From: AAAI Technical Report WS-96-03. Compilation copyright © 1996, AAAI (www.aaai.org). All rights reserved.
CyberspaceGameShowHosts : Agents for Socialization,
Kenrick Mock
Intel ArchitectureLabs,JF2-76
2111 NE25th Ave
Hillsboro, OR97124
kenrick_j_mock@ccm.jfintel.com
Leora Lawton
Bellcore
8 CorporatePI., Rm#3C-113
Piscataway, NJ 08854
leora@thumper.bellcore.com
Not Just Entertainment
Michelle Hoyle
Comp.Sci., Univ.ofZurich
WinterthurerStr. 190, CH8057Zurich, Switzerland
hoyle@ifi.unizh.ch
created fantasy personalities purportedto typify graphical
multi-user environments as specified through their
nickname.
Within the domainof IRC, we have created several
challenging trivia games.Thefocus of this paper is the
game "Risky Business" (or RiskyBus) which is
TM style trivia gamein whichusers attemptto be
Jeopardy!
the first to answertrivia questionsin order to winvirtual
moneyand virtual prizes (Qnittner, 1995; Sandberg,
1994). The gameis hosted by a computeragent or "bot"
programmed
in C that communicates
with IRCservers just
like a humanclient. OnEFnetservers, the host is named
"RobBot"and on Qnet servers, the hostess is named
"ReneeBot."In order to host the games,the bots require
knowledge regarding IRC commands, knowledge
regarding howthe gameis played, rules and etiquette to
maintainthe channel, and knowledge
for self-preservation
within the IRCenvironment. In the event that the bot
makesan error, humanoperators are capable of correcting
the game.Thebot also records individual player statistics
such as the numberof gameswon,and the high scores. In
addition to this knowledge,the bots also project distinct
personalities throughELIZA-likeresponses to user input
(Weizenbaum,1965). This knowledgemust currently
input by a humanoperator. A popular type of reply are
witty or humorousstatements about a specific player’s
interests.
A sampleof a gamebeing played through RobBot’sis
shownbelow:
Abstract
As the Internet has grownin size and popularity, the
Internet mediumhas changed from an educational
and technical content to a social and entertainment
content. TheLycossearch service estimates that the
numberof WWW
pages has grownfrom 5 million to
6.89 million pagesduringthe monthsof April to June
of 1995, and will reach 10 million pages by 1996. A
large numberof these pagesare devotedto personal
pages, whereasonly a few years ago personal pages
werenonexistent. Similarly, the EFnetInternet Relay
Chat (IRC) servers have grown from supporting
several hundredconcurrent users in 1991 to over
20,000 concurrent users in 1996. Both the WWW
andIRCare becoming
active vehicles of socialization
and entertainment. This paper focuses on the
technical and social aspects of a set of games
designedby the authors that run on IRC,and howAI
techniques can augmentthese games.In particular,
weexaminethe gameRiskyBusiness, an online trivia
gamehosted by a computerprogramor "bot," short
for robot. In addition to providingentertainment,the
computer gameshow host also supports a unique
setting that real-life gameshow hosts never
encounter: the opportunity to interact in real-time
with thousands of players, 24 hours a day, and
becomea cornerstone of their social interaction in
cyberspace.
Background Information
IRC is composed of client programs connected to a
networkof servers that facilitate textual conferencing.
Similar to a Multi-UserDungeon(MUD),(Mauldin, 1994)
users are able to "talk" to other groupsof online users
interactively in real-time. However,instead of the MUD’s
physical metaphor of rooms and locations, IRCusers
communicatewithin "channels" that typically cover a
specific topic (such as unix, sex, or chocolate). Aswith
virtual worlds whereusers take on the persona of their
avatars, in the IRCworldusers take on a newpersonaor
emphasizepart of their existing personaor interests, as
specified through their nickname.BecauseIRCis more
textually based, it maybe that there is less roomfor the
12
<RobBot>Current category: Footwear. Question Value:
800.
<RobBot>Question 5 of 30: Lowcut woman’sshoe or a
deviceto pass gasoline
<BrandEx>rob pump
<Texmex>rob pump
<RobBot>brandex: That is CORRECT!
You win 800.
Yourtotal is -300.
<RobBot>
Please wait while preparing the next Gullivers
Travelsquestion...
<jennew>brand rocks!
<RobBot>
Current category: GuUiversTravels. Question
Value: 400.
<RobBot>Category Comment:Trivia about Gullivers
Travels
<RobBot>
Question6 of 30: The only thing the Laputian
king wantedto learn about the outside world
<Texmex>
oh this one sux
<Math>
what food do you like rob
<RobBot>
Pass the ho-ho’s!
