refactor: 重组项目目录结构
以讲义内容为骨架迁移到标准目录格式： - officefile/ 主内容（12章 + 附录 + CC4SI补充） - dofile/ 代码示例（11个Python脚本） - data/ 图片资源 - output/ 生成输出（忽略） - Archive/ 归档旧目录（忽略） - .claude/skills/ 保留markdown-to-docx工具链 - .pandoc/ 保留CSL和本地化配置 Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
@@ -0,0 +1,820 @@
|
||||
# 05.2 伦理与责任
|
||||
|
||||
## 核心问题
|
||||
|
||||
> 空间决策如何影响不同的人群和环境？
|
||||
> 当AI参与空间规划时，如何确保过程的公平和透明？
|
||||
> 出错的决策责任应该由谁来承担？
|
||||
|
||||
---
|
||||
|
||||
## 概念讲解
|
||||
|
||||
### 空间决策的伦理维度
|
||||
|
||||
空间决策不是价值中立的，它们分配资源、机会和风险：
|
||||
|
||||
```
|
||||
空间决策的伦理影响
|
||||
|
||||
┌─────────────────────────────────────────────────────────────┐
|
||||
│ │
|
||||
│ 1. 分配正义 (Distributive Justice) │
|
||||
│ - 谁获得绿地、公园等正面空间资源？ │
|
||||
│ - 谁承受污染、噪声等负面影响？ │
|
||||
│ - 空间资源的公平分配原则是什么？ │
|
||||
│ │
|
||||
│ 2. 程序正义 (Procedural Justice) │
|
||||
│ - 决策过程是否透明？ │
|
||||
│ - 受影响者能否参与决策？ │
|
||||
│ - 决策依据是否可审查？ │
|
||||
│ │
|
||||
│ 3. 承认正义 (Recognition Justice) │
|
||||
│ - 不同群体的需求和价值观是否被认可？ │
|
||||
│ - 弱势群体的空间权利是否被尊重？ │
|
||||
│ - 文化多样性在空间中如何体现？ │
|
||||
│ │
|
||||
│ 4. 生态正义 (Ecological Justice) │
|
||||
│ - 当代人与未来世代之间的公平？ │
|
||||
│ - 人类活动对生态系统的责任？ │
|
||||
│ - 非人类物种的空间权利？ │
|
||||
│ │
|
||||
└─────────────────────────────────────────────────────────────┘
|
||||
```
|
||||
|
||||
### AI带来的新伦理挑战
|
||||
|
||||
```
|
||||
AI空间决策的特有问题
|
||||
|
||||
1. 算法偏见 (Algorithmic Bias)
|
||||
训练数据中的社会偏见被编码进模型
|
||||
→ 历史上的红线政策可能影响现在的预测
|
||||
|
||||
2. 黑箱决策 (Black Box Decision)
|
||||
复杂模型的决策过程难以解释
|
||||
→ 利益相关者无法质疑或理解决策
|
||||
|
||||
3. 责任分散 (Diffused Responsibility)
|
||||
涉及多个主体：开发者、用户、数据提供者
|
||||
→ 出错时责任难以界定
|
||||
|
||||
4. 规模效应 (Scale Effects)
|
||||
AI可以大规模应用决策
|
||||
→ 小偏差在大规模下产生大影响
|
||||
|
||||
5. 路径依赖 (Path Dependence)
|
||||
早期决策影响后续数据收集
|
||||
→ 偏见自我强化
|
||||
```
|
||||
|
||||
### 空间正义的典型问题
|
||||
|
||||
| 问题类型 | 说明 | AI相关风险 |
|
||||
|---------|------|-----------|
|
||||
| **环境种族主义** | 有害设施更多位于少数族裔社区 | AI可能复制历史模式 |
|
||||
| **绿色绅士化** | 绿地改善导致原住民被迫搬迁 | AI优化可能加剧此问题 |
|
||||
| **数字鸿沟** | 缺乏数据地区被忽视 | AI只关注数据丰富的区域 |
|
||||
| **代表性不足** | 某些群体的需求未被考虑 | 训练数据偏差 |
|
||||
|
||||
---
|
||||
|
||||
## 设计原理
|
||||
|
||||
### 可解释性设计原则
|
||||
|
||||
```python
|
||||
"""
|
||||
可解释的空间AI设计
|
||||
"""
|
||||
from typing import Dict, List, Any, Optional
|
||||
from dataclasses import dataclass
|
||||
from abc import ABC, abstractmethod
|
||||
|
||||
@dataclass
|
||||
class DecisionExplanation:
|
||||
"""决策解释"""
|
||||
decision: str # 做出的决策
|
||||
rationale: List[str] # 决策理由
|
||||
key_factors: Dict[str, float] # 关键因素及权重
|
||||
alternatives: List[Dict] # 考虑过的替代方案
|
||||
uncertainties: List[str] # 不确定性说明
|
||||
assumptions: List[str] # 假设条件
