Socially Adept Technologies

hurtpotatocreekΑσφάλεια

5 Νοε 2013 (πριν από 3 χρόνια και 5 μήνες)

43 εμφανίσεις

Socially Adept Technologies

Steve Marsh

National Research Council Canada

steve.marsh@nrc.ca


http://www.stephenmarsh.ca/


March 21st, 2002

Motivation


An introduction to the field


Pointers to relevant work


Questions about ‘suitability’


Suggestions for future projects


A wake
-
up call


A ‘call to arms’


Propaganda :
-
)



Outline


Introduction


What is Social Adeptness?


What is a Socially Adept Technology?


Examples of work in Social Adeptness


Questions


Problems


Answers?


Conclusions and more questions (from you? :
-
)


Here’s a thought...

Introductions

When an individual enters the presence of others,
they commonly seek to acquire information about
him or bring into play information about him
already possessed ... Information about the
individual helps to define the situation, enabling
others to know in advance what he will expect of
them and what they may expect of him.


GOFFMAN, E., page 13, 1959; “
The Presentation of Self in Everyday Life
”, Penguin: Middlesex.

So? What does this mean?


Humans have a sense of self (Mead) and through this
they adapt to situations and decide how to interact with
others (by trying to figure out the ‘self’ of the other
person)


Human
-
human interaction is social


It is also cultural


Our culture dictates what is and what is not acceptable in
given

situations


Novel situations are handled by prior similar situations if
possible (cf scripts)


And there may be rules, implicit or explicit, we follow...

Fair enough, but so what?

We’re AI (agents?!) people


This ability to behave socially and culturally correctly
towards people (or entities) with whom we are interacting is
what we call
‘social adeptness’



What’s en entity?


People, animals, agents…


So what?


So, we argue that if this social adeptness works so well for humans,
why shouldn’t it work just as well for machines?


(and we’re not the only ones, as you’ll see)


So, in other words, human
-
human
and

human machine
and

machine
-
machine interactions are social

Social Adeptness


One sensible definition of social adeptness is:



The ability to behave correctly in any given situation
according to the culture of the agents with whom one
is interacting in any social setting.



And
correctly

means sensibly, carefully, and in an
‘expected’ manner (which is, granted, a slight
tautology…)

What is it, in a practical sense?


Social adeptness is an understanding of, reasoning
about, and behaviour according to social norms such
as:


Ethics


Emotions


Morality


Trust


Personality


A sense of self


A sense of others


Cultural awareness


Social awareness (which is often the same thing)






(Marsh, 1995)

Socially Adept Technologies (or Agents)?


A Socially Adept Technology is capable of reasoning
with these norms in order to determine correct
behaviour in any given interaction


The word
interaction

is important here…


Correct social behaviour may or may not be necessary in
private

(Remember the tree falling in an empty forest?)


But, interactions can be asynchronous (as with email)


Basically, the onus is on us to behave correctly toward
those we are interacting with

?


It should be clear that, in order to allow technologies
(agents, interfaces, etc.) to reason with these social
norms, we need to have formal, or at least
computationally tractable, models of them


Obtaining these models is the goal of research in
Social Adeptness…


Interdisciplinary


Wide ranging


And hard… :
-
)

Actually, it’s quite easy to imagine...


Imagine:


Cellphones that don’t ring when you’re at the theatre (or at a
lecture…)


Robotic vacuum cleaners that don’t vacuum when you’re in
the middle of a dinner party


… or a good movie…


Interfaces that can adapt to your mood


Tools that can help you interact with people from different
cultures, in various situations


Agents that organise meetings according to your personal
requirements


… and all without ever having to be
told

what is right…


All of these are Socially Adept Technologies...

… whilst quite hard to do


Work in the field is inherently multidisciplinary, ranging
over topics such as


Philosophy


Sociology


Computer science


AI


Psychology


etc...

Work in Social Adeptness


Several researchers are working in and around the
topic. To name some (and discuss fewer)...


Socially Intelligent Agents


Dautenhahn (Hertfordshire, England)


Artificial Morality


Danielson (UBC)


Trust


Marsh, Dibben (St Andrews, Scotland), Davenport (Napier,
Scotland), Esfandiari, Chdrasekharan (Carleton),
Castelfranchi (NRC Italy)


Personality


Meech (AmikaNow!), Reeves & Nass (Stanford)


Interface Agents


Extempo (Hayes
-
Roth), Microsoft


Socially Intelligent Agents


Prinicipally, this is Kerstin Dautenhahn’s work


Dautenhahn has organised several workshops in this
field, and a book is forthcoming


See her web pages for details


The basic premise is similar to SATs


Make systems that can interact properly with humans


Robotics in fact play a large part in this work


Dautenhahn’s web pages can be found at:

http://homepages.feis.herts.ac.uk/~comqkd/

Artificial Morality


Peter Danielson, UBC (see his book:
Artificial Morality:
Virtuous Robots for Virtual Worlds
, Routledge, 1992)


Danielson provides simple agents with an
understanding of morality and its workings


He has extended his work in several very interesting
areas (including ecology and the business world)


See his web pages at:

http://www.ethics.ubc.ca/pad/

Trust


(A topic dear to my heart…!)