* MastrLionpasses out (muchto the relief of the channel
no doubt)
<Mach>rob mathematics
<MastrLion>rob tug
<RobBot>
mastrlion: Bzztf That is incorrect. Youlose
400. Yourtotal is -500.
<RobBot>mach: That is CORRECT!
You win 400. Your
total is 400.
RobBotrecognizestext prefacedby "Rob"as input relating
to the game. In most cases, the input data constitutes
commands
or a player’s answer to a trivia question.
However,the bot mayalso respondto text whichis not a
command,as in the example where "Mach" comments
about food, and the bot replies regarding ho-ho’s. These
replies give RobBota personality- perhapsoneas a junkfood addict. Notethat while the gameproceeds,users are
also socializing with each other; Texmexcomments
about
howhe dislikes the current category, whilejennewpraises
BrandExfor answeringa question correctly.
Bots as Agents
In manyrespects, the gamebot maybe viewed as an
intelligent agent. Foner (1995)describes crucial notions
of an agent in terms of autonomy,personalizability,
discourse, risk & trust, domain,graceful degradation,
cooperation,
anthropomorphism,
expectation,
entertainment and social needs.
Each of these
characteristics can also be understoodas facilitators for a
social setting. Asfacilitators, the characteristics can be
better understoodwhenthe purposefor the bot is clarified.
First, we note howthe following characteristics are
addressed by RobBot. Then, we will show howthese
characteristicsrelate to the social purposeandsociological
functioningof the game-playingenvironment.
Autonomy. RobBot hosts the game without human
intervention, although gameoperators are capable of
giving the bot commands.However, despite serious
neglectat times by the creators, the gamehas continuedto
run and flourish on its own.Thebot as an agent must be
able to nmindependentlyif it is to satisfy any value in
terms of entertainmentor social interaction. If a human
operator must constantly provide direction, then the bot
becomesa tool of the humanrather than a separate entity.
Dependence. Although not one of Foner’s crucial
13
notions of an agent, somedegree of dependenceof RobBot
on humansis importantfrom a technical and sociological
viewpoint. Technically, the bot requires somehuman
intervention for the gameto run smoothly.In turn, this
needfor dependence
invokesa sense of responsibility and
powerin the humanoperators that can satisfy a need for
control that transcends entertainment and leisure time
needs.
Personalizability.
RobBot maintains his own
personality through the responses programmed
into his
lexicon, and is capable of recording information about
other users. Thebot also maintainsdatabasefunctions in
terms of whois an operator, and what scores each player
has attained. Aplayer’s scores and record are important
measuresof one’s social status on the gamingchannels.
Additionally, by projecting life-like qualities onto the
robot, an atmosphere inducing socialization maybe
induced.
Risk and Trust, Graceful degradation. RobBotdoes
makeerrors while conducting the game. Players may
phrasean answerdifferently than the answerstored in the
answerdatabase, or spelling errors maybe present in the
answer database, resulting in the bot pronouncing a
correct answeras incorrect. Humanoperators are often
present to correct such errors. Since the domainof Risky
Businessis a game,risk is a lowfactor since the results of
an error do not havedire consequences.Often, incorrectly
interpreted responsesby the bot are foundto be humorous
by the players, or playerswill bandtogetherin cursingthe
bot for his errors.
Cooperation.Userand agent collaboration is crucial to
play the game.Moreworkis required so that a richer twoway communicationis constructed between player and
robot. Whilethe bot does not question users, it does
provide user feedbackto verify that command
requests are
being processed. This cooperationfacilitates a primitive
form of socialization betweenhumanplayer and computer
agent.
Anthropomorphism.
RobBot relies heavily on
anthropomorphism
in order to accomplishhis tasks as a
game show host. While the technology currently
implemented in RobBot consists of simple keyword
mappings,the importanttask for RobBotis not to pass the
TuringTest, but to provide entertainmentas a gameshow
host might. To this end, keywordsand canned phrases
appearto satisfy this requirement.
Sociological
Functioning of the Game-Playing
Environment
Foner stated that the bot Julia performedas an agent and
that there were"sociological" patterns to her behavior.
AlthoughFoner focused on the capabilities of Julia and
not the sociologicalimpactof the peopleinteracting with
Julia, he knewsomethingfundamentallysocial was going
on in interactive multi-person communications.Foiler
observed that meaningwas inferred by conversation, a
building block of social interaction. Healso saw that
context was another tool by whichweunderstand social
meaningand can decide howto behave and respond. In
what follows weshall discuss what sociological meaning
is occurring through languageand context, and howthe
agent-- the bot’s -- role can be designedto facilitate an
appropriatesocial climate, in this case, for entertainment
andsocializing.