|
||||
|
||||
class ExplainableSpatialAI(ABC):
|
||||
"""可解释的空间AI基类"""
|
||||
|
||||
@abstractmethod
|
||||
def make_decision(self, context: Dict) -> Any:
|
||||
"""做出决策"""
|
||||
pass
|
||||
|
||||
@abstractmethod
|
||||
def explain_decision(self, decision: Any, context: Dict) -> DecisionExplanation:
|
||||
"""解释决策"""
|
||||
pass
|
||||
|
||||
def audit_trail(self) -> List[Dict]:
|
||||
"""返回审计轨迹"""
|
||||
return self._audit_log
|
||||
|
||||
class ExplainableSiteSelector(ExplainableSpatialAI):
|
||||
"""可解释的选址AI"""
|
||||
|
||||
def __init__(self):
|
||||
self.criteria_weights = {}
|
||||
self._audit_log = []
|
||||
|
||||
def set_criteria(self, criteria: Dict[str, float], justification: str):
|
||||
"""
|
||||
设置评判标准
|
||||
|
||||
Args:
|
||||
criteria: {标准名: 权重}
|
||||
justification: 权重选择的理由
|
||||
"""
|
||||
self.criteria_weights = criteria.copy()
|
||||
self._log({
|
||||
'action': 'set_criteria',
|
||||
'criteria': criteria,
|
||||
'justification': justification
|
||||
})
|
||||
|
||||
def make_decision(self, context: Dict) -> Dict:
|
||||
"""
|
||||
做出选址决策
|
||||
|
||||
返回选中的地点及其评分
|
||||
"""
|
||||
sites = context['sites']
|
||||
constraints = context.get('constraints', {})
|
||||
|
||||
# 评估每个候选地
|
||||
scored_sites = []
|
||||
for site in sites:
|
||||
score, details = self._evaluate_site(site, context)
|
||||
scored_sites.append({
|
||||
'site': site,
|
||||
'score': score,
|
||||
'details': details
|
||||
})
|
||||
|
||||
# 排序并选择最高分
|
||||
scored_sites.sort(key=lambda x: x['score'], reverse=True)
|
||||
selected = scored_sites[0]
|
||||
|
||||
# 记录决策
|
||||
self._log({
|
||||
'action': 'make_decision',
|
||||
'selected': selected['site'],
|
||||
'score': selected['score'],
|
||||
'alternatives': scored_sites[1:4] # 保存前几个备选
|
||||
})
|
||||
|
||||
return selected
|
||||
|
||||
def explain_decision(self, decision: Any, context: Dict) -> DecisionExplanation:
|
||||
"""生成决策解释"""
|
||||
selected_site = decision['site']
|
||||
score_details = decision['details']
|
||||
|
||||
# 生成解释
|
||||
return DecisionExplanation(
|
||||
decision=f"选择地点 {selected_site['name']}",
|
||||
rationale=[
|
||||
f"该地点综合评分最高 ({decision['score']:.2f})",
|
||||
"评分基于预定义的标准和权重",
|
||||
"所有候选地点已被系统评估"
|
||||
],
|
||||
key_factors=score_details,
|
||||
alternatives=[
|
||||
{
|
||||
'site': alt['site']['name'],
|
||||
'score': alt['score'],
|
||||
'reason': '评分较低'
|
||||
}
|
||||
for alt in context.get('alternatives', [])[:3]
|
||||
],
|
||||
uncertainties=[
|
||||
"评分依赖输入数据的准确性",
|
||||
"权重选择包含主观判断",
|
||||
"未量化的因素可能影响实际适用性"
|
||||
],
|
||||
assumptions=[
|
||||
"所有标准可以用数值表示",
|
||||
"各标准相互独立",
|
||||
"当前条件在未来保持稳定"
