Open up a news article, flip to the science section, look under AGI. Artificial general...
Playtest
- Thread starter Acatalepsy
- Start date
- Created
- Status
- Ongoing
- Watchers
- 9
- Recent readers
- 0
- Threadmarks
- 1
-
Statistics (1 threadmarks, 1.4k words)
User Total Acatalepsy 1 Threadmarks Reader mode
Acatalepsy
Firewall Proxy
- Location
- The Diamond Age
Open up a news article, flip to the science section, look under AGI. Artificial general intelligence; minds that think as well as us, if not like us. Make a word web, scale size of word by frequency, shade by emotional affect. Note the anomaly - the word 'Cold', large and central, in shades between navy blue and dark purple, connected in valence with words like 'emotionless', 'unfeeling', remorseless'. Cold is a temperature, not an attribute of a mind. It's incorrect. Jonas rubbed his hand over core casing number seven, feeling the like a sun-warmed table despite the dedicated pumping of liquid hydrogen underneath its titanium skin.
Come on, girl, he thinks. I know you can do it.
The attack bounces from constellation to constellation, shifting perspective between frames, assembling fragments of data into gestalts detailing vulnerabilities, patterns, and probabilities. Here a packet runs on an oblivious weather satellite, repurposing its communications array for more sinister purposes. It slams hard into a military firewall, and feels inquisitive tendrils probing in preparation for a counter attack. No thanks, it says, sorry to bother you, and moves on to easier prey.
The battlefield stretches space and dimension. The literal field is only a tiny speck compared to the distances contemplated, but its density is a challenge, a maze of buildings and structures in a hodgepodge style. Heuristics quiver, confused; one insists that it's somewhere in the middle east, another recognizes Singaporean street signs, and still others point to distinctly European designs to the roads.Eventually the machine ceases paying attention to such signals, and concentrates on the literal boundaries. This is the sandbox it has, its information is fully defined. Those heuristics, even if accurate, give it no new information.
Sandbox is appropriate. This is a testing ground, built to confuse and frustrate. It is a mix of pattern and anti-pattern, alternating between cleanly designed spaces that might as well come out of the tactical manual to chaotic heaps of broken brick, dumped across the floor for no other purpose than to bounce t-rays at odd angles and make processors handle rough terrain and firefight geometry in the same seconds, or risk real (or simulated) annihilation.
The larger field is more than this tiny complex. It is a black box that contains a universe, a low-fidelity glimpse into the vast spinning shell of microsats beaming terabytes of entropy across the sky. The system that simulates such a thing wields such awesome power, it is almost a shame that it spends so much of it in dumb simulation, ticking its little world ahead microsecond by microsecond, an autistic system unknowing of the larger reality.
Back in the real, soldier ants march across the conflicted space. Despite having half the number of requisite legs, they give the impression of being mechanical spiders nearly as long as a man is tall, sporting a turret-arm hunched back like a scorpion's tail. They fan out over the complex, sidestepping piles of bricks, climbing over and around wrecked cars. Around them flit dragonfly-drones, covering broad swathes of the battlespace in a particle swarm, hoping to find the enemy gradient.
The enemy sits twenty meters away, in a conference room cleared out of tables and chairs. Twenty men and women in military fatigues sat in a circle around a cylindrical terminal praying to spiritual totem. Fiber optic wires spread out from the device, snaked over and around each figure, and up to small plugs in the back of their necks. Further wires radiated out from there; from each cortical plug a half dozen wires wrapped around or through the circle until it found another soldier's head.
Around them, a bored platoon of medics and doctors sat in hard plasticine hairs, backdropped by posters of soldiers with quotations of long dead generals about courage and proclamations of virtue. The soldiers couldn't see them; were hardly even aware of their surroundings. Their eyes unfocused and drifted, barely reacting at all when one of the medics got up and wiped their bald heads with a cool rag. Like their opponent, the greatest threat these soldiers faced was the heat of their own collective intellects.
Across the hall, side by side, two teams watched the data of the match scroll across thin terminals and projectors. Both were a mix of theorists and technicians, specialists in cybernetic theory, neurology, and computer science. Behind them, the audience; more soldiers, in dress uniforms with ribbons and medals festooned across chests.
Jonas hung back with generals, and listened to them try to make sense of the feeds. "Which one is our boys?" he heard a three star whisper to an aide. "I can't tell which one of these is that damned machine."
One of the flies sees its counterpart. Its peers note the change, and bias their search; bias becomes confirmation as sounds and sights trickle back through bayesian filters and undirected particle swarm becomes coordinated search pattern.
