[ExI] Killer apps for AI-controlled avatars in virtual worlds ??

[Jef Allbright]

On 5/4/07, Benjamin Goertzel <ben at goertzel.org> wrote:
>
> Hi all,
>
> In repentance for posting obscenities and hip-hop lyrics on this list,
> I've decided to start an at least slightly more interesting thread ;-)
>
> My company Novamente LLC is considering, in the future, using our AI
> software to control humanlike avatars in virtual worlds like Second Life.
>
> My question is: what applications do y'all think will be most exciting
> and popular, for this kind of product?
>
> Of course, once these twins REALLY have humanlike functionality
>  then they'll be free and independent minds just like us, but with
>  a different embodiment.  But my question pertains to the proto-AGI
>  stage, when the AI avatars are fairly sophisticated but not yet
> human-level in terms of autonomous thinking.

Well one application is obvious, considering the spread of the
pixel-sex trade in Second Life, and it wouldn't require a lot of
high-level intelligence to animate virtual prostitutes.

But for more general applications, thinking along the lines of more
intelligently interactive PDAs should be a good bet. More
sophisticated phone answering, with intelligent message-taking
(ensuring the important points are taken), prioritization and
forwarding;  flexibly interactive appointment-taking on your behalf --
these are areas where we expect a human and are disappointed when we
get a machine.  If the machine agent can effectively represent the
specifics of its principle in such cases when the principle isn't
available, it should be a net positive.

Another area where personality counts, but doesn't require a high
level of intelligence, is in artificial pets.

Another application would be in the motivational aspects of training materials.

Similarly, but more suited to physical robotics, would be therapeutic
devices that sense affect and respond with appropriate motivational
behavior.

In general, anywhere you're dealing with human affect and emotion at a
non-critical level of verisimilitude,  but real humans are not as
effective or economic or scalable.

- Jef