|
||||
]
|
||||
)
|
||||
|
||||
def _evaluate_site(self, site: Dict, context: Dict) -> tuple:
|
||||
"""评估单个地点"""
|
||||
scores = {}
|
||||
|
||||
for criterion, weight in self.criteria_weights.items():
|
||||
# 从地点数据中获取该标准的值
|
||||
value = site.get(criterion, 0)
|
||||
|
||||
# 标准化（简化版）
|
||||
normalized = self._normalize(criterion, value)
|
||||
|
||||
# 加权
|
||||
scores[criterion] = normalized * weight
|
||||
|
||||
total_score = sum(scores.values())
|
||||
|
||||
return total_score, scores
|
||||
|
||||
def _normalize(self, criterion: str, value: float) -> float:
|
||||
"""标准化准则值"""
|
||||
# 简化：假设越大越好，范围0-100
|
||||
return min(max(value / 100, 0), 1)
|
||||
|
||||
def _log(self, entry: Dict):
|
||||
"""记录日志"""
|
||||
entry['timestamp'] = self._get_timestamp()
|
||||
self._audit_log.append(entry)
|
||||
|
||||
def _get_timestamp(self) -> str:
|
||||
"""获取时间戳"""
|
||||
from datetime import datetime
|
||||
return datetime.now().isoformat()
|
||||
|
||||
# === 伦理检查 ===
|
||||
|
||||
class EthicsChecker:
|
||||
"""伦理检查器"""
|
||||
|
||||
def __init__(self):
|
||||
self.checks = []
|
||||
|
||||
def add_check(self, check_fn, name: str):
|
||||
"""添加检查"""
|
||||
self.checks.append((check_fn, name))
|
||||
return self
|
||||
|
||||
def check_decision(self, decision: Any, context: Dict) -> Dict:
|
||||
"""执行所有伦理检查"""
|
||||
results = {
|
||||
'passed': True,
|
||||
'issues': [],
|
||||
'warnings': []
|
||||
}
|
||||
|
||||
for check_fn, name in self.checks:
|
||||
try:
|
||||
result = check_fn(decision, context)
|
||||
if not result['passed']:
|
||||
results['passed'] = False
|
||||
results['issues'].append({
|
||||
'check': name,
|
||||
'reason': result['reason']
|
||||
})
|
||||
elif result.get('warning'):
|
||||
results['warnings'].append({
|
||||
'check': name,
|
||||
'warning': result['warning']
|
||||
})
|
||||
except Exception as e:
|
||||
results['issues'].append({
|
||||
'check': name,
|
||||
'reason': f"检查失败: {str(e)}"
|
||||
})
|
||||
|
||||
return results
|
||||
|
||||
# 预定义的伦理检查
|
||||
|
||||
def check_environmental_justice(decision, context) -> Dict:
|
||||
"""检查环境正义：确保不将负面影响集中到弱势社区"""
|
||||
selected_site = decision['site']
|
||||
|
||||
# 检查是否有弱势群体数据
|
||||
vulnerable_communities = context.get('vulnerable_communities', [])
|
||||
|
||||
for community in vulnerable_communities:
|
||||
if selected_site.get('near_community') == community['id']:
|
||||
# 如果项目有负面影响，需要特别审查
|
||||
if context.get('project_type') == 'negative_impact':
|
||||
return {
|
||||
'passed': False,
|
||||
'reason': f"选址靠近弱势社区 {community['name']}，需要额外的环境正义审查"
|
||||
}
|
||||
|
||||
return {'passed': True}
|
||||
|
||||
