Research in human-like agent behavior?

Archived from groups: comp.ai,comp.ai.alife,comp.ai.games (More info?)

I'm looking for any research on achieving truly human-like behavior in
synthetic agents (in particular virtual agents such as computer game
characters). Specifically, I'm interested in human-like decision
making, not other issues like motor control which don't directly
pertain to AI. Is anyone aware of any research in this? The only
work I'm familiar with is integrating together a lot of techniques to
try to make the agent's controller come as close to human intelligence
as possible.

Thanks in advance for any info.

[ comp.ai is moderated. To submit, just post and be patient, or if ]
[ that fails mail your article to <comp-ai@moderators.isc.org>, and ]
[ ask your news administrator to fix the problems with your system. ]
7 answers Last reply
More about research human like agent behavior
  1. Archived from groups: comp.ai,comp.ai.alife,comp.ai.games (More info?)

    Jonathan Dinerstein <jondinerstein@yahoo.com> wrote:

    >I'm looking for any research on achieving truly human-like behavior in
    >synthetic agents (in particular virtual agents such as computer game
    >characters). Specifically, I'm interested in human-like decision
    >making, not other issues like motor control which don't directly
    >pertain to AI. Is anyone aware of any research in this? The only
    >work I'm familiar with is integrating together a lot of techniques to
    >try to make the agent's controller come as close to human intelligence
    >as possible.
    >
    >Thanks in advance for any info.
    >

    You may want to take a look at the Soar Project:

    http://sitemaker.umich.edu/soar

    -Sayan.

    [ comp.ai is moderated. To submit, just post and be patient, or if ]
    [ that fails mail your article to <comp-ai@moderators.isc.org>, and ]
    [ ask your news administrator to fix the problems with your system. ]
  2. Archived from groups: comp.ai,comp.ai.alife,comp.ai.games (More info?)

    bhattach@engin.umich.edu (Sayan Bhattacharyya) wrote in message news:<41073f63$1@news.unimelb.edu.au>...
    > Jonathan Dinerstein <jondinerstein@yahoo.com> wrote:
    >
    > >I'm looking for any research on achieving truly human-like behavior in
    > >synthetic agents (in particular virtual agents such as computer game
    > >characters). Specifically, I'm interested in human-like decision
    > >making, not other issues like motor control which don't directly
    > >pertain to AI. Is anyone aware of any research in this? The only
    > >work I'm familiar with is integrating together a lot of techniques to
    > >try to make the agent's controller come as close to human intelligence
    > >as possible.
    > >

    The intelligence required to select a muscle to move is similar
    to the intelligence required to select a strategy or direct thoughts,
    so I don't see why you think motor control isn't a legitimate part of
    AI.
    Since people don't really understand how the mind works,
    programming an AI is difficult and the past half century or so of
    research has been unsuccessful. For games though, there might be
    little tricks for making the character appear intelligent. I would
    think along those lines and not waste time on AI for game characters.

    [ comp.ai is moderated. To submit, just post and be patient, or if ]
    [ that fails mail your article to <comp-ai@moderators.isc.org>, and ]
    [ ask your news administrator to fix the problems with your system. ]
  3. Archived from groups: comp.ai,comp.ai.alife,comp.ai.games (More info?)

    Jonathan Dinerstein wrote:
    > I'm looking for any research on achieving truly human-like behavior in
    > synthetic agents (in particular virtual agents such as computer game
    > characters). Specifically, I'm interested in human-like decision
    > making, not other issues like motor control which don't directly
    > pertain to AI. Is anyone aware of any research in this? The only
    > work I'm familiar with is integrating together a lot of techniques to
    > try to make the agent's controller come as close to human intelligence
    > as possible.
    >
    > Thanks in advance for any info.

    There are references to a fair amount of this type of work here:

    http://www.soartech.com/publications.php

    [ comp.ai is moderated. To submit, just post and be patient, or if ]
    [ that fails mail your article to <comp-ai@moderators.isc.org>, and ]
    [ ask your news administrator to fix the problems with your system. ]
  4. Archived from groups: comp.ai,comp.ai.alife,comp.ai.games (More info?)

    wtkiii@hotmail.com wrote:
    > Since people don't really understand how the mind works,
    > programming an AI is difficult and the past half century or so of
    > research has been unsuccessful. For games though, there might be
    > little tricks for making the character appear intelligent. I would
    > think along those lines and not waste time on AI for game characters.

    I don't think I agree that it's been "unsuccessful." We've learned a
    lot.