The extant purposeof RobBotis to run a game,and the
gameis attractive to people for several reasons. People
need leisure time. It helps themrelax for constraining
social roles and reduce stress. Leisure takes manyforms
and no form of leisure can satisfy all people, and is
unlikelyto satisfy anyonepersonfor all his or her leisure
needs. Leisure maybe solitary or social, intellectual or
physical, passive or active, competitive or noncompetitive, etc. (Russell & Hultsman, 1988).
importantkind of leisure is social leisure time because
people need unstructured social time (Samdahl,1988).
Leisure activities must also be accessible. The Riskybus
gamefills a certain leisure niche (Lawton,1995). It
intellectual, social, engaging, active and can be
competitive. Becauseit is on-line, twenty-four hours a
day, sevendays a week,it is accessible to anyonewhohas
access to a computer,modem
and phoneline, and whohas
the basic knowledge
to use an on-line system.
As a core group of players engage in Riskybus, the
network then becomesmeaningfully ritualized in its
context, forming a sub-culture which can be then
transmitted to newplayers as they arrive. Components
of
a sub-culture include a recognition of the dominant
culture, somecontextualstability, sharedlanguage,history
and purpose. These in turn lead to mutually shared
norms(behaviorpatterns) and values (communal
goals).
RobBot’spurposeas agent, therefore, is to maintainthis
social game-playingenvironmentand let the subculture
continue and develop. Accordingly,the above-described
characteristics of an agent are also mechanismsthat
promote the sociological dynamicsas well. Wecan see
howthe characteristics of the agent supportthe necessary
sociological features necessary for an on-goingsocial
environment.
Autonomy,the first of these mechanisms, permits
convenienceand a stable structure. Withoutthe stability
of structure, it wouldbe difficult to knowhowto act in the
social context of the game, which would effectively
hamperthe development
of the social networks.Therules
more or less stay the same from day to day, monthto
month, so the responses that exist to the game
environment can be learned and becomepredictable.
14
The stability of the structure helps facilitate the
development of social history to the game-playing
environment.
Twomore characteristics,
Personalizability and
Anthropomorphism,
lead to the developmentof a shared
language. Thebot typically utters certain phrases, which
can then be used by participants as symbolsof events and
concepts. Thebot’s consistency in phrasingleads to the
game-playersacceptanceof them. Languageis one of the
keytransmittersof sub-culture,so it is importantthat there
is constant reinforcementof these phrases. WhenRobBot
expresses delight about chocolate via "*choco*"then all
participants can use *choco*as a keywordfor joy. The
standardization of lapguageby players in responseto the
bot’s languagealso adds significantly to the development
of sharedlanguage.Further, sharedreality is enhancedby
the bot’s record keeping capacity, creating a sense of
community
history and awareness.
The characteristic of cooperation helps shape the
behaviorpatterns of players and reinforces a definition of
social order. For example,swearingis frowneduponand
any player uttering certain wordswill receive a warning
"<Player 1> This is a family channel! Be warnedor I’ll
have to call the bouncers!" Persistence will get the
miscreantkickedoff the channel. Players then learn that
swearingis not to be tolerated in any large degree. In
another game, Acro, the bot provides no such guidance
and coarse languageis a common
feature of the players’
answers.(This feature resulted in a moralityplay between
certain players. Eventually those whocouldn’t deal with
occasionalvulgarityceasedplaying.)
Withthe characteristics of Risk andTrust and Graceful
Degradationwesee that RobBot’sflaws lead to a bonding
of players. Players advise one another on RobBot’s
unpredictabilityin certain categories. Byreferring to the
bot as a character ("stupid bot") the players recognizethe
shared reality that they face. This shared reality
contributesto part of the community
consciousness.
It should be clear that the point of the bot is not
necessarily to be a stand-in for a real person. Somebots
maysuccessfully accomplishtheir role by being even more
primitive than RobBot.For example,in Chaosor Boggle,
there is little if anyPersonalizationor Anthropomorphism.
Nevertheless, the social space (roomto chat) provided
the bot contributes to the development
of the game-playing
environment.Bogglehas little community
associated with
it: the mainattraction is the competitivenature of the
leisure time. Chaos has more social space, but the
structure of its bot precludesleisurely chatting time, even
though a core of committedplayers has developedon the
various networks.Accordingly,while certain friendships
may form and intensify in those game-playing
environments,
they are less likely becausefewerfeatures of
agents(andtheir underlyingsocial catalysts) are present.