Introduced for autonomous agents in 1991
-
2 (Marsh)


Trust is the basis of sociability


Without trust, society would cease to exist (Bok)


Thus, an understanding of and concrete implementations of
trust are vitally important to the study of Social Adeptness,
acting as the keystone of a Socially Adept (or Intelligent)
Technological thrust


As evidence of its importance, it has received more
attention than most other SA attributes, especially
recently (because of E
-
Commerce, which we’ll come to
sooner or later…)

Trust contd.
-

Other work


The past 3 years have had workshops on Deception,
Fraud and Trust in Agent Societies organised at the
Autonomous Agents conferences


The proceedings from these workshops are an
invaluable aid to finding out more

Personality


An understanding of and subsequent representation of
personality is an important part of any interaction


This applies to human
-
technology interactions just as much as
human
-
human


The most visible work in this area is that of Reeves
and Nass from Stanford, reported in their book,
The
Media Equation


A significant result from this work was that people like
to interact with systems that show the same
personality as them


e.g., dominant with dominant, submissive with submissive, etc.

Personality contd.
-

Reeves and Nass


The Media Equation presents compelling evidence for
this and other findings


Although subsequent work may have put this in doubt…


Quite simply however, people anthropomorphise


They ascribe personalities to technology


(do you talk to your car?)


… and because of this they find it easier to interact with
(and put up with) the technology


This is powerful stuff
-

understanding it gives us a key
to designing more acceptable systems and interfaces

Personality contd.
-

Meech


Meech, for his thesis, looks into the Media Equation’s
results and takes them further into the design of
human
-
computer interfaces


The conclusions drawn are similar
-

that people like to
interact with like personalities


Such personalities can also be promoted, even in
textual interfaces, with different wording, emphasis,
etc.


We are using this work in a novel web site
architecture, as you’ll see later

Time for a look back...


There are many more examples of work in Social Adeptness


Too many to cover in this talk, including:


Emergent behaviour (Artificial Life, studies of societies…)


Emotions (e.g. Roz Picard’s Affective Computing, MIT)


Attention
-
based systems (e.g. Roel Vertegaal, Queens U)


Narrative and communicative systems (Bickmore &
Cassel, MIT; Mateas and Sengers, CMU)


Systems that make jokes… (Kim Binstead, Sony)


etc.


Bringing them together under a single moniker is worthwhile and
informative


…and forward, and ...


Given what we have seen so far in the area, it’s time to
think about the ultimate goal (implicit or explicit,
worked towards or not) of this combined research:


The creation of (potentially physically) embodied social
agents capable of existing in the ‘real’ human social
world, behaving correctly according to the norms of
society and culture.



Such agents may not be
artificially intelligent
, but they
will undoubtedly be
socially intelligent


But we’re a way away from that yet

… sideways


Given the theoretical work, what’s being done
practically?

(Now we come to the promotional part…)


My work at NRC is specifically concerned with
Socially Adept Technology, its uses and how to apply
it to different avenues of work


This will proceed in a new lab, with the code name
Project Mole Rat :
-
)


I’ll discuss some of the relevant work here


A Prolegomenon for all Future Social
Technologies Research…


(with apologies fo Immanuel Kant)


I believe (and you’re free to disagree…):


Technology should be seen as a social actor (cf. Reeves and Nass)


Incorporation of social norms into technology can result in increased user
comfort and efficiency (no second guessing)


Social norms can be incorporated both in the interface between human and
technology, but also


Within the technology itself


In the interface between technologies


Many of these beliefs stem from my focus on Multi
-
Agent Systems (in
itself a ‘European’ concept)

Trust, contd.
-

Marsh


My own work in trust was devoted towards


Better understanding how
cooperative

trust worked


Developing a
computationally tractable

formalisation of
Trust


Allowing for trust reasoning agents


Allowing for social science studies involving formal
models of trust


Implementing and testing the model


For in
-
depth details, see the website

Formalising Trust?


Some basic terminology:


Trusting entities have 3 kinds of trust:


Basic



T
x


The amount of trust you might have in the world


General



T
x
(
y
)


Trust you have in a specific person in general


Situational


T
x
(
y
,
a
)


Trust you have in a specific person in a specific
situation


Trust values are in the range [
-
1,+1) (now, is that odd?)

Formalising Trust?


