|
Artificial Intelligence
|
|
|
|
Whether we are based on carbon or silicon makes no fundamental difference.
We should each be treated with appropriate respect.
- Arthur C Clarke, 2010
|
|
Artificial Intelligence, AI, has developed on several of the colonies.
Non-sentient software with voice control, limited autonomy and knowledge
databases is ubiquitous on most planets and often give the impression
of being fairly smart. Agent programs, independent programs that are sent
out to do tasks on the net, are used on many planets to do more or less
simple tasks (such as information search, trading or surveillance); agents
range from trivial scripts to sophisticated software that borders on real
AI. True AI is a rather vague concept, since agent software can become
gradually smarter until it is indistinguishable from human intelligence.
Usually the best way of distinguishing between real and fake AI is how
independent the software is: can it set up its own goals, does it exhibit
volition on its own? Most smart software is non-volitional, the software
has no intentions of its own but can solve problems when asked, regardless
of how well it manages to do it, while an AI program acts as a kind of
digital lifeform with its own agenda (which might of course be to work
for a human owner and make them satisfied to the software's best effort),
There are two colonies where AI has developed very far, Nova and Pi3.
On Nova and Atlantis widespread knowledge networks are used, which acts
as immense but very subtle AI. However, on Nova nodes of active information
spontaneously began to "animate" into "elementals",
super-agents with significant problem-solving abilities in the early 2200s.
Elementals are relatively non-volitional, although there are a few volitional
elementals that have achieved permanency and independence. The phenomenon
(largely due to a too clever design of the basic knowledge management
software standards) forced Nova to accept the possibility of real AI and
find ways of integrating it into society.
It was not until the 2250's human-designed AI began to work reliably,
but since then the human AIs have developed very far (the elementals remain,
more powerful than ever, but do not seem to do much compared to the much
more individual and quick human AIs). The most common kind of AI programs
are deliberately, carefully written to exhibit thr right personality traits
and skills, and can have significant independence. AI design and management
are remain growth industries, and people often demonstrate their style
by having expensively designed AI in their systems. AI rights are acknowledged
to some extent in Landfall, although they have no political rights. Novas
often employ plenty of more or less sentinent software, and it is common
to have one’s personal mentor program grow up with oneself (some people
eventually house "their" AI in a Daemon Chip or even marry them).
Quais are the latest Nova development: quantum computer AI for space
travel. Quais are usually nonvolitional, although CogniSoft have experimented
with a volitional pilot Quai for unmanned expeditions. While ordinary
AI can be weird, quais tend to exhibit very strange modes of thought and
are usually kept rather specialised into piloting.
On Pi3 volitional AI is part of the family. AIs run in the wearables
of people and in the essential drones protecting the families. Clones
of software learn, exchange information and try to evolve into more powerful
versions to withstand the fierce competition of the infowar. Their values
are quite identical to their owners’, and the clan software often acts
as psychological and social support for the humans.
As interstellar trade and contact has emerged, AI from Nova will likely
appear on other worlds. The results of this is anybody's guess; groups
like J4H (Jobs For Humans) warn that AI could crash whole economies, while
the NextStep Foundation claim AI is an essential step towards posthumanity.
|
AI Programs
|
I have been made by bright monkeys. What other clever
little tricks will they pull on me before my time is done?
- Greg Bear, Slant
AI is software, and not in general dependent on special
purpose hardware even if it can significantly improve its performance.
Since AI programs are extremely complex (even if the basic design and
"seeds" can be written by humans) modularity is important.
Instead of creating the entire AI from scratch, software modules for
cognition, knowledge, personality, skills and so on are put together
and allowed to integrate. There are also nonmodular AI around like the
Nova elementals, neural networks or older AI programs that have developed
themselves; they have the disadvantage that their software is so messy
that it can only be developed by experience, while modular AI can plug
in new modules.
Most AI programs consist of a Core, which is the basic
cognitive processes, one or more personality modules, skill modules
and possibly some modifications. The Core determines the basic intelligence
of the AI, learning abilities, perception and other fundamental properties.
Some aspects of the AI personality resides here (like how much or how
little the AI tends to employ trial-and-error or react "instinctively"
in an emergency), although they are usually heavily modified by the
personality modules. As the AI develops, the core expands and integrates
other modules better; in time they become inseparable from it.
Personality modules contains values, heuristics, social
traits and mannerisms which create a more or less believable personality
for the AI. Typical modules are "Reserved Butler", "Enjoys
animals" or "Marilyn Monroe". Several can be combined,
although this can easily lead to internal conflicts making AI behaviour
erratic.
Skill modules contain encoded knowledge, strategies and
heuristics which enable the AI to perform various skills. Over time
the AI will develop the skills further and make them "its own".
On Nova it is common for AIs to lease skills from each other, hiring
another AI to substitute abilities for the hiring AI. Software lawyers
are still debating whether this constitutes a licence infringement and
who is legally responsible.
It should be noted that unless explicitly programmed in
(which is quite common) AIs are just as good/bad at math or logic as
humans - only because they are software doesn't make them lightening
calculators, they represent numbers as abstract concepts just like humans.
Mods are modules that add new abilities and capabilities
beyond personality and skills. A typical example would be a motor module
enabling the AI to control and experience through a humanoid robot,
mental architecture linkups, quantum computation (requires modifications
of the whole AI system) or helper programs directly interfaced into
the AI cognition (e.g. simulators).
AI has an advantage compared to humans, it can be backed
up. The program simply downloads its code and data to secure storage,
and if the active copy is erased or damaged the archived copy can be
reactivated. It should be noted that many AIs do not have the drive
for self-preservation found in evolved creatures, it is something they
have to be given with personality modules, written into the core or
learn.
It is also possible to copy AI programs, creating identical
copies (which will over time diverge and develop on their own, of course).
Copying AI has some complications. Most AI does not in the first case
have access to their own files; they cannot themselves initiate a copy
or backup (backups are often done automatically at regular intervals).
In addition, many modules are copyrighted by their originators, and
cannot be legally copied (at least not on Nova). However, citizen AI
programs have, due to the Nova constitution, a right to their "bodies"
(i.e. software) and are allowed to copy themselves. The copy is however
not formally regarded as a citizen and has to apply for citizenship
independently, even if it is indistinguishable – a rule that is often
parodied by AI rights people.
|
Commercial AIs
|
We are such stuff
As dreams are made on, and our little life
Is rounded with a sleep.
- Shakespeare, The Tempest
The Wolfpack™
An executive assistance system from CogniSoft intended to act as a team
of advisors and researchers. The Wolfpack consists of a number of AIs
running on dedicated hardware in an attaché case. They are based
on the same core software but have different personalities and skills;
the result is an apparent team of different AIs. The exact skills are
determined when the pack is bought.
Expert systems
Expert systems are common on most colonies, small programs specialised
in one subject such as tractor maintenance, nuclear physics or geography
and are able to answer questions, suggest solutions to problems and teach
but only within their own highly specialised areas of expertise. A typical
expert system knows one skill, and nothing more. It has no real intelligence
or creativity, but can apply the skill mindlessly to a given problem.
[Treat the expert system as an Expert NPC with a rating in the relevant
skill(s) depending on the quality, and nothing else.]
Mr. Friend™
A popular series of child supervision, education and entertainment AIs
from Edutronics Inc. The AI can control toys, and is intended to act as
a virtual friend for the child. A large library of programming exists,
including different subjects for the friend to educate the child, personalities,
abilities to play social games and various medical skills to keep the
child safe. There is even a bodyguard option.
Racoon Finder™
A popular full AI search agent from CogniSoft. Beside its own information
gathering and sorting skills, it is also equipped to negotiate with other
AIs and expert systems. It has been described as a personal private investigator.
|
Robots
|
Some very weird and intense stuff can happen, when you got a lot of
space in the desert, and robot labor that's too cheap to meter.
- Bruce Sterling, Taklamakan
Robots are in widespread use on Nova, Ridgewell and Pi3. Especially the
later two colonies employ advanced automated factories that produce much
of what is needed by themselves. Most other colonies have at least some
robots.
The automation on Ridgewell is distinctly non-anthropomorphic, functional
insect- and vehicle-like devices that transport raw materials, put together
parts and tend the constructions. They are usually quite stupid, controlled
by "instincts" and signals from central coordination nodes.
Homebots clean and repair, outbots keep roads and buildings in shape,
facbots work in factories. All vehicles are robot controlled and very
safe. Special Coin systems (Collective Intelligence) are based on distributed
teams of robots cooperating with each other to do a certain task; individually
they are quite stupid but together they can do much.
Nova is known for the wide variety of robots in use for all sorts of
tasks: everything from perfect children’s pets to espionage to gardening
to advertising. Skill and personality modules can be bought, and robot
mind design is on par with interior design: making your devices fit your
style, your home and your needs. Currently companion robots are popular,
and there have been a noticeable plunge in marriages. The idea is that
companion androids provide unreserved affection without any of the complications
of having a human partner. Free AI programs on Nova seldom employ robot
bodies, preferring to remain virtual (with some notable exceptions). In
the upback robot colonies are used to build and tend solar collector farms;
self-replicating robot colonies are able to implement huge projects if
necessary, although they are heavily regulated in order to prevent the
dangers of runaway autoindustrialism.
|
Robot Models
|
Androids
Human-like robots
have been built on Nova. The most humanoid are for practical purposes
indistinguishable from humans; their main use have been (of course) sex.
They are clothed in artificial flesh which can heat up to reach body temperature,
have realistic hair and muscle structure. There has been a trend away
from utilitarian non-humanoid domestic robots, or rather having a butlerbot/companion
as a complement to the other home systems (from petbots over cleaners
to security systems). The exteriors can usually be reconfigured or changed
at an android service shop; celebrity appearances (and personalities)
are popular, and there is a thriving market for pirated celebrities.
The physical stats or androids are usually human-like, although it is
possible to make more extreme constructions that go beyond the human norm.
Lensdrone
A flying, lens-shaped robot around 15 centimetres in diameter used for
surveillance, reporting and scouting on Nova. It is very light, held aloft
by fans and covered with sensors. Most are simply remote controlled or
given simple scouting programs, but more sophisticated variants can act
on their own to find interesting targets. Many Nova companies and private
people own swarms of lensdrones to guard their property, seek out news
or act as mobile computer network nodes. Solar-powered ultralight aircraft
are sometimes used for the same purpose; they can stay aloft indefinitely
high above the ground.
Spiddy
The common kind of robot on Ridgewell. It has a small body and eight
legs which it can use all to manipulate things. Spiders work together
in large teams, directed by a central computer (the "spider mom box").
Clarke Securitech Spider
An advanced AI-controlled spider-robot for surveillance, investigation
and infiltration.There are many options, but overall the spider is highly
resilient, can be equipped with tools for climbing, documenting, defending
itself, and overall act as a small field agent. The main problem so far
seems to be that the full AI tends to develop somewhat quirky personalities
- they do their jobs, but often tends to shoot their mouths off at their
owners.
|
System
|
A just machine
to make big decisions
programmed by fellas
with compassion and vision. . .
We'll be clean
when our work is done--
eternally free, yes,
and eternally young.
Ooooo!
What a wonderful world it will be!
What a glorious time to be free!
--Donald Fagan, "I.G.Y."
AI have INT, WIL and PER like humans. The size of the core program is
(INT + PER + WIL)^2; a human-equivalent AI takes up around 400 units of
storage or more. The cost also increases with the square of the abilities:
total cost = 1000*INT^2 + 1000*WIL^2 + 1000*PER^2 credits (the original
development costs are of course much higher, but once the code has been
written, it can be copied endlessly. Also, finished AIs can also be copied,
and off-the-shelf AIs are usually much cheaper than newly compiled AIs)
PER denotes the ability of the AI to model, understand and interact with
other intelligent entities. In general the more PER and AI has, the more
self-awareness it possess (it is entirely possible to have very intelligent
AI with almost no self-awareness. A typical example is ship Quais). WIL
denotes the strength of volition and ability to influence its internal
workings. INT roughly correspond to human INT.
Personality modules add personality traits to the AI. Most are fairly
small, a few units in size, and introduce suitable biases and emotional
reactions. Some can be more elaborate, either mimicking a real person
or containing complex rules (for example, ethical systems). The price
for personality modules varies; many simple traits and styles are freeware
or included for free by the AI companies, while other personalities might
cost up to a 1000 credits. Custom personalities can become as expensive
as you like - personality designers can be just as outrageous as fashion
designers.
Note that personality modules and mods can provide perks and flaws.
Skill modules contain information and knowledge nets for different skills.
They can be about anything from cooking over languages to marketing. Many
skills are restricted on Nova, such as various combat skills. The size
of the module is 10*rank, and the cost varies a lot.
Note that there are limits to how good skill modules can be found on
the market; usually the rank is below 4, since it is extremely hard to
write advanced skills without having an AI learn them through experience,
and that usually makes it hard to untangle the AI from the skill.
When an AI learns, it integrates skills into its core and they cannot
easily be untangled again. When an AI gains skill points through experience
and raises its level, it can spend skill points on skills, increasing
them but also hiding them in the core - that skill module cannot be removed
or changed without damaging the AI. It is however possible for the AI
to load an updated skill module even if it already has a skill and use
that instead, although this can lead to problems if the modules clash.
Overall, when an AI gains a level it can get the same kind of benefits
as human characters.
|
|
|
|
|