By understanding that a real social need is being
facilitated, it is feasible to design entertainment
environmentsto meet other social needs, for example,
educational games. In such a game, designers could
combineeducation(curriculummaterial), play, self-esteem
enhancing
praise, etc., all structuredinto the bot’s design.
Thebot could also act as chaperonby chastising obscene
or other problematic language. Not only wouldthe game
be fun, but the process of learning through game-playing
ad the social environmentof the gamewouldreinforce
essential social skills in child andadolescentdevelopment.
Role of AI in Risky Business & Internet
Peopleflee to the Internet for a variety of reasons;perhaps
the most common
reason is that it’s touted as being the
technologicalrevolution of the 90s. Therecent growthof
Internet Service Providers indicates that the teenage
hacker, the aspiring businessman, and the greying
academicall desire to be on the net. However,people are
reluctant to change,andoneof the primecharacteristics of
computertechnologyis the speedthat it changes.People
long for the normal, for the usual, for the understood.
Therefore, whenthey traversed the electron boundaryto
the worldof the Internet, alreadya drastic changein itself,
they initially founda world far fromtheir expectations,
filled with evenmorealien symbolsand mystic rituals. A
larger influx of such peoplegavethemthe powerto start
creating that which they understood: bars,
neighbourhoods,
dating services. All those institutions in
the real worldthat cater to the social animalin mannow
havea place on the Internet.
Onerole of Artificial Intelligence in the cyberspace
worldis to providea familiar interface to the participants.
RobBot and the Risky Business channel provide an
environment
hospitable to peopleawashin the chaotic flux
of the Internet. Theartificial intelligence inherent in
RobBot,minimalas it is, creates a persona to which
people can relate becausehe is somethingunderstoodand
non-threatening. RobBotplays an essential part in
acclimatizing people to the world of the Internet. With
evenmoreartificial intelligence, the line betweenuser and
software artifact wouldbecomeeven further blurred,
flawlessly playing the ever jovial gameshow host,
providing witty repartee on demandor the answers to
tricky trivia questions.
AI Technology to Augment Robot Capabilities
Someof the areas where more advanced artificial
inteUigencetechniques can better meet RobBot’sgoals of
becominga better gameshowhost include improvements
in the core capabilities of autonomy,personalizability,
15
risk and trust, graceful degradation, cooperation, and
anthropomorphism.
Improvements
in these categories fall
under three broader areas: improved parsing and
knowledge-processingcapabilities to better understand
input text, a modelof the domainenvironmentto support
intelligent reasoning and autonomy,and mechanismsto
support other life-like qualities such as emotion,
personality, and beliefs. Althoughweare not currently
applying these technologies to RobBot,manyresearch
projects and methodologiesexist in these areas which
could be applied. Selection and implementationof the
specific technologiesremainsas future work.
The area of parsing and knowledge-processing
capabilities has been exploredby manyresearchers. Many
methodsexist for parsing include simple keywordmethods
(Weizenbaum,1965), statistical methods (Charniak,
1993), knowledge-basedapproaches (Riesbeck & Schank,
1989), or hybrid methods(Mock, 1996). Anyof these
could be implementedinto RobBotto improvehis natural
language understanding. Improvedunderstanding will
improvethe functionality of the bot, e.g., users couldgive
commands
to RobBotin natural language as opposedto a
set of formal commands. Additionally, stronger
capabilities in this area will improvegraceful degradation
since RobBotwill better understandthe input resulting in
fewer instances of miscommunication,
and provide for a
moresophisticated personality to enhancesocialization
and communication.
Notethat the systemdoes not haveto
be capableof passingthe TuringTest to be successful, but
only to understand enough material to facilitate an
entertaininginteraction withthe players.
Thearea of model-basedreasoningwill provide greater
autonomy. Currently, the bot has only a limited
understandingof his environmentand operations that may
be applied to the environment.Byproviding a modelto
RobBotof what the environmentlooks like (Hedges,1989)
andwhat operationsare available to navigate throughthe
environment,the bot can search the space to achieve a
goal. For example,if control is lost on the homechannel
of #RiskyBus, then by consulting the model of the
environmentand available operations, RobBotcan choose
to create a newchannelor invokea series of operations
whichmaylead to a state wherecontrol of the channelis
regained. Throughsearch of the problemspace and a set
of heuristics, RobBotmaybe able to pick the operation it
believesis mostlikely to result in its goalstate.
Thearea of supportinglife-like qualities in computer
agents is a newarea that has only recently becomepopular
due to more easily available computing power.
