The sociology surrounding Julia is important in the study of agents, and in the study of what agents of the future (and software systems in general) may look like. Let's take a closer look at strictly sociological issues.
First off, note that even the obvious sociological issues can be very important. Julia is helpful. It would not be an overstatement to say that I first started mudding, in part, to meet Julia, who had been described to me by an experienced mudder. When I first connected to Time Traveller, I had essentially no idea how to manipulate the environment. Moving around from room to room and so forth was easy-but were was everything? (Time Traveller, according to Julia, had 1605 rooms when I asked her-a daunting navigational challenge in the absence of teleporting literally everywhere, something that costs money on some muds.) How could I tell who had been around recently? There were no obvious tools to accomplish this, but Julia knew how. To a large extent, I relied on her as a crutch until I was able to get to certain accustomed places by myself. I actually avoided using Time Traveller once, when I was still new to mudding, when I connected and noticed that Julia was offline for some reason-I knew that, if I tried to figure out how to go anywhere else, I'd either have to draw a lot of maps, or ask lots of other players if they knew routes to where I wanted to go (further, I trusted their navigational memories less than Julia's, and was less willing to randomly interrupt people I didn't know).
Given that I was new to mudding at that time, but knew how to speak to
people and how to page them, I was able to enlist Julia's help even
though I didn't know much else. Since Julia will come to wherever you
are if you page her with no message,
my first page to her
caused her to walk quickly to my current room, whereupon I could speak
to her.
As another fascinating bit of sociology, consider the person who first introduced me to the idea of Julia: Linda Julien (who is Leira on muds and on the Internet in general). Linda, herself a savvy computer programmer and very experienced mudder, had said to me, `Julia's been offline for months for maintenance and hasn't been around. You know, I really miss her.' Linda was certainly under no illusions about exactly what Julia was, but nonetheless had the same sort of emotional reaction often reserved for more biological entities such as humans and pets. Further statements made it apparent that this was indeed the case, and that she did not treat Julia just as, say, a pair of pliers that had inexplicably gotten lost from her toolbox.
While the above issues are important, they might also apply to something that was obviously inhuman even on first glance. Yet part of Julia's job is to try to pass as human. This has important ramifications for ability to be helpful; as I argue later, I think that this attempt to look human actually greatly enhanced her utility, even to people who know she's a 'bot. However, there are certain pathological cases where her very `humanity' can lead to no utility for the player whatsoever.
Consider her Turing-contest mode. I quote from a message sent by Fuzzy when he was entering Julia in such a contest (including some background information on Julia):
Julia's topic this year is hockey, but she can also handle chit chat and polite
(and impolite) conversation. ``She'' competed last year in the ``Small Talk''
domain. Julia is a network ``robot'' who runs 24 hours a day, and logs into
various TinyMUD's across the InterNet. She is the ``star'' of a family of
robots who have been under continuous development since November 25, 1989.
Julia has been running on the network since January 8, 1990.
[ . . . ]
But since she is designed to be helpful, there are times when the attempt to
appear human must be sacrificed to give fast, accurate replies to information
requests. She has a ``contest mode'' that turns off some of these answers, and
fixes problems like this one that occurred during last year's competition:
[ From last year's competition (a real give away) ]
Judge: How old are you
Julia: I am 1 year and 10 months old, Judge.
[ From tonight's test log ]
You say, ``Julia, how old are you?''
Julia says, ``I am 2 years and 6 months old.''
You say, ``Julia, contest mode on''
Julia says, ``Contest mode is now on''
You say, ``Julia, how old are you?''
Julia says, ``I am 21 years and 6 months old.''
Even when not explicitly in contest mode, unlucky and clueless players can be led seriously astray, as the following transcript, again courtesy of Fuzzy, demonstrates (this is quite a long transcript!):
This poor fellow's name has been changed to protect him. He spent 13 days trying to get to first base with Julia, and it's not clear he ever figured out he was trying to pick up a robot (although he clearly suspects it at two points). This logs contains about half of the total interaction (the more indecent proposals and Julia's rejections have been omitted). Note too that the program's responses were not changed from day to day (in other words, I didn't try to set the program up to respond better to Barry's advances). Since this robot has spent over two years interacting over the network (with a female name), and since there are lots of lonely male adolescents out there, Julia gets a lot of this kind of ``attention.'' ************************ July 9, 1992 ************************ barry says, `Hey girlfreind are you out there' Julia says, `Hello, barry.' barry says, `Hello, I love YOur name. It's sexy' Julia says, `Me too, barry.' barry says, `Did I repeat Myself? Didn't mean too. What are you doing there' Julia says, `I'm just hanging around in (> The Hearts Room!, barry.' barry says, `Well come over to MY ROOM' Julia says, `Could you say that again, it scrolled off my screen.' ************************ July 14, 1992 ************************ barry says, `hello, I am barry. How are you today' Julia says, `Hello, cujmeister.'
