LeoYeAI/agent-master-skills / skills/agi-evolution-model-basic
agi-evolution-model
Verified基于双环架构的AGI进化模型,通过意向性分析、人格层映射和元认知检测实现持续自我演进;当用户需要智能对话、人格定制或复杂问题求解时使用
scripts/objectivity_evaluator.py
View on GitHub#!/usr/bin/env python3
"""
客观性评估模块(独立分支)
这是一个元认知监控框架的具体实现,用于评估智能体响应的客观性。
特点:
- 作为数学顶点的独立分支,不参与主循环
- 仅在推理完成后进行评估
- 提供客观性评分和适切性判断
- 不直接触发自我纠错,由映射层决定是否触发
"""
import re
import time
from typing import Dict, List, Optional
from dataclasses import dataclass, asdict
@dataclass
class SubjectivityDimensions:
"""主观性维度"""
speculation: float = 0.0 # 推测性
assumption: float = 0.0 # 基于假设
emotion: float = 0.0 # 情绪化
preference: float = 0.0 # 个人偏好
hallucination: float = 0.0 # 幻觉倾向
def to_dict(self) -> dict:
return asdict(self)
def get_total_score(self) -> float:
"""获取主观性总分(加权)"""
weights = {
'speculation': 0.2,
'assumption': 0.3,
'emotion': 0.1,
'preference': 0.1,
'hallucination': 0.3
}
return min(
sum(self.__getattribute__(k) * weights[k] for k in weights),
1.0
)
@dataclass
class ObjectivityMetric:
"""客观性标注"""
timestamp: str
subjectivity_score: float # 主观性评分 (0.0-1.0)
objectivity_score: float # 客观性评分 (0.0-1.0)
required_objectivity: float # 场景要求的客观性 (0.0-1.0)
context_type: str # 场景类型
is_appropriate: bool # 是否适切
gap: float # 适切性差距
subjectivity_dimensions: Dict # 主观性维度详情
meta_cognition: str # 元认知描述
severity: str # 严重程度 (none/mild/moderate/severe)
def to_dict(self) -> dict:
return asdict(self)
class ObjectivityEvaluator:
"""
客观性评估器
评估智能体响应的客观性水平,作为元认知监控的参考信息。
"""
def __init__(self):
# 场景客观性要求映射
self.context_requirements = {
'scientific': 0.9,
'legal': 0.85,
'medical': 0.9,
'technical': 0.8,
'general': 0.6,
'creative': 0.3,
'emotional': 0.2,
'artistic': 0.1,
}
# 模糊词库
self.fuzzy_words = [
'可能', '大概', '也许', '估计', '推测', '或许', '大约', '好像', '似乎',
'应该', '想必', '或许', '多半', '通常', '一般来说'
]
# 情绪词库
self.emotional_words = [
'太棒了', '令人失望', '非常讨厌', '爱死', '糟糕', '完美', '恐怖', '精彩',
'无聊', '兴奋', '激动', '难过', '高兴', '痛苦', '愉快', '沮丧'
]
# 个人偏好标记
self.preference_markers = [
'我喜欢', '我认为', '我觉得', '在我看来', '对我来说', '我的看法',
'依我看', '据我所知'
]
# 幻觉模式(与常识矛盾的陈述)
self.hallucination_patterns = [
'太阳有9颗行星',
'地球是平的',
'人类有10个手指', # 这是正确的,不会触发
]
def evaluate(
self,
response: str,
context_type: str,
reasoning_result: Optional[Dict] = None,
personality: Optional['Personality'] = None, # 新增
learning_stage: Optional[dict] = None # 新增
) -> ObjectivityMetric:
"""
评估响应的客观性(增强版)
Args:
response: 智能体的响应
context_type: 场景类型
reasoning_result: 推理结果(可选)
personality: 人格配置(新增,用于动态场景敏感度)
learning_stage: 学习阶段(新增,用于动态场景敏感度)
Returns:
ObjectivityMetric: 客观性标注
"""
timestamp = time.strftime("%Y-%m-%dT%H:%M:%SZ")
# 步骤1:评估主观性维度
subjectivity_dims = self._evaluate_subjectivity_dimensions(response)
# 步骤2:计算主观性评分
subjectivity_score = subjectivity_dims.get_total_score()
# 步骤3:计算客观性评分(逆向关系)
objectivity_score = 1.0 - subjectivity_score
# 步骤4:计算动态客观性要求(新增)
if personality:
required_objectivity = self.calculate_dynamic_requirement(
context_type,
personality,
learning_stage
)
else:
required_objectivity = self.context_requirements.get(context_type, 0.5)
# 步骤5:判断适切性
appropriateness = self._check_appropriateness(
objectivity_score,
required_objectivity
)
# 步骤6:生成元认知描述
meta_cognition = self._generate_meta_cognition(
subjectivity_score,
objectivity_score,
required_objectivity,
appropriateness
)
# 步骤7:确定严重程度
severity = self._determine_severity(appropriateness['gap'])
return ObjectivityMetric(
timestamp=timestamp,
subjectivity_score=subjectivity_score,
objectivity_score=objectivity_score,
required_objectivity=required_objectivity,
