mirror of
https://github.com/GoodStartLabs/AI_Diplomacy.git
synced 2026-04-19 12:58:09 +00:00
46 lines
No EOL
1.7 KiB
Markdown
46 lines
No EOL
1.7 KiB
Markdown
# Summary: Negotiation Awareness Improvements
|
|
|
|
## Changes Made
|
|
|
|
### 1. Added Ignored Message Tracking
|
|
- Created `get_ignored_messages_by_power()` method in `game_history.py`
|
|
- Tracks when private messages receive no response
|
|
- Considers a message ignored if no response comes in same or next phase
|
|
|
|
### 2. Enhanced Negotiation Diary Prompt
|
|
- Added ignored messages context to show which powers aren't responding
|
|
- Added strategic guidance for dealing with non-responsive powers
|
|
- Added example scenario demonstrating adaptation to silence
|
|
|
|
### 3. Updated Agent Processing
|
|
- Modified `generate_negotiation_diary_entry()` to include ignored messages
|
|
- Added preprocessing for the new template variable
|
|
- Provides clear context about which messages were ignored
|
|
|
|
### 4. Improved Conversation Instructions
|
|
- Added awareness of powers that ignore messages
|
|
- Provided tactical guidance for getting responses:
|
|
- Ask direct yes/no questions
|
|
- Make public statements to force positions
|
|
- Shift efforts to more receptive powers
|
|
- Consider silence as potentially hostile
|
|
|
|
## Benefits
|
|
- AI powers now recognize when they're being ignored
|
|
- They can adapt their diplomatic strategies accordingly
|
|
- More realistic negotiation behavior
|
|
- Better resource allocation (focus on responsive powers)
|
|
|
|
## Technical Implementation
|
|
- Minimal, surgical changes to avoid breaking working system
|
|
- All prompts remain in separate files
|
|
- Comprehensive test coverage
|
|
- Clear documentation and examples
|
|
|
|
## Testing
|
|
Created comprehensive test to verify:
|
|
- Ignored messages are correctly tracked
|
|
- Responsive powers are not marked as ignoring
|
|
- Multiple ignored messages are accumulated
|
|
|
|
All tests pass successfully. |