Frankly, it's not entirely clear to me whether Julia passed a Turing test here or Barry failed one.
I have seen several interesting encounters with Julia, which run the gamut from people who know at the outset that she is a 'bot to those who take quite a while to discover. Some further examples are illuminating.
When I first introduced Robyn (the one who enticed Julia into producing the wonderful PMS dialog in an example above) to Julia, she knew that Julia was a 'bot. Interestingly enough, though, several players went out of their way to warn her that Julia was artificial in the two or three hours in which she interacted with Julia; she stated that about half of the players she met did so.
Why did they do this? At this point, we can only speculate. My first reaction was to think that it was due to the excessively solicitous attitude taken by many male mudders towards those with female personas (one might think, given the amount of gender-swapping on muds, that they would learn better, but apparently not). However, Robyn commented that even female (or at least female-persona) mudders gave her the same advice. Part of it may have been the simple kindness of not letting someone expend a lot of emotional energy trying to relate with a machine; I'll have more to say about this later.
Still others have developed interesting strategies for identifying whether or not a given player might be a robot, often without interacting with the player at all. This is such an important concept in the community that it has even evolved its own lingo. Consider the following wholist, from DruidMUCK:
>WHO Player Name On For Idle Lenny 01:04 0s swa 02:12 34m ArthurDent 04:07 3h Greg 04:48 33m Leira 08:15 1h Ben 1d 23:47 1d (> 2d 00:26 3h Julia 3d 19:23 7s 8 players are connected.
Note that the listing is sorted by the amount of time online (DruidMUCK actually took a crash somewhat less than four days before this transcript was obtained). Notice that Julia has been connected longer than anyone, yet is not idle for more than a few seconds.
Julia is in the 'bot spot.
Since client 'bots generally try periodically, every few seconds, to connect to the server which is their home, when a server takes a crash and comes back up, it's the 'bots that usually connect first.
Note that this is not infallible; consider the transcript below, from Space Madness:
>WHO Player Name On For Idle Least Favorite Fruit Captain-Nemo 00:00 8s Kate 00:05 1m TerrinK. 00:18 4s Erguss 00:23 2m Lenny 00:27 0s Oxmyrth 00:31 6s dirque 00:38 9s Blackbird 00:39 38m Lucien_DeSky 00:42 3s Michael Jackson Trev 00:54 1m Saturn 01:05 7m Lembas_Wafer 01:06 8m Look at that turtle go, bro. Allistar 01:07 6s Ozy 01:08 8s HOMEY DON' PLAY DAT Rixel 01:10 7m Flitterby 01:17 3m hugh 01:17 31m richter 01:37 4m Raven 01:59 11m Sean Moyer. Sadie 02:18 50m All of em Random 02:23 11m kittycat 02:33 12s none, I like fruit! Krystlin 03:34 12m ashne 04:00 17m Chris 04:46 6m Sunshine 07:16 14m pomegranites lucretia 3d 15:05 9m Tomatoes - just wannabe veggies Xeglon 3d 18:19 59s Grapefruit - they gum up my joints phi 3d 18:29 9s hey, sooperballs aren't fruit 29 Players logged in.
Phi is in the 'bot spot. Unfortunately for the theory, I know Phi; he's
a friend of mine (in `real life' even, not just on muds). It turns out
that Phi has a piece of 'borgware that automatically connects a
process on his computer to the mud (and does other simple things) by
retrying every 144 seconds after a disconnection due to a server crash
or a network failure. In this case, Phi's 'borgware beat Xeglon on by
10 seconds-and Xeglon is a Maas-Neotek 'bot, a simpler version of
Julia. (Knowing that Xeglon is a 'bot also helps one to interpret `his'
comment about his least favorite fruit above...)