    We have better search techniques, better algorithms for machine
    vision, better methods of spatial representation, and scads and
    scads of useful and workable heuristics for finding "good"
    solutions to intractable problems - even if they're not necessarily
    the "best" solutions.

    We've learned a lot about language and parsing, a lot about efficient
    methods of doing mathematical regressions, about the mathematical
    properties of intractable problems, etc.

    We've been able to make a lot of successful applications, too: from
    chess and poker playing games to useful (though limited) natural-
    language interfaces and systems for searching vast amounts of
    NL text for writing about particular subjects, to expert systems
    that have proven incredibly useful especially in industrial
    control applications. None of them is as smart as a person, nor
    smart in exactly the same *way* as a person, but they can do things
    that we need done, so they're useful.

    Various people have held forth goals and achieved them. Various
    others have contented themselves with pointing at the goals that
    no one has achieved yet. Every time something becomes an acceptable
    engineering technique, we quit calling it "AI". But that doesn't
    change the fact that a lot of successful engineering techniques
    come from AI research.

    And people are still making new discoveries....

    Bear

    [ comp.ai is moderated. To submit, just post and be patient, or if ]
    [ that fails mail your article to <comp-ai@moderators.isc.org>, and ]
    [ ask your news administrator to fix the problems with your system. ]
  5. Archived from groups: comp.ai,comp.ai.alife,comp.ai.games (More info?)

    Ray Dillinger wrote:
    > wtkiii@hotmail.com wrote:
    >
    >> Since people don't really understand how the mind works,
    >>programming an AI is difficult and the past half century or so of
    >>research has been unsuccessful. For games though, there might be
    >>little tricks for making the character appear intelligent. I would
    >>think along those lines and not waste time on AI for game characters.
    >
    >
    > I don't think I agree that it's been "unsuccessful." We've learned a
    > lot.
    >
    > We have better search techniques, better algorithms for machine
    > vision, better methods of spatial representation, and scads and
    > scads of useful and workable heuristics for finding "good"
    > solutions to intractable problems - even if they're not necessarily
    > the "best" solutions.
    >
    > We've learned a lot about language and parsing, a lot about efficient
    > methods of doing mathematical regressions, about the mathematical
    > properties of intractable problems, etc.
    >
    > We've been able to make a lot of successful applications, too: from
    > chess and poker playing games to useful (though limited) natural-
    > language interfaces and systems for searching vast amounts of
    > NL text for writing about particular subjects, to expert systems
    > that have proven incredibly useful especially in industrial
    > control applications. None of them is as smart as a person, nor
    > smart in exactly the same *way* as a person, but they can do things
    > that we need done, so they're useful.
    >
    > Various people have held forth goals and achieved them. Various
    > others have contented themselves with pointing at the goals that
    > no one has achieved yet. Every time something becomes an acceptable
    > engineering technique, we quit calling it "AI". But that doesn't
    > change the fact that a lot of successful engineering techniques
    > come from AI research.
    >
    > And people are still making new discoveries....
    >
    > Bear
    >

    If the quest to land 'a man on the moon' had been equally as (un)successful as
    the quest to build an artificial system that exhibited human levels of
    intelligence, all of our space flights would have:

    - sat inert on the launch pad
    - fizzled out or exploded somewhere in flight
    - crash landed on asteroids and wrong planets
    - stopped communicating
    - been lost in space

    Over and over and over...

    Worse yet, each outcome would have occurred almost randomly, with no apparent
    progress toward achieving the original mission except that we might know
    something about where NOT to point the rocket.

    IMO, the level of performance of today's AI systems is not measurably better
    than it was 40 years ago. Yes, we've discovered some interesting (or
    entertaining) byproducts. But if the Apollo missions had made a comparable lack
    of progress as has AI, nobody would be lauding the merits of its derivative
    technology. NASA would be rightly embarrassed by their zero percent success
    rate and their unquantifiable rate of progress. And like the good engineers
    NASA is/was, they'd quietly go back to their drawing boards.

    Unfortunately, I think AI still has most of its journey yet to go, and it's
    still not clear whether the rocket is going in the right direction.

    Randy

    --
    Randy Crawford http://www.ruf.rice.edu/~rand rand AT rice DOT edu

    [ comp.ai is moderated. To submit, just post and be patient, or if ]
    [ that fails mail your article to <comp-ai@moderators.isc.org>, and ]
    [ ask your news administrator to fix the problems with your system. ]
  6. Archived from groups: comp.ai,comp.ai.alife,comp.ai.games (More info?)

    <wtkiii@hotmail.com> wrote in message news:4109fc9e$1@news.unimelb.edu.au...
    > bhattach@engin.umich.edu (Sayan Bhattacharyya) wrote in message
    news:<41073f63$1@news.unimelb.edu.au>...
    > > Jonathan Dinerstein <jondinerstein@yahoo.com> wrote:
    > >
    > > >I'm looking for any research on achieving truly human-like behavior in
    > > >synthetic agents (in particular virtual agents such as computer game
    > > >characters). Specifically, I'm interested in human-like decision
    > > >making, not other issues like motor control which don't directly
    > > >pertain to AI. Is anyone aware of any research in this? The only
    > > >work I'm familiar with is integrating together a lot of techniques to
    > > >try to make the agent's controller come as close to human intelligence
    > > >as possible.
    > > >
    >
    > The intelligence required to select a muscle to move is similar
    > to the intelligence required to select a strategy or direct thoughts,
    > so I don't see why you think motor control isn't a legitimate part of
    > AI.
    > Since people don't really understand how the mind works,
    > programming an AI is difficult and the past half century or so of
    > research has been unsuccessful. For games though, there might be
    > little tricks for making the character appear intelligent. I would
    > think along those lines and not waste time on AI for game characters.
    >

    Hey all,

    I agree with wtkiii, that full-on AI research is not necessary for a casual
    or even professional game programmer. I have done some cursory web research
    in A-Life, and a great deal of success was achieved in the "Creatures"
    series, as far as believable behavior was concerned, with >Neural Nets<.

    It's a very advanced topic, but for gaming purposes, you would have a
    handfull of nodes.

    1. Gather/Predation Node
    2. Flee Node
    3. Sleep Node
    etc.

    Each of these nodes responds to stimuli, and through a system of fractional
    weights, >fuzzy logic<, the net determines which of the nodes has the most
    influence. For example, your bot's hunger numbers may be very high, but
    there is also an enemy, that it is far away. Food node draws lines to
    nourishment, while Flee node always points away from enemies, the reult is
    an aggregate action, which brings the bot to food, but in a direction that
    takes it away from the enemy. It's not straight-up logic, that would make
    one call to "find food" after deciding it was best, then ignore enemies.

    That's my input. I think searching "Neural Nets" and "Fuzzy Logic" would
    take you far in the quest for quasi-intelligence-done-quick.

    [ comp.ai is moderated. To submit, just post and be patient, or if ]
    [ that fails mail your article to <comp-ai@moderators.isc.org>, and ]
    [ ask your news administrator to fix the problems with your system. ]
  7. Archived from groups: comp.ai,comp.ai.alife,comp.ai.games (More info?)

    >Jonathan Dinerstein <jondinerstein@yahoo.com> wrote:
    >
    >
    >>I'm looking for any research on achieving truly human-like behavior in
    >>synthetic agents (in particular virtual agents such as computer game
    >>characters). Specifically, I'm interested in human-like decision
    >>making, not other issues like motor control which don't directly
    >>pertain to AI. Is anyone aware of any research in this? The only
    >>work I'm familiar with is integrating together a lot of techniques to
    >>try to make the agent's controller come as close to human intelligence
    >>as possible.


    Two projects come to mind, they're not entirely independent either:

    1) Commonsense Computing Group at MIT Media Lab
    http://csr.media.mit.edu/RoboverseHome.htm
    - Developing a virtual reality for testing intelligent agents in a
    limited environment.
    - Working on integrating Marvin Minsky's proposed mind architectures
    (Emotion Machine, Society of Mind)

    2) CoSy - UK & EU university collaboration on implementing a human
    architecture, many schools working on different parts of the problem.
    http://www.cs.bham.ac.uk/research/projects/cosy/
    - using Aaron Sloman's architecture

    Note, Sloman also submitted this project as a UK Grand Challenge in
    Computing: See http://www.cs.bham.ac.uk/research/cogaff/gc/

    Check out the proceedings from the St. Thomas common sense Symposium
    this year, available in most recent ed. of AI Magazine and here:
    www.cs.bham.ac.uk/research/ cogaff/AIMag/StThomas-AIMag.pdf

    Best
    Dustin

    [ comp.ai is moderated. To submit, just post and be patient, or if ]
    [ that fails mail your article to <comp-ai@moderators.isc.org>, and ]
    [ ask your news administrator to fix the problems with your system. ]
Ask a new question

Read More

Video Games