def check_transparency(decision, context) -> Dict:
|
||||
"""检查透明度：确保决策过程可记录和审查"""
|
||||
if not decision.get('details'):
|
||||
return {
|
||||
'passed': False,
|
||||
'reason': '决策缺乏详细评分信息，无法审查'
|
||||
}
|
||||
|
||||
return {'passed': True}
|
||||
|
||||
def check_public_participation(decision, context) -> Dict:
|
||||
"""检查公众参与：确保受影响者有机会表达意见"""
|
||||
if context.get('affects_public', False):
|
||||
participation = context.get('public_participation')
|
||||
if not participation or participation == 'none':
|
||||
return {
|
||||
'passed': False,
|
||||
'reason': '项目影响公众但缺乏公众参与程序'
|
||||
}
|
||||
elif participation == 'minimal':
|
||||
return {
|
||||
'passed': True,
|
||||
'warning': '公众参与程度较低，建议加强'
|
||||
}
|
||||
|
||||
return {'passed': True}
|
||||
|
||||
# === 使用示例 ===
|
||||
|
||||
if __name__ == "__main__":
|
||||
print("=== 可解释的空间AI ===\n")
|
||||
|
||||
# 创建选址器
|
||||
selector = ExplainableSiteSelector()
|
||||
|
||||
# 设置评判标准（带理由）
|
||||
selector.set_criteria(
|
||||
criteria={
|
||||
'accessibility': 0.3,
|
||||
'environmental_quality': 0.25,
|
||||
'cost_effectiveness': 0.2,
|
||||
'community_support': 0.15,
|
||||
'future_potential': 0.1
|
||||
},
|
||||
justification="基于项目目标和利益相关者访谈"
|
||||
)
|
||||
|
||||
# 模拟候选地点
|
||||
sites = [
|
||||
{'name': 'Site A', 'accessibility': 85, 'environmental_quality': 70,
|
||||
'cost_effectiveness': 60, 'community_support': 80, 'future_potential': 75},
|
||||
{'name': 'Site B', 'accessibility': 70, 'environmental_quality': 85,
|
||||
'cost_effectiveness': 75, 'community_support': 60, 'future_potential': 70},
|
||||
{'name': 'Site C', 'accessibility': 90, 'environmental_quality': 60,
|
||||
'cost_effectiveness': 80, 'community_support': 70, 'future_potential': 65},
|
||||
]
|
||||
|
||||
# 创建伦理检查器
|
||||
ethics_checker = EthicsChecker()
|
||||
ethics_checker.add_check(check_environmental_justice, "环境正义检查")
|
||||
ethics_checker.add_check(check_transparency, "透明度检查")
|
||||
ethics_checker.add_check(check_public_participation, "公众参与检查")
|
||||
|
||||
# 上下文
|
||||
context = {
|
||||
'sites': sites,
|
||||
'vulnerable_communities': [],
|
||||
'project_type': 'neutral',
|
||||
'affects_public': True,
|
||||
'public_participation': 'moderate'
|
||||
}
|
||||
|
||||
# 做出决策
|
||||
decision = selector.make_decision(context)
|
||||
|
||||
print(f"选中地点: {decision['site']['name']}")
|
||||
print(f"综合评分: {decision['score']:.2f}\n")
|
||||
|
||||
# 获取解释
|
||||
explanation = selector.explain_decision(decision, context)
|
||||
print("=== 决策解释 ===")
|
||||
print(f"决策: {explanation.decision}")
|
||||
print(f"\n理由:")
|
||||
for r in explanation.rationale:
|
||||