Formal models can be used to model trust through
interactions:


T
x
(
y
,
a
) =

U
x
(a)
* I
x
(
a
)

*

T
x
(
y
)




Cooperation threshold:


C_T
x
(
y
,
a
) = (R
x
(
a
) / C
x
(
y
,
a
)) * I
x
(
a
)




Marsh(1994) see: http:// www.iit.nrc.ca/~steve/pubs/Trust

But


The models aren’t perfect


They were never meant to be


But they are simple


And they do work (even with humans (Dibben, 1998))


We’re applying them to E
-
Commerce, as you’ll see
later

ACORN

Portions of this work were carried out in collaboration
with researchers at University of New Brunswick’s
Faculty of Computer Science…

Thanks to Profs Ali Ghorbani and Virendra Bhavsar in
particular, who have taken ACORN from its humble
beginnings to new heights…

Students and programmers that have worked on this
project are: Youssef Masrour, Hui Yu, Leigh Wetmore,
and Jonathan Carter.

ACORN
-

Introduction and Motivation


ACORN is a ‘tool
-
based’ SAT, whose relevance becomes clearer
with some thought


ACORN is a peer to peer multi
-
mobile
-
agent architecture based
on community
-
oriented communication paths in human society
-

Stanley Milgram’s Small World Problem


(how many buzzwords do you need in a sentence…?)


ACORN was conceived as a replacement for ‘static’ information
systems such as bog standard email and static web servers


We see information in this sense as a dynamic entity which
has to work to exist in the world…


ACORN is ‘one of those’ acronyms: Agent
-
based Community
Oriented Routing Network

ACORN
-

Basics


In ACORN,
every
piece of information is (potentially
represented by) an autonomous mobile agent
-

the
InfoAgent


sounds, images, movies, frames, documents and parts of
documents, files, links, and so on…


Note
-

anything you can send via email, you can send in
ACORN too


Every InfoAgent carries with it


metadada for and a link to its information (not necessarily the
information itself)


owner information (it is given)


community information (it learns and can be given)


community ‘paths’ (it builds itself and can be given)

Uses of ACORN


As an email replacement
-

‘email with attitude…’


As community building and enhancing technology


As a people finder


As a novel peer review system


As a personalised directed information architecture


(directed ads, anyone...?)


B2B and B2C applications

Current Status


ACORN is fully implemented in Java (uses JSP)


There will be a port to C this summer


Development is ongoing in privacy, anonymity, and
‘thin’ InfoAgents


Integration of summarisation and additional search
technologies are also ongoing


Socially Adept Web Sites


An application of SAT to adaptive web site technology, this project aims
to show how some simple rules can be applied to already existing
technology in order to facilitate its better usage and integration in society


It’s also an approach to answering Etzioni’s (1997) call for adaptive webs


Finally, although it is applied presently to eCommerce and web
interfaces, we believe some at least of what we’ve learned can be
applied to other interfaces.


This work was carried out jointly with John Meech. Our thanks also goes
to Ala’a Dabbour for a first implementation of the prototype site

Static Trust Factors in E
-
Commerce


Seals of Approval


Brand


Navigation


Fulfillment


Presentation


Technology

Studio Archetype/Sapient

Web Site As Agent


Web site acts as an intelligent, adaptive interface
-

can
be viewed as an agent


Constructs a user profile from interactions, history and
other data


Uses models of trust, personality and context to
evaluate user behaviour


Adapts web page content/structure accordingly


A prototype site has been developed for this paradigm

Web Site Architecture

SociAware


SociAware is ‘Socially Aware’ technology
-

a simple
means of thinking realistically about SATs in general.


It was first introduced at MICON in August 2001


The most basic aspect of SociAware is the extension
of the trust model in simple ways to enable social trust
reasoning (that is, to allow society to reason about how
it trusts ‘things’ such as information.


SociAware Applications: infoDNA


A standard of Trust in information agents, implemented
as an extension to the ACORN architecture


Problem: agents judging information in ACORN…


i.e., which pieces of information to forward to owner, and
which to discard


Solution: each piece of information is socially rated


Then each agent can use these ratings in decision
making


Note that this solution is not perfect


Societies can be fooled into believing things that are not
true…

ACORN and infoDNA


Each piece of information carries with it additional infoDNA:


Originator and signature


Set of reader ratings and signatures


Ratings in our system are [
-
1,+1) but any suitable representation
would work


Agents can judge information based on these societal rankings


Naturally, much more information is also available


Owner of information


Metadata


This is a simple application but worthwhile, also it gives us a set
of results to work with when implementing more complex
approaches


Social Web Technology


Socially Adept Web Site adapts to User personality,
trust


However, initial stages of adaptation are problematic


Unknown user, unknown requirements


Site strange to user


Potential privacy concerns with adaptation, user profile


Using SociAware technology, we will be addressing
these concerns

Social Web Technology


User represented by SociAware Agent


This maintains user profile


Site represented by Site Agent


At first visit, negotiations between user and site agents result in
pre
-
built user profile with no identifying capacity (except through
user agent, which reveals only what is necessary)


In addition, because SociAware, user agent can query society
(e.g. via SociAware server) for views on site policies, etc.