This might be a distraction from the real fight in fake-space, but distraction implies that attention is a conserved resources. The machine is overflowing with attention, left to wonder tangential cases between bits of useful data. It is a chorus of heuristic sub processes, sandwiched between executive functions and incomplete models.
Its agents are competent in their own sphere. They can walk and shoot without difficulty, find paths across rough terrain, identify threats, take cover, and signal allies.They understand agency, and divide the world into the trichotomy of RED/GREEN/BLUE - hostile/neutral/friendly. As machines go, they are brilliant savants. They lack only concepts of tactics, or purpose, or even something as banal as 'mission objectives'. They are tools to be given direction by forces greater than themselves.
The machine shifts through a dozen fire plans and choose something diffuse, something that will bring its opponents out of their hiding, to mass and reveal themselves. As of yet, it has yet to imagine the enemy's will in detail; now it simply falls back on doctrine, heuristics, and sums over probabilities.
Probing gunfire rips across faux courtyards and ricochets off concrete barricades as subagents see RED, request and receive permission to fire. Flocks of airborne spies merge and split, microjamming and probing with infosec seeking to isolate and confuse, but encryption and entropy hold. It's been decades since battlefield codebreaking has been viable.
Return fire disables a subagent. There. The machine sees the shape of the enemy attack. It winnows the thousands of plans considered to more manageable numbers. Now it begins to reason on the enemy's intentions, and sends incremental updates.
Updates that vanish into a burst of static and noise. It localizes bursts of electromagnetic radiation with speed of light arithmetic. It's smaller agents flutter helplessly, crippled or confused, while its larger agents are left between updates and plans. The enemy has fired a salvo of electromagnetic pulse bombs, spread in payload-grenade shells across a broad arc above its perimeter.
The machine considers once again the enemy. Reasoning on the enemy's intentions is a new art, for a machine. That module is new, untested against this type of enemy. Humans use empathy to, metaphorically, 'put themselves in the shoes' of others. The machine doesn't have a mechanism to hypothesis in detail about a gestalt mind. Nonetheless, it does have the capability to make a hypothesis about opposition in general.
Even more difficult than understanding a hostile mind is understanding that a hostile mind may, itself, be modelling the machine itself. And yet is a natural structure, a necessary one, one the research team developed painstakingly and tested at low levels against human and machine opponents. This mechanism is trained to recognize attempts to exploit the machine's own reasoning. The same training, by backpropagation, can tell when one of the machine's plans might be exploitable, and avoid doing so again.
These plans had been stored, cached, and integrated into executive processes. When the time comes, they are recalled in microseconds, and plans devised around it.
A human would have thought: Hey. You did that on purpose!
Come on, girl, he thinks. I know you can do it.
The attack bounces from constellation to constellation, shifting perspective between frames, assembling fragments of data into gestalts detailing vulnerabilities, patterns, and probabilities. Here a packet runs on an oblivious weather satellite, repurposing its communications array for more sinister purposes. It slams hard into a military firewall, and feels inquisitive tendrils probing in preparation for a counter attack. No thanks, it says, sorry to bother you, and moves on to easier prey.
The battlefield stretches space and dimension. The literal field is only a tiny speck compared to the distances contemplated, but its density is a challenge, a maze of buildings and structures in a hodgepodge style. Heuristics quiver, confused; one insists that it's somewhere in the middle east, another recognizes Singaporean street signs, and still others point to distinctly European designs to the roads.Eventually the machine ceases paying attention to such signals, and concentrates on the literal boundaries. This is the sandbox it has, its information is fully defined. Those heuristics, even if accurate, give it no new information.
Sandbox is appropriate. This is a testing ground, built to confuse and frustrate. It is a mix of pattern and anti-pattern, alternating between cleanly designed spaces that might as well come out of the tactical manual to chaotic heaps of broken brick, dumped across the floor for no other purpose than to bounce t-rays at odd angles and make processors handle rough terrain and firefight geometry in the same seconds, or risk real (or simulated) annihilation.
The larger field is more than this tiny complex. It is a black box that contains a universe, a low-fidelity glimpse into the vast spinning shell of microsats beaming terabytes of entropy across the sky. The system that simulates such a thing wields such awesome power, it is almost a shame that it spends so much of it in dumb simulation, ticking its little world ahead microsecond by microsecond, an autistic system unknowing of the larger reality.
