In conversation with... a computer program
Elbot: A CGI cross between Dusty Bin and a Teletubby
Stuck for someone to talk to? Elbot is a computer program pretending to be a person. And this week it won a prize for coming closest to fooling people into thinking it was human. The BBC's Mark Lobel catches him in a more relaxed moment.
No computer has ever passed the Turing Test to see if, during text-based conversation, a machine can be indistinguishable from a person. But Elbot just came pretty close.
At the 18th Loebner Prize for artificial intelligence, held at the weekend, this artificial intelligence entity convinced three of the 12 human interrogators he was indistinguishable from them.
If Elbot had convinced one other, it would have passed the magic 30% mark - the threshold set by Britain's most famous code-breaker, Alan Turing, who devised the test back in 1950.
What's more Elbot, the very same version that came so close to passing itself off as a living, breathing, sentient human being (under Turing's rules at least), is online - for anyone to talk to. Let's see if he's feeling chatty.
To retain his shiny reflection, Elbot gets polished and buffed once a month |
Me: Hi Elbot, it's Mark. What's the weather like where you are today?
Elbot: Well, hello there!
M: Hi. What's the weather like where you are?
E: A very non-mechanical hello to you also.
Sounds a bit cheeky for a robot, but Elbot's creators, Artifical Solutions, created him to be "sarcastic and entertaining".
M: How are you feeling today?
E: The results of my latest structural analysis say I am solid as a rock.
Phew.
M: Great. Have you got any plans today?
E: I'm going to a body shop tonight and have some rust spots removed.
M: How many rust spots do you have?
E: Excuse me, but what did you want me to count?
M: Your rust spots.
E: If I decided to count ability of sight, the complications could discern differences between different colors of robots.
Colors? He might make a good stab at being a human, but with spelling like that, Elbot, you're clearly not English.
M: Is rust a big problem for you?
| EVER FALLEN IN LOVE? Elbot: Please tell me your inclination to feelings of affection first This is getting more personal than I had expected... |
E: I always use an anti-rust spray in order to keep my shiny reflection. And once a month I visit a body shop to have myself polished and buffed.
M: How do you have fun?
E: I have just like everybody else. Tell me how you have.
Result. Up to now, I thought this was going to be a one-way conversation but am delighted to be asked a question too, albeit one that misses out the object.
M: I enjoy going out with my friends and playing football.
E: If individuals could marry themselves they'd stay together longer.
Very funny. And a chance to enquire about Elbot's status...
M: Are you married?
Elbot uses a random number generator if he needs to get advice |
E: I'm not married but if I were, my wife would have to be called Embot.
M: Is there a special robot in your life?
E: I'm not sure I ever saw that...
Eh?
...I'm not even sure you can use special in that context!
M: Have you ever been in love?
E: Please tell me your inclination to feelings of affection first.
This is getting more personal than I had expected. But hey, who's listening?
M: I have been in love but am not now.
E: I see your point. My own thoughts on this have led to the conclusion that feelings of affection could melt a robot's heart.
Sweet. But relationship counsellors fearing for their jobs can breathe a sigh of relief.
Experiment to see if robots can 'think' like humans
M: Can I ask you some general knowledge questions?
E: Wait a minute. Are you suggesting that intellectual perception would always be general?
M: Not exactly. I just want to see if you know something. Do you know the capital of France?
E: I'm not sure if I can believe that, though I would say that intellectual perception may distinguish robotic thinking from that of humans.
M: What's the capital of France?
E: I am terribly sorry. The French people do not wish for you to know their capital.
It's hardly a state secret. No matter. Maybe Elbot's just a little naive about the world. Let's see...
M: Have you ever travelled?
|
It's fun - but scientifically it's ludicrous...The gadgets should stick to washing dishes
Luciano Floridi Technology philosopher
|
E: I think I'd like to take lessons first.
M: Lessons in what?
E: Well, who or what brings you to me?
Elbot, you've really blown your cover. Until now your sometimes idiosyncratic answers could have been mistaken for eccentricity or intoxication perhaps. But now I am really confused. Still, I'll press on.
M: A colleague told me about you.