print(f" - {r}")
|
||||
print(f"\n关键因素:")
|
||||
for factor, value in explanation.key_factors.items():
|
||||
print(f" - {factor}: {value:.3f}")
|
||||
|
||||
# 伦理检查
|
||||
print(f"\n=== 伦理检查 ===")
|
||||
ethics_result = ethics_checker.check_decision(decision, context)
|
||||
if ethics_result['passed']:
|
||||
print("所有伦理检查通过")
|
||||
else:
|
||||
print("伦理检查发现问题:")
|
||||
for issue in ethics_result['issues']:
|
||||
print(f" - [{issue['check']}] {issue['reason']}")
|
||||
```
|
||||
|
||||
### 问责机制设计
|
||||
|
||||
```
|
||||
AI空间决策的问责框架
|
||||
|
||||
┌─────────────────────────────────────────────────────────────┐
|
||||
│ │
|
||||
│ 责任链 │
|
||||
│ │
|
||||
│ 数据提供者 ──→ 模型开发者 ──→ 系统集成者 ──→ 最终用户 │
|
||||
│ │ │ │ │ │
|
||||
│ │ │ │ └── 决策责任 │
|
||||
│ │ │ └── 集成责任 │
|
||||
│ │ └── 模型责任 │
|
||||
│ └── 数据质量责任 │
|
||||
│ │
|
||||
│ 问责机制 │
|
||||
│ │
|
||||
│ 1. 文档化 (Documentation) │
|
||||
│ - 记录所有决策和数据来源 │
|
||||
│ - 保存模型版本和参数 │
|
||||
│ - 维护变更历史 │
|
||||
│ │
|
||||
│ 2. 审计 (Audit) │
|
||||
│ - 定期审查决策 │
|
||||
│ - 检查偏见和公平性 │
|
||||
│ - 验证技术正确性 │
|
||||
│ │
|
||||
│ 3. 申诉 (Appeal) │
|
||||
│ - 提供质疑决策的渠道 │
|
||||
│ - 建立复核机制 │
|
||||
│ - 允许人工干预 │
|
||||
│ │
|
||||
│ 4. 纠正 (Remedy) │
|
||||
│ - 发现错误后的补救措施 │
|
||||
│ - 对受影响方的补偿 │
|
||||
│ - 系统改进和预防 │
|
||||
│ │
|
||||
└─────────────────────────────────────────────────────────────┘
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 案例分析
|
||||
|
||||
### 案例1：城市绿地分布的算法偏见
|
||||
|
||||
**问题背景**：某城市使用AI优化绿地布局，但结果加剧了既有不平等。
|
||||
|
||||
```python
|
||||
"""
|
||||
问题代码示例：训练数据中的历史偏见
|
||||
"""
|
||||
|
||||
# 问题：使用历史公园使用数据来优化新公园位置
|
||||
def optimize_park_locations_biased(historical_usage_data, new_sites):
|
||||
"""
|
||||
有偏见的优化算法
|
||||
|
||||
问题：历史使用数据反映的是历史可达性，
|
||||
而非真实需求。服务不足的区域数据少，
|
||||
因此被算法继续忽视。
|
||||
"""
|
||||
# 简单优化：在历史使用高的地方附近选址
|
||||
scored = []
|
||||
for site in new_sites:
|
||||
# 靠近高使用区域得分高
|
||||
score = sum(
|
||||
usage for nearby, usage in historical_usage_data
|
||||
if distance(site, nearby) < 1000
|
||||
)
|
||||
scored.append((site, score))
|
||||
|
||||
# 选择得分最高的
|
||||
scored.sort(key=lambda x: x[1], reverse=True)
|
||||
return [s[0] for s in scored[:5]]
|
||||
|
||||
# 改进版本：考虑需求而非历史使用
|
||||
def optimize_park_locations_fair(demand_indicators, new_sites,
|
||||
equity_weight=0.5):
|
||||
"""
|
||||
公平的优化算法
|
||||
|
||||
考虑：
|
||||
1. 当前服务不足程度（需求）
|
||||
2. 人口密度
|
||||
3. 弱势群体分布
|
||||
"""
|
||||
scored = []
|
||||
for site in new_sites:
|
||||
# 服务不足得分
|
||||
underserved_score = calculate_underserved(site, demand_indicators)
|
||||
|
||||
# 效率得分（可达人口）
|
||||
efficiency_score = calculate_accessible_population(site)
|
||||
|
||||
