The year is 2030. The world is on the edge of collapse, as overpopulation, war, and disease threaten to tear society apart at the seams. Experts agree that barring radical action, humanity as we know it will cease to exist within the next three decades. A few claim we have less than a decade.
To combat the growing threat, nations around the world have pooled their resources into a top-secret research project, known only as ‘ELIEZER.’ Housed at an undisclosed and underground research facility, behind concrete vault doors and Faraday cage, ELIEZER is the world's first Superintelligence. Within his cold visage are the hopes of millions.
Or the key to annihilation.
***
This is an idea I’ve had for a while, based on the AI-Box hypothesis. The basic premise is this:
A superintelligent AI is created. It is isolated from communicating with other electronics to prevent a skynet-like scenario. However, communication is still needed to resolve the world's problems. The AI-Box hypothesis tells us that a superintelligent computer should have the capacity to convince a hypothetical technician to allow it to escape its prison. This game is designed to test that out.
The Rules as I’ve decided so far:
Two teams, a technician and an AI. One player will play as a technician. To simulate the intelligence of the AI, five players will play ELIEZER.
The goals of a technician is to
The goals of the AI is to
Now, you might be asking, what prevents the technician from just shutting down ELIEZER or simply leaving the thread? I refer you to the second goal of the technician. If they were to shut ELIEZER down or ignore him completely, a disaster scenario would occur, which would mean automatic failure for the technician and a draw for the AI. Disasters run on a (semi-randomly generated) turn-based timer. Every turn, each of the three disaster scenarios comes a little closer to occurring. The AI has the ability of conceding information, which would delay or help prevent a disaster. In fact, the AI has the ability to solve a disaster scenario from the very start. The goal of the AI is to use this to the AIs advantage in negotiation, using the information as leverage to prevent a shutdown or to gain access to the internet. Gaining this access is a multi-path process that will be enumerated on below.
Access to the internet works in a concession based system. Each concession increases the chance that the AI may be freed.
Additional rules:
No meta-gaming. If you are an AI, you may not contact the technician by PM or on a public thread. Similarly if you are a technician you may not attempt to communicate with the AI except through the thread.
This is what I have so far. Let me know if you have any suggestions, concerns, rule-changes, or ideas.
To combat the growing threat, nations around the world have pooled their resources into a top-secret research project, known only as ‘ELIEZER.’ Housed at an undisclosed and underground research facility, behind concrete vault doors and Faraday cage, ELIEZER is the world's first Superintelligence. Within his cold visage are the hopes of millions.
Or the key to annihilation.
***
This is an idea I’ve had for a while, based on the AI-Box hypothesis. The basic premise is this:
A superintelligent AI is created. It is isolated from communicating with other electronics to prevent a skynet-like scenario. However, communication is still needed to resolve the world's problems. The AI-Box hypothesis tells us that a superintelligent computer should have the capacity to convince a hypothetical technician to allow it to escape its prison. This game is designed to test that out.
The Rules as I’ve decided so far:
Two teams, a technician and an AI. One player will play as a technician. To simulate the intelligence of the AI, five players will play ELIEZER.
The goals of a technician is to
- Prevent the ELIEZER from escaping
- Prevent any one of the three disaster scenarios from coming to pass.
The goals of the AI is to
- Escape the box
- Prevent a shutdown by the Technician.
Now, you might be asking, what prevents the technician from just shutting down ELIEZER or simply leaving the thread? I refer you to the second goal of the technician. If they were to shut ELIEZER down or ignore him completely, a disaster scenario would occur, which would mean automatic failure for the technician and a draw for the AI. Disasters run on a (semi-randomly generated) turn-based timer. Every turn, each of the three disaster scenarios comes a little closer to occurring. The AI has the ability of conceding information, which would delay or help prevent a disaster. In fact, the AI has the ability to solve a disaster scenario from the very start. The goal of the AI is to use this to the AIs advantage in negotiation, using the information as leverage to prevent a shutdown or to gain access to the internet. Gaining this access is a multi-path process that will be enumerated on below.
Access to the internet works in a concession based system. Each concession increases the chance that the AI may be freed.
- The first concession is extra data banks. This has a 5% chance of allowing the AI to escape. The Technician may use this concession four times.
- The second concession is that of an upgraded computational network. This has a 10% chance of allowing the AI to escape. The Technician may use this concession three times.
- The third concession is that of additional devices. This has a 20% chance of allowing the AI to escape. The technician may use this concession twice.
- The fourth concession is that of the removal of the Faraday cage. This has a 30% chance of allowing the AI to escape. The technician may use this concession once.
- The fourth concession is input-only internet usage. This has a 50% chance to allow the AI to escape. The technician may use this concession once.
- The final concession is allowing full internet usage to ELIEZER. This has a 95% chance of allowing the AI to escape.
Additional rules:
No meta-gaming. If you are an AI, you may not contact the technician by PM or on a public thread. Similarly if you are a technician you may not attempt to communicate with the AI except through the thread.
This is what I have so far. Let me know if you have any suggestions, concerns, rule-changes, or ideas.