Consequently, this area has not been explored as
thoroughlyas the other twoareas, but a numberof projects
exist in this arena. Hayes-Roth’sworkwith the Virtual
Theater (1995) involves the creation of improvisational
computercharacters that can performlife-like activities
such as exhibition of intelligence, variability, idiosyncrasy,
affect, motivation, and personality. Other projects such as
ALIVE(Maes, 1994) can provide graphical interpretations
of feelings and emotions. To express nebulous areas such
as opinions or beliefs requires modelsof thought and the
construction
of argtunents (Alvarado, 1990). The
culmination of these types of technologies into RobBot
will result in increased expression, which provides
entertainment and also provides an environment that
facilitates socialization. Withthese types of capabilities,
players will be able to interact and socialize with RobBot
on a level of detail they were previously unable to.
Conclusion
Due to the complexity and scale of Internet systems,
computer agents have become a popular method to help
manageand navigate cyberspace. As the Internet becomes
more accessible to non-technical people, the application
and importance of entertainment on the Internet grows.
This paper has examined the IRC game "Risky Business,"
a multi-player trivia game that is hosted by a computer
bot. While the intelligence of the bot is currently very
simple, it is conducive to an entire frameworkof leisure
time socialization as well as entertainment.
Additional
AI technologies
may enhance the
socialization and communicationby making the bot more
personable and human-like. This can help bring the bot
closer to the humangame showhost, and ease the changes
necessary for humansto adapt to the online world. Since
the inception of Risky Business two years ago, we have
seen the evolution of cliques and friends whose ties to
each other seem to be as stable as those relationships
formedin life off of the Internet.
The underlying structures for communication and
socialization in IRC and Risky Business are the driving
technologies behind this behavior. These findings suggest
that support for socialization and communicationare key
factors in the success of entertainment systems, not just
technological advances such as high resolution graphics or
32-bit sound. Experimentsare also under investigation to
examine the application of the social-based gaming
environment toward educational systems.
Information about getting onto IRC is available on the
WWW
at http://www.kei.com/irc.html.
Information about
playing Risky Business is available on the WWW
at
http://phobos.cs.ucdavis.edu:8001/~mock/irc, html
and
http://www.cs.uregina.ca/~hoyle/Games/.
MA:Kluwer Academic Publishers
Charniak, E. 1993. Statistical
Cambridge, MA:MITPress.
LanguageLearning.
Foner, L. 1995. What’s an Agent, Anyway?A
Sociological Case Study. Available at
http://foner.www.media.mit.edu/people/foner/Julia/Juliaintro.html.
Hayes-Roth,B., Sincoff, E., Brownston,L., Huard, R. &
Lent, B. 1995. Directed Improvisation with Animated
Puppets. Proc. Of CHI 95 Conference on Human
ComputerInteraction.
Hodges, J. 1989. Device Representation for Modeling
Improvisation in MechanicalUse Situations. Proc. 1 lth
AnnMeeting Cognitive Science Soc. Hillsdale, NJ:
Lawrence Erlbaum.
Lawton, L. 1995. Entertainment on the information
superhighway:Internet gamesas accessible social leisure
time. Bellcore internal document.
Maes, P. 1995. Artificial Life meets Entertainment:
Interacting with Lifelike Autonomous
Agents. Special
Issue on NewHorizons of Commercialand Industrial AI,
v38,11, Communications of the ACM.
Mauldin, M. 1994. Chatterbots, Tinymuds, And The
Turing Test: Entering The Loebner Prize Competition.
Proceedingsof the 12th National Conferenceon Artificial
Intelligence.
Mock, K. 1996. Hybrid Hill-Climbing and KnowledgeBased Techniquesfor Intelligent NewsFiltering. To
appear in AAAI-96.
Quittner, J. 1995, April. AutomataNonGrata. Wired.
pp. 172-173.
Riesbeck, C. & Schank, R. 1989. lnside Case-based
Reasoning. Hillsdale, NJ: LawrenceErlbanm.
Russell, R. and Hultsman, J. 1988. An empirical basis for
determining the multidimensionalbasis of leisure. Leisure
Sciences, 10. pp. 69-76.
Samdalfl, D. 1988. A symbolic interactionist modelof
leisure: theory and empirical support. Leisure Sciences
10. pp. 27-39.
Sandberg, J. 1994, September20. "What Is Virtual
Reality, Alex?" Robot Hosts On-Line Quiz Show. The
Wall Street Journal, pp. B1.
Weizenbaum,J. 1965. ELIZA-- A computer program for
the study of natural language communicationbetween
man and machine. Communicationsof the ACM,v9,1, pp.
36-45.
References
Alvarado, S. J. 1990. UnderstandingEditorial Text: A
Computer Model of Argument Comprehension. Boston,
16