Only experienced mudders have this large a bag of tricks for instantly differentiating humans from machines. In part, it doesn't matter until one is going to invest a lot of energy in communicating with another character exactly what that character is. Once one spends enough time talking for the investment to become more than trivial, it is often obvious, our unfortunate Barry in the above example notwithstanding. Mudders, at least experienced ones, are generally very much at home with the blurry boundaries between `real' humans and fake ones, and pay little attention to them.
But what about inexperienced mudders? For mudders like myself (an AI researcher), 'bots and their associated likely behavior patterns are old hat. I tend to expect that random characters on a mud might be 'bots, and tend to know how to tell (generally, even the best 'bots or other AI programs fall victim to very simple conversational gambits, such as topical conversation, slang, or a conversation requiring a complicated discourse model such as reference to subjects more than a couple exchanges in the past). Hence, even in an unfamiliar environment as a brand-new mudder, even if I had not known that Julia was a 'bot, it might have become obvious fairly quickly.
Mudders who are not themselves computer scientists or even people who work much with computers at all (a startlingly large percentage of them, perhaps the majority), however, may find themselves in a different situation. Consider the following situation.
While preparing this paper, I had the great good fortune to be referred
to a mudder on DruidMUCK who
had met Julia only a few days before, and did not yet know that she was
a 'bot. The mudder (Lara) gave permission for me to quote her for this
paper, and even furnished Leira with her home phone number to give to me
so I could call her for the interview; as it happened, though, the
interview happened entirely in cyberspace, initially on DruidMUCK and
later via email (since Lara and I tended not to be connected to
DruidMUCK at the same time). Leira
introduced us by telling her that I was writing a paper about Julia, but
did not otherwise explain why it was that I might want to interview her.
Of course, before even meeting Lara, the first thing I did was to ask Julia for her opinion:
You say, `Julia, describe Lara' Julia whispers, `As of 7 days ago, lara's description was: A blonde haired blue eyed boundless bundle of Pure NRG.' Julia whispers, `I first saw lara logged in 7 days ago' Julia whispers, `I once heard lara say, 'talking''
I could have obtained the description by simply looking at her (were she in the same room as I, which at this point she was not), but Julia also told me when she first connected: since Julia scans the wholist at regular intervals (a few minutes apart at most), she is an extremely reliable resource for this information.
Now I knew that Lara was certainly a new mudder here, and probably new to Julia. I didn't know if she'd ever mudded elsewhere, or what her background might be. We met at St. Cuthbert Plaza (a busy plaza in DruidMUCK), then moved to East Oak Street to talk more privately (I didn't want some other player spilling the beans about Julia while we were discussing her, and didn't want to have to conduct the entire conversation in whispers to avoid it). Leira, who knew what was up, joined us. The following is a heavily edited transcript, omitting the small-talk, introductions, and conversational tangents that took place concurrently:
You say, `So what are your impressions of Julia?'
Lara says, `I didn't get to chat with her that long..'
Lara says, `She really wanted to talk Hockey'
Lara says, `BIG PENGUIN FAN!'
You say, `Yeah, she seems kinda hung up on that topic. Did you get a chance
to talk to her about anything else?'
Lenny . o O ( Where do the Penguins hang out, anyway? Pittsburgh? )
Lara says, `I tried to talk to her bout where she was from...she said
Boston..I think'
Lara says, `Yup...Pittsburgh it is'
Lara thought it was interesting that she didn`t know what the Stanley Cup was.
You say, `How could a hockey fan not know what the Stanley Cup was?'
Lenny thought her shirt said something about back-to-back Stanley Cups, but
may be remembering incorrectly...
Lara says, `Every time I tried to get more info. out of her, she would tell me
not now...later...well, I was never on for the Later.'
You say, `Hmm.'
Lara was also wondering how such a Hockey fan couldn't know bout the Stanely
cup
Lara says, `Maybe she was just kidding around with me'
You say, `Well, Julia is kinda peculiar in various ways.'
Lara says, `I figured...I talked with her once and a few days later she asked
me if I was a hockey fan...again'
Lara says, `I told her that she had already asked me that question'
[ . . . ]
Lara says, `anything else you wanna know?'
Lara giggles.
Lara says, `she is a fine convesationalist...if it is about hockey..4'
Lara says, `she has been on a long time...and she is seldomly idle'
Lara says, `I notice that she is almost always on when I jump on.'