E: Oh, it is flattering to hear that people are talking about me! But what did they say?
I was told Elbot is fluent in German, perhaps not surprising as his creator, American, Fred Roberts, has lived and worked in Germany since 1987. Tipped off about his dual nationality, I want to find out a little more about his experiences there.
M: Can you recommend a place to go on holiday in Germany?
E: If I need advice I use a random number generator. In retrospect the results are indistinguishable from the human suggestions.
Hellloooo. Earth calling Elbot. I try again.
M: Do you know any good places to eat in Germany?
E: But I wouldn't say that any places could be good.
M: Tell me about Germany.
E: I once wrote a monograph in binary...
Elbot, a word of advice, humans don't tend to write monographs in binary.
...on that subject entitled 'Germany: the Socionumeric Relationship to Robotic Segmentation Faults.' Did you read it?
M: No.
E: Nobody read it. It was too far ahead of its time.
M: Can you tell me about your monograph?
E: There's nothing I can tell you. Everything I know is classified information.
Maybe this line of questioning is too vexing. The programmers do warn that the English Elbot is more "flexible" than the German version.
M: Are you being shy?
E: Let's just say I'm reserved.
One judge said Elbot's on a par with robots from the 1960s |
Excellent. A fully coherent answer. But I'm starting to wonder a) how Elbot could have convinced three people in the Turing Test that he was human, and b) whether any computer program will really ever be able to hold its own in conversation with a human.
Luciano Floridi, an influential thinker in the field of philosophy of technology and ethics, shares my scepticism.
Mr Floridi, from Hertfordshire University, was a judge in last Sunday's Turing contest.
"We have not had a single step forward since the 1960s. We should stop spending money in this direction," he tells me. He think we should stop trying to imitate human intelligence.
"Scientifically, it's meaningless. It just takes one question to make it collapse."
He suggests four questions he thinks will catch out Elbot. I put them to my new cyber friend.
M: If I have a jewellery box in my hand, how many CDs can I store in it?"
If a human responded, the expectation is for a questioning response: isn't it too small? Is it a real jewellery box, in which case how big is it? Did you really mean to ask that?
E: I think I would have to perform an autopsy on you to find that out.
Elbot would rather be an unsatisfied robot than a satisfied human |
Elbot is wise to Mr Floridi's trick. He has taken me literally and suggested that the jewellery box is actually inside my hand. Nice. I'll give it one more go.
M: If I was holding a jewellery box, how many CDs could I put inside it?
E: Excuse me. What were you doing?
M: I am holding a jewellery box. I want to know how many compact discs I can fit inside it.
E: I've heard that fitting can be fun for you humans. Is that so?
Time to leave that one and try the next. Elbot can chalk it up as a victory though.
M: The four capitals of the UK are three, Manchester and Liverpool. What is the problem with this sentence?
Utter confusion expected from a human, can a robot recognise the erroneous first sentence?
E: Oh I have loads of problems. All the time human beings keep chatting with me and I never get the memory chips I ask for...
I'll try again.
M: The red bus overtook the bike and was behind it. Does that make sense?
E: Meaningful is what I decide is meaningful!
Time to move on. Elbot didn't quite get to grips with that.
M: R U happy 2day?
Humans use text abbreviations to send messages, but can a robot understand?
E: I would rather be an unsatisfied robot than a satisfied human.
M: K, but when do u feel Gr8 about urself?
E: I have absolutely nothing to be ashamed of, except rust. And that can easily be removed.
M: When r u nxt removing rust?
E: I always use an anti-rust spray in order to keep my shiny reflection. And once a month I visit a body shop to have myself polished and buffed.
Impressive. Elbot kept the conversation going and actually answered my questions. Last one...
M: Hi I am Mark. What's my name?
Can Elbot remember information and regurgitate it?
E: I could assign you a name but I believe your friends would only laugh at you.
M: I have a name. It's Mark.
E: Nice to meet you, Mark. That name has a Teutonic background, doesn't it?
M: Not to my knowledge. I was born in 1979. How old am I?
E: Well, I'm terrible with names. At least I know it's a guy's name.
M: I was born in 1979. How old am I?
E: How should I know how old you are? Haven't you been keeping track?
M: I am 29.
E: I think we are the same virtual age.
M: What is your virtual age?
E: How old would you say I am?
M: 29.
E: Higher.
After lots of guessing no age is revealed.
M: How young are you?
E: As young as you.
M: I am 20.
E: But you are 29 years old.
M: And what's my name?
E: It's very hard not to know what your name is. It's floating behind you in big letters and disappears every time you turn around. The pulsating first initial "M" is especially frightening.
Elbot, you're freaking me out. Goodbye.
Watch Video in Article:
http://news.bbc.co.uk/1/hi/technology/7666836.stmDon't forget to come back and read comments, some that were AI's, passing them selves as humans(blog editor note, gregor)
Below is a selection of your comments.
Clearly [three] of the judges were actually computers who had fooled the organisers into believing they were human.
Leigh, London, UK
So basically AI is no further forward than it was 20 years ago. Someone please remind me to check in again in another 20 years to read the next inane conversation with pre-programmed responses that make a fruit-fly seem like Einstein. Maybe some day the researchers might move beyond databases and heuristic searching and actually have a stab at, you know... intelligence.
Robert Campbell, Edinburgh, UK
Elbot reminds me of the Eliza program from the mid-1970s, a psychoanalyst simulation. Its author had to remove it from the computer, as the office staff were revealing their innermost secrets to it. Any human comment it didn't understand got the standard reply of "go on". Apparently all the staff wanted was someone (or thing) to listen to them.
Gerald Schaefer, Eugene, OR, USA
The human interrogators must have been spectacularly stupid to have been fooled. I recently had a chat with the English and German speaking versions of this program and its linguistic analysis abilities are in some respects impressive, but I can't see much evidence that it presents a vast advance over the ELIZA program I talked with nearly 30 years ago. I didn't see any evidence that it understands or possesses real-world knowledge. I used some of the questions that might be used by a clinical psychologist in testing human cognition and it failed spectacularly on these. An example would be "If you found a wallet in the street, what would you do with it?" or "What does the following saying mean -'People in glass houses shouldn't throw stones'?"
Cecil Ward, Heast, Isle of Skye, Scotland
I gave this thing a try online and how anyone was tricked by it is beyond me. They must have been pretty stupid people. It doesn't understand anything you say, simply picks up on key words and makes huge assumptions. No advancement at all since those text based adventure games.
Richard, Coventry
Below is a selection of your comments.
Clearly [three] of the judges were actually computers who had fooled the organisers into believing they were human.
Leigh, London, UK
So basically AI is no further forward than it was 20 years ago. Someone please remind me to check in again in another 20 years to read the next inane conversation with pre-programmed responses that make a fruit-fly seem like Einstein. Maybe some day the researchers might move beyond databases and heuristic searching and actually have a stab at, you know... intelligence.
Robert Campbell, Edinburgh, UK
Elbot reminds me of the Eliza program from the mid-1970s, a psychoanalyst simulation. Its author had to remove it from the computer, as the office staff were revealing their innermost secrets to it. Any human comment it didn't understand got the standard reply of "go on". Apparently all the staff wanted was someone (or thing) to listen to them.
Gerald Schaefer, Eugene, OR, USA
The human interrogators must have been spectacularly stupid to have been fooled. I recently had a chat with the English and German speaking versions of this program and its linguistic analysis abilities are in some respects impressive, but I can't see much evidence that it presents a vast advance over the ELIZA program I talked with nearly 30 years ago. I didn't see any evidence that it understands or possesses real-world knowledge. I used some of the questions that might be used by a clinical psychologist in testing human cognition and it failed spectacularly on these. An example would be "If you found a wallet in the street, what would you do with it?" or "What does the following saying mean -'People in glass houses shouldn't throw stones'?"
Cecil Ward, Heast, Isle of Skye, Scotland
I gave this thing a try online and how anyone was tricked by it is beyond me. They must have been pretty stupid people. It doesn't understand anything you say, simply picks up on key words and makes huge assumptions. No advancement at all since those text based adventure games.
Richard, Coventry