Back in the real, soldier ants march across the conflicted space. Despite having half the number of requisite legs, they give the impression of being mechanical spiders nearly as long as a man is tall, sporting a turret-arm hunched back like a scorpion's tail. They fan out over the complex, sidestepping piles of bricks, climbing over and around wrecked cars. Around them flit dragonfly-drones, covering broad swathes of the battlespace in a particle swarm, hoping to find the enemy gradient.
The enemy sits twenty meters away, in a conference room cleared out of tables and chairs. Twenty men and women in military fatigues sat in a circle around a cylindrical terminal praying to spiritual totem. Fiber optic wires spread out from the device, snaked over and around each figure, and up to small plugs in the back of their necks. Further wires radiated out from there; from each cortical plug a half dozen wires wrapped around or through the circle until it found another soldier's head.
Around them, a bored platoon of medics and doctors sat in hard plasticine hairs, backdropped by posters of soldiers with quotations of long dead generals about courage and proclamations of virtue. The soldiers couldn't see them; were hardly even aware of their surroundings. Their eyes unfocused and drifted, barely reacting at all when one of the medics got up and wiped their bald heads with a cool rag. Like their opponent, the greatest threat these soldiers faced was the heat of their own collective intellects.
Across the hall, side by side, two teams watched the data of the match scroll across thin terminals and projectors. Both were a mix of theorists and technicians, specialists in cybernetic theory, neurology, and computer science. Behind them, the audience; more soldiers, in dress uniforms with ribbons and medals festooned across chests.
Jonas hung back with generals, and listened to them try to make sense of the feeds. "Which one is our boys?" he heard a three star whisper to an aide. "I can't tell which one of these is that damned machine."
One of the flies sees its counterpart. Its peers note the change, and bias their search; bias becomes confirmation as sounds and sights trickle back through bayesian filters and undirected particle swarm becomes coordinated search pattern.
This might be a distraction from the real fight in fake-space, but distraction implies that attention is a conserved resources. The machine is overflowing with attention, left to wonder tangential cases between bits of useful data. It is a chorus of heuristic sub processes, sandwiched between executive functions and incomplete models.
Its agents are competent in their own sphere. They can walk and shoot without difficulty, find paths across rough terrain, identify threats, take cover, and signal allies.They understand agency, and divide the world into the trichotomy of RED/GREEN/BLUE - hostile/neutral/friendly. As machines go, they are brilliant savants. They lack only concepts of tactics, or purpose, or even something as banal as 'mission objectives'. They are tools to be given direction by forces greater than themselves.
The machine shifts through a dozen fire plans and choose something diffuse, something that will bring its opponents out of their hiding, to mass and reveal themselves. As of yet, it has yet to imagine the enemy's will in detail; now it simply falls back on doctrine, heuristics, and sums over probabilities.
Probing gunfire rips across faux courtyards and ricochets off concrete barricades as subagents see RED, request and receive permission to fire. Flocks of airborne spies merge and split, microjamming and probing with infosec seeking to isolate and confuse, but encryption and entropy hold. It's been decades since battlefield codebreaking has been viable.
Return fire disables a subagent. There. The machine sees the shape of the enemy attack. It winnows the thousands of plans considered to more manageable numbers. Now it begins to reason on the enemy's intentions, and sends incremental updates.
Updates that vanish into a burst of static and noise. It localizes bursts of electromagnetic radiation with speed of light arithmetic. It's smaller agents flutter helplessly, crippled or confused, while its larger agents are left between updates and plans. The enemy has fired a salvo of electromagnetic pulse bombs, spread in payload-grenade shells across a broad arc above its perimeter.
The machine considers once again the enemy. Reasoning on the enemy's intentions is a new art, for a machine. That module is new, untested against this type of enemy. Humans use empathy to, metaphorically, 'put themselves in the shoes' of others. The machine doesn't have a mechanism to hypothesis in detail about a gestalt mind. Nonetheless, it does have the capability to make a hypothesis about opposition in general.
Even more difficult than understanding a hostile mind is understanding that a hostile mind may, itself, be modelling the machine itself. And yet is a natural structure, a necessary one, one the research team developed painstakingly and tested at low levels against human and machine opponents. This mechanism is trained to recognize attempts to exploit the machine's own reasoning. The same training, by backpropagation, can tell when one of the machine's plans might be exploitable, and avoid doing so again.
These plans had been stored, cached, and integrated into executive processes. When the time comes, they are recalled in microseconds, and plans devised around it.
A human would have thought: Hey. You did that on purpose!