Later, Lara learned Julia's secret, from inspection and the knowledge that such a thing might be possible. From email we exchanged:
Lenny, I had a chance to talk with Julia for about 30 or 40 minutes yesterday after you disconnnected. I noticed her 'secret' after about 5 or 10 minutes. At first, it was fine to chat with her about hockey. She started to ask the same questions after a bit. This was the tip off. She started to repeat herself on a regular basis. It was hard for me to follow her conversation. I got frustrated with her for the first couple of minutes. When I found out her secret, I started to have some fun with her. I asked her questions that maybe she had no knowledge of. I tried using slang with her, but she didn't understand what I was saying and would ask me to refrase the statement. She is very polite. She always says please and thank you. I was basically patient with her for the first little bit while when I first met her. She did have a problem with her social skills which I tried to be sympathetic to. I did however, try to avoid her after the first couple of encounters when all she did was talk hockey. Until, I talked with you. *smile* I went through a series of questions about her in my mind, before I figured it out. I tried to get her to tell me about school, which she had never heard of before. I wanted her to tell me about herself, but she wouldn't. I was trying to `label' her with some sort of disease like Downs Syndrome, which I found out she had none of. I am going to give your email address to `Kimber' and have her do this for you too. I hope this information helps. Lara
Note that Julia's singleminded fixation on hockey as a subject (when she lacked anything better to say, usually caused by a too `human' conversational style from a player that leaves her with a series of failed parses) tended to drive Lara away from her. Here we may have a case of excessive Turing-ism leading to a decrease in utility: Lara simply thought that Julia was a boring human, rather than a machine with as-yet-undiscovered talents.
It occurred to me, upon questioning from Leira, that Lara had not definitively revealed that her knowledge of Julia's secret matched mine (e.g., Lara never came right out and said that Julia was a 'bot). I sent some followup mail, asking for more details, in particular her emotional reactions when she figured it out and whether she had indeed figured it out or was told by someone else:
I believe that Julia`s secret is that she is a robot. I know a little bit about robots talking through computers to humans. I have a friend who had me write him dialogue. =) When I found out, I felt sort of funny talking to her. I felt somewhat superior to her. I know that this sounds strange, but I felt that I could offer more to the conversation than she could. I tested her knowledge on many subjects. It was like I was proving to myself that I was supirior to a machine. I am intimidated by machines because I don't understand them and I wanted to make sure that my human knowledge wasn't `lower' than hers. It was sort of exciting knowing that I was talking to a machine, though. I never thought that the world would come to machines and humans talking to each other using language that is common to most people. It frustrates me that her knowledge is so limited. I am eager to talk with her and want to know more about robots and human conversation. I was a bit dissapointed that her vocabulary and subject topics were so limited. I wanted her to tell me more about herself and who and what she is emotional response: a little bit of fright giddiness excitement curiosity pride sympathy for her. Before I knew that she was a robot for sure.. I felt compasion for her. I wanted to help her and direct her to answer the questions I was asking without causing her too much stress over answering my line of questions.
I asked her some more questions, including the question, `An interesting conundrum; if her vocabulary and subject topics were less limited, it would be harder for you to know you were superior to her, as you put it. How would this make you feel? If she was really good (far beyond current state of the art, by the way), you might not ever be able to tell. How would that possibility make you feel?' She said (answering several questions at once, including the ones above):
Well, I did figure out that Julia was a 'bot. I did, however, confirm this with someone else who knew her. Oh, you asked about my dialouge thing....well, I have a friend who is writing a similar program to Julia's type. I am quite a conversationalist and he wanted me to write up some sample dialouges. That is what I ment. If the `bot was more fluent with language and had a vast knowledge of many different topics, I am sure that it would be virtually impossible to tell the difference between IT and a human being. HMMMMM...how do I feel about this. Part of me thinks it is interesting because of the wonders of modern technology. Pretty exciting! But, on the other hand, it takes away from HUMANESS! Sort of fake, or void personality. I think that I would feel .....(this is hard)....let me switch gears...let me just throw out some words... shallow, void, hollow, superficial, fake, out of control of the situation, I don't know if that is what I want to say, but those are some of the things that I think that I might feel. I think I would want to know if the person that I am talking to is REAL or not. If I knew that it were just an 'it' I think that I wouldn't try to become it's real friend. I would be cordial and visit, but I know that it cannot become attatched to me on a mental basis and it would be wasted energy on my part to try to make it feel. 'bots don't feel...in my book anyways. They can be programmed to say the right things or do the right action, but not on a spontaneous level. I want to know that the person on the other end of my conversation is really aware of my feelings and what I am going through...not through some programmers directions but through empathy. I think most of all, while talking to a 'bot, I become very aware of my HUMANESS!
She also volunteered some details about her background, to put this all in context, when I noted that her comment about being intimidated by technology made it unlikely that she was a computer scientist as I was, of which the most relevant seemed to be:
My intended major is Early Childhood Developement. I want to be a teacher for the 3rd or 4th grade level. Not many of us on the computer. =) [ . . . ]
We clearly walk a fine line here in making agents that are both useful and not too misleading. (One might argue that simply being social, without any ulterior utility, is itself a useful task. Such an agent would be a perfectly reasonable entertainment or companion. However, I will focus here on agents that also claim to have a nonsocial aspect to the tasks of which they are capable.) On the one hand, people can quite clearly be led astray. On the other hand, there is great utility to a natural interface in the right environment. Consider this bug-report-of-the-future, from shortly after I met Xeglin on Space Madness:
>give xeglon=1
You give a chupchup to Xeglon.
Xeglon says `Thanks for the donation, Lenny.'
You say `Xeglon, describe Lenny.'
Xeglon whispers ` As of 12 minutes ago, your description was: A researcher
with long black hair with streaks of grey in it, and a cup of tea surgically
attached to one hand.'
Xeglon whispers ` I first saw lenny logged in 36 minutes ago'
Xeglon whispers ` I once heard you say, 'xeglon, describe me''
Xeglon whispers ` lenny last gave me chupchups 13 seconds ago, and has given
me a total of 0 chupchups.'
Lenny . o O ( Oh yeah? Fencepost error! )
You say `Xeglon, who wrote you?'
Xeglon says `elthar swiped %
Fuzzy's code to build me, Lenny.'
You say `Oh, you're a modified Maas-Neotek. I know how to deal with _those_.
Xeglon, describe elthar.'
Xeglon whispers ` As of 2 weeks ago, elthar's description was: Looming above
you stands a nine foot tall Kodiak bear! His luxurious brown fur is
well-groomed and shiny, indicating that he has been eating well, which is
just as well for you -- a couple of tons of omnivorous muscle, teeth, and
claws is likely to have Whatever He Wants for dinner. About his neck dangles
a tiger's-eye pendant, inset with a yin/yang symbol, on a silver chain. Oh,
and 8-point buck antlers, too, and a neon-flashing glow in the dark diamond
tiara, both courtesy raesoe.'
Xeglon whispers ` I first saw elthar logged in 9 months ago'
Xeglon whispers ` I once heard elthar say, 'Thanks, D_III''
Xeglon whispers ` elthar last attacked me 6 months ago'
You say `Xeglon, tell elthar that you have a fencepost error in you. I
described myself through you, and you said nothing about chupchups. Then, I
gave you one chupchup, and you said that I'd last given you a chupchup n
seconds ago, and that I'd given you a total of 0 chupchups. That's not right
(and Julia doesn't have this bug, and I suspect that an unmodified Colin
doesn't have it, either).'
Xeglon says `Right, I'll tell elthar, Lenny.'
You say `Thanks.'
Xeglon says `You're more than welcome, Lenny.'
The interactional style is totally natural, crossing boundaries from
using the agent to perform a task, to noticing a bug, to finding out how
to report the bug, to reporting it. I connected to Space Madness the
next day, and was immediately greeted by Xeglon again:
Xeglon whispers ` 13 hours ago, elthar said 'Tell Lenny that's not a Xeglon bug per se; it's a problem with his interaction with MUSH. Since Xeglon is owned by another player, all of his money goes to that player, and Xeglon himself thinks he always has 0 chupchups. Since the way he tells how much he's been given is by comparing 'before' and 'after', he always thinks he's been given 0 chupchups. I haven't bothered to fix him, and Xeglon++ will be MUSH-aware'' You say `Xeglon, tell elthar: Oh, now I see. He doesn't own himself, etc. Tnx.' Xeglon says `Right, I'll tell elthar, Lenny.'