Socia?Aware server maintains data. Also becomes indispensable
in browsing new unknown sites


Other value added
-

negotiation via SociAware server preserves
even more privacy/control

Table Manners
-

a ‘physical’ SAT


For this implementation of the SAT concept, we wanted to take some physical aspect of
collaborative technology and use it as a base toolset with which to experiment on various
topics


Group formation in distributed settings


The detection and facilitation of group dynamics in local and distributed settings


The locus of ‘command’ in a collaborative technology


Remote control and tele
-
haptic technologies


Computer Supported Collaborative Play


For this, we chose to implement two HI
-
Space tables over two sites in Ottawa (CRC’s
Virtual Classroom and NRC’s MoleRat lab), with an option to network further tables


The umbrella name of this technology toolset is ‘
Table Manners’
(thanks to Monica…)


We’re still building the tables
-

they will be online by June

Architecture


Table Manners will use agents to represent individual
users


Each user will then have a model the agent can use to
predict behaviour, analyse the same, and come up
with worthwhile group building/reinforcing structures
amongst the other users and their agents


This raises interesting questions of privacy, sensing,
avatar potential, etc.


Table Manners continued


We will have


2 networked tables over a dedicated research fibre for
high bandwidth


2 large plasma displays for video conferencing


Palm pilot control and personalisation of table and
associated information


Wireless Haptic capabilities (prototypes based on
MindStorms stuff, with more to come)


Several potential projects to implement and observe


(and I’m very excited about the potentials!)


Potential Table Research Approaches


Remote control of robotics


Agent based user modeling
-

truly socially adept
agents…


Avatars


Advanced video conferencing


Active environments


Collaborative gaming


Privacy and trust


Ubiquitous individual information handling


The Wacky Idea File


Wouldn’t it be nice if…


Your PDA could guide you in real time about the customs and
social expectations of the new country you’ve just arrived in…


Your PDA could link with an active environment and show you
how what you see now relates to what you saw (perhaps in
another country) last week


Wireless was so ubiquitous you really were always wired, and
your machines were always contextually and socially aware


Your ‘machines’ really were invisible, really were ‘personal’
and really did ‘know’ what do do at any given time


We really could trust our technology to do the ‘right’ thing


Technology simply faded away when you didn’t need it…


Questions


Work on Socially Adept Technologies raises its own
questions


Ironically, of ethics and morality amongst others


Some of them I mention briefly here, others may be
plain to you


Is this a good thing to do?


Are we hurting people by deceiving them?


Is anthropomorphising technology good for people?


Can ‘bad people’ use this in naughty ways? How?


What does this give us regular AI (or even dumb) technologies
don’t?


Do we need Artificial Intelligence if we have Social
Intelligence?


Where is all this going?

Problems


The main concern with all of this work is that it may
indeed not be possible


Is it, for example, possible to provide a ‘good enough’
representation of Trust for artificial agents? What about
emotion?


And given this, does the agent
really
trust or emote,
for example?


All of which raises interesting philosophical
questions, if nothing else


And which I leave in your capable hands… but...

Some Answers?


The obvious answer to the questions and problems raised is to
ignore them and hope they go away


(This is not a particularly scientific, or moral, approach…)


The most powerful answer is that, by researching these topics
and by asking these questions
at the same time
, we are doing two
very powerful things:


Attempting to use what we find to achieve some answers


Asking the questions (another tautology?!)


As scientists, we have a responsibility for asking these questions
before others do the work without asking them, and who knows,
we might just find some answers...

Conclusions


‘Socially Adept Technologies’ relates to two distinct things:


Technologies capable of behaving properly in the social world


A collection of research topics relating to the concept of social behaviour


There are several other parts of the puzzle out there


Socially Adept Technologies can provide for novel, worthwhile,
and above all, satisfying models of user
-
machine interaction


In addition, understanding and modeling social behaviour may
lead to a better understanding of and support systems for
human societal behaviour (cf. Dibben, 1998)


Finally, looking at this topic raises interesting moral and ethical
questions which will not go away, and which it is our
responsibility to address

A final word

Without systems being able to
sense

that the user is
there, who they are, what they’re doing or need, and
lots of other physical and mental things,
this stuff is
doomed to ultimate failure


(consider the average traffic light)


The human in the loop is the key to Social Adeptness...

Thanks for listening…




Any questions?