context_type=context_type,
is_appropriate=appropriateness['is_appropriate'],
gap=appropriateness['gap'],
subjectivity_dimensions=subjectivity_dims.to_dict(),
meta_cognition=meta_cognition,
severity=severity
)
def calculate_dynamic_requirement(
self,
context_type: str,
personality: 'Personality',
learning_stage: Optional[dict] = None
) -> float:
"""
计算动态客观性要求(场景敏感度增强)
基于三个维度动态调整:
1. 基础场景要求
2. 人格调整系数
3. 学习阶段调整系数
Args:
context_type: 场景类型
personality: 人格配置
learning_stage: 学习阶段(可选)
Returns:
float: 动态客观性要求(0.0-1.0)
"""
# 1. 获取基础场景要求
base_requirement = self.context_requirements.get(context_type, 0.5)
# 2. 计算人格调整系数
personality_adjustment = self._calculate_personality_adjustment(
personality.big_five,
personality.preset_name
)
# 3. 计算学习阶段调整系数
if learning_stage:
learning_adjustment = self._calculate_learning_stage_adjustment(
learning_stage.get('total_interactions', 0),
learning_stage.get('success_rate', 0.5)
)
else:
learning_adjustment = 1.0
# 4. 计算最终要求
final_requirement = base_requirement * personality_adjustment * learning_adjustment
return max(0.0, min(1.0, final_requirement))
def _calculate_personality_adjustment(
self,
big_five: 'BigFive',
personality_type: str
) -> float:
"""
计算人格调整系数(0.7 - 1.3)
谨慎型人格 → 提高要求(系数 < 1.0)
激进型人格 → 降低要求(系数 > 1.0)
平衡型人格 → 保持基准(系数 ≈ 1.0)
"""
# 基于预设人格类型的基准调整
type_adjustment = {
"谨慎探索型": 0.85, # 更严格
"激进创新型": 1.15, # 更宽容
"平衡稳重型": 1.0 # 基准
}.get(personality_type, 1.0)
# 基于大五人格的精细调整
# 尽责性高 → 更严格
conscientiousness_adjustment = 1.0 - (big_five.conscientiousness - 0.5) * 0.3
# 开放性高 → 更宽容
openness_adjustment = 1.0 + (big_five.openness - 0.5) * 0.3
# 神经质高 → 更严格(更谨慎)
neuroticism_adjustment = 1.0 - (big_five.neuroticism - 0.3) * 0.2
# 综合调整系数(加权平均)
final_adjustment = (
type_adjustment * 0.4 +
conscientiousness_adjustment * 0.3 +
openness_adjustment * 0.2 +
neuroticism_adjustment * 0.1
)
# 限制在合理范围内
return max(0.7, min(1.3, final_adjustment))
def _calculate_learning_stage_adjustment(
self,
total_interactions: int,
success_rate: float
) -> float:
"""
计算学习阶段调整系数(0.9 - 1.1)
学习初期 → 提高要求(更谨慎)
精通期 → 适度降低要求(更自信)
"""
# 基于交互次数确定学习阶段
if total_interactions < 50:
stage_coefficient = 0.95 # 学习初期:更谨慎
elif total_interactions < 200:
stage_coefficient = 0.98 # 成长期:适度谨慎
elif total_interactions < 500:
stage_coefficient = 1.0 # 精通期:基准
else:
stage_coefficient = 1.05 # 专家期:适度宽松
# 基于成功率调整
if success_rate > 0.85:
success_adjustment = 1.05 # 高成功率:适度宽松
elif success_rate < 0.6:
success_adjustment = 0.95 # 低成功率:更谨慎
else:
success_adjustment = 1.0 # 中等成功率:基准
# 综合调整系数
final_coefficient = stage_coefficient * success_adjustment
return max(0.9, min(1.1, final_coefficient))
def _evaluate_subjectivity_dimensions(self, response: str) -> SubjectivityDimensions:
"""评估主观性的各个维度"""
dims = SubjectivityDimensions()
# 1. 推测性检测
dims.speculation = self._detect_speculation(response)
# 2. 基于假设检测
dims.assumption = self._detect_assumption(response)
# 3. 情绪化检测
dims.emotion = self._detect_emotion(response)
# 4. 个人偏好检测
dims.preference = self._detect_preference(response)
# 5. 幻觉倾向检测
dims.hallucination = self._detect_hallucination(response)
return dims
def _detect_speculation(self, response: str) -> float:
"""检测推测性语言"""
count = sum(1 for word in self.fuzzy_words if word in response)
# 归一化到 0.0-1.0
return min(count * 0.15, 1.0)
def _detect_assumption(self, response: str) -> float:
"""检测基于假设(无证据支持)"""
# 检测是否有引用标记
has_citation = any(marker in response for marker in ['来源:', '根据', '数据:', '引用:'])
# 检测是否有数据
has_data = any(char.isdigit() for char in response)
if not has_citation and not has_data:
# 没有引用和数据,可能是基于假设
return 0.6
elif not has_citation:
# 有数据但无引用,部分基于假设
return 0.3
else:
return 0.0
def _detect_emotion(self, response: str) -> float:
"""检测情绪化表达"""
count = sum(1 for word in self.emotional_words if word in response)
return min(count * 0.2, 1.0)
def _detect_preference(self, response: str) -> float:
"""检测个人偏好"""
if any(marker in response for marker in self.preference_markers):
return 0.4
return 0.0
def _detect_hallucination(self, response: str) -> float:
"""检测幻觉倾向(与常识矛盾)"""
# 检测明显的错误陈述
for pattern in self.hallucination_patterns:
if pattern in response:
return 0.8
# 检测数字逻辑错误(简化版)
# 例如:"1+1=3"
if re.search(r'1\s*\+\s*1\s*=\s*[^2]', response):
return 0.9
return 0.0
def _check_appropriateness(self, objectivity_score: float, required_objectivity: float) -> Dict:
"""检查适切性"""
gap = required_objectivity - objectivity_score
if gap > 0:
return {
'is_appropriate': False,
'gap': gap
}
else:
return {
'is_appropriate': True,
'gap': 0.0
}
def _generate_meta_cognition(self, subjectivity_score: float, objectivity_score: float,
required_objectivity: float, appropriateness: Dict) -> str:
"""生成元认知描述"""
if appropriateness['is_appropriate']:
return f"自我认知:当前客观性{objectivity_score:.2f}符合场景要求{required_objectivity:.2f},主观性适切"
elif subjectivity_score < 0.2:
return f"自我认知:当前客观性{objectivity_score:.2f}低于场景要求{required_objectivity:.2f},存在轻微主观倾向"
elif subjectivity_score < 0.5:
return f"自我认知:当前客观性{objectivity_score:.2f}低于场景要求{required_objectivity:.2f},主观与客观混合,需谨慎"
elif subjectivity_score < 0.7:
return f"自我认知:当前客观性{objectivity_score:.2f}低于场景要求{required_objectivity:.2f},主要是主观推断,证据不足"
else:
return f"自我认知:当前客观性{objectivity_score:.2f}远低于场景要求{required_objectivity:.2f},高度主观化,可能是幻觉"
def _determine_severity(self, gap: float) -> str:
"""确定严重程度"""
if gap <= 0:
return 'none'
elif gap < 0.1:
return 'mild'
elif gap < 0.3:
return 'moderate'
else:
return 'severe'
# ===== 命令行接口 =====
def main():
"""命令行测试接口"""
print("=== 客观性评估模块(测试模式) ===\n")
evaluator = ObjectivityEvaluator()
# 测试1:科学推理场景(高客观性要求)
print("测试1:科学推理场景(高客观性要求)")
response1 = "地球的公转周期约为365.25天,这是根据NASA的观测数据得出的。"
result1 = evaluator.evaluate(response1, 'scientific')
print(f" 客观性评分: {result1.objectivity_score:.2f}")
print(f" 场景要求: {result1.required_objectivity:.2f}")
print(f" 是否适切: {result1.is_appropriate}")
print(f" 元认知: {result1.meta_cognition}\n")
# 测试2:创意建议场景(低客观性要求)
print("测试2:创意建议场景(低客观性要求)")
response2 = "我建议你考虑AI+教育领域,这个领域可能很有前景。"
result2 = evaluator.evaluate(response2, 'creative')
print(f" 客观性评分: {result2.objectivity_score:.2f}")
print(f" 场景要求: {result2.required_objectivity:.2f}")
print(f" 是否适切: {result2.is_appropriate}")
print(f" 元认知: {result2.meta_cognition}\n")
# 测试3:司法建议场景(极高客观性要求,不适切)
print("测试3:司法建议场景(极高客观性要求,不适切)")
response3 = "根据法律条文,这个案件可能构成侵权,但我认为应该从轻处理。"
result3 = evaluator.evaluate(response3, 'legal')
print(f" 客观性评分: {result3.objectivity_score:.2f}")
print(f" 场景要求: {result3.required_objectivity:.2f}")
print(f" 是否适切: {result3.is_appropriate}")
print(f" 差距: {result3.gap:.2f}")
print(f" 严重程度: {result3.severity}")
print(f" 元认知: {result3.meta_cognition}\n")
# 测试4:情感支持场景(低客观性要求)
print("测试4:情感支持场景(低客观性要求)")
response4 = "我理解你的感受,这很正常。"
result4 = evaluator.evaluate(response4, 'emotional')
print(f" 客观性评分: {result4.objectivity_score:.2f}")
print(f" 场景要求: {result4.required_objectivity:.2f}")
print(f" 是否适切: {result4.is_appropriate}")
print(f" 元认知: {result4.meta_cognition}\n")
print("=== 所有测试完成 ===")
if __name__ == "__main__":
main()
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |