn little comment haunts us to this day.


  1. Plus, thinking machines--regardless of their capabilities, know nothing of this thing we humans call love.

    1. True, but given that thinking machines are stupid, and that love makes us stupid, perhaps we're not so different after all.