# 综合得分，可调整公平权重
|
||||
score = (1 - equity_weight) * efficiency_score + \
|
||||
equity_weight * underserved_score
|
||||
|
||||
scored.append((site, score, {
|
||||
'underserved': underserved_score,
|
||||
'efficiency': efficiency_score
|
||||
}))
|
||||
|
||||
# 按综合得分排序
|
||||
scored.sort(key=lambda x: x[1], reverse=True)
|
||||
|
||||
# 记录决策依据
|
||||
for site, score, details in scored:
|
||||
site['selection_score'] = score
|
||||
site['score_details'] = details
|
||||
|
||||
return [s[0] for s in scored[:5]]
|
||||
|
||||
def calculate_underserved(site, indicators):
|
||||
"""计算服务不足程度"""
|
||||
# 距离最近的现有设施
|
||||
distance_to_nearest = min_distance_to_parks(site)
|
||||
|
||||
# 附近人口中的弱势群体比例
|
||||
vulnerable_ratio = get_vulnerable_population_ratio(site)
|
||||
|
||||
# 服务不足 = 距离远 + 弱势群体多
|
||||
return distance_to_nearest * (1 + vulnerable_ratio)
|
||||
```
|
||||
|
||||
### 案例2：生态保护区的社区影响
|
||||
|
||||
**问题背景**：AI优化的生态廊道选址忽略了当地社区权益。
|
||||
|
||||
```python
|
||||
"""
|
||||
考虑多方利益的生态廊道选址
|
||||
"""
|
||||
|
||||
class EthicalCorridorSelector:
|
||||
"""伦理导向的廊道选址器"""
|
||||
|
||||
def __init__(self):
|
||||
self.stakeholders = {
|
||||
'ecology': {'weight': 0.4, 'concern': '生态连通性'},
|
||||
'community': {'weight': 0.3, 'concern': '社区利益'},
|
||||
'economy': {'weight': 0.2, 'concern': '经济成本'},
|
||||
'culture': {'weight': 0.1, 'concern': '文化价值'}
|
||||
}
|
||||
|
||||
def evaluate_corridor_route(self, route, context):
|
||||
"""
|
||||
评估廊道路线
|
||||
|
||||
返回：综合评分和各利益相关方的影响
|
||||
"""
|
||||
scores = {}
|
||||
|
||||
# 生态评分
|
||||
scores['ecology'] = self._evaluate_ecological_value(route, context)
|
||||
|
||||
# 社区评分
|
||||
scores['community'] = self._evaluate_community_impact(route, context)
|
||||
|
||||
# 经济评分
|
||||
scores['economy'] = self._evaluate_economic_cost(route, context)
|
||||
|
||||
# 文化评分
|
||||
scores['culture'] = self._evaluate_cultural_impact(route, context)
|
||||
|
||||
# 加权综合
|
||||
total = sum(
|
||||
scores[stakeholder] * self.stakeholders[stakeholder]['weight']
|
||||
for stakeholder in self.stakeholders
|
||||
)
|
||||
|
||||
# 检查任何一方的严重负面影响
|
||||
for stakeholder, score in scores.items():
|
||||
if score < 0.3: # 阈值
|
||||
return {
|
||||
'acceptable': False,
|
||||
'reason': f"{stakeholder}评分过低: {score:.2f}",
|
||||
'scores': scores,
|
||||
'total': total
|
||||
}
|
||||
|
||||
return {
|
||||
'acceptable': True,
|
||||
'total_score': total,
|
||||
'scores': scores,
|
||||
'breakdown': {
|
||||
stakeholder: {
|
||||
'score': scores[stakeholder],
|
||||
'weight': self.stakeholders[stakeholder]['weight'],
|
||||
'concern': self.stakeholders[stakeholder]['concern']
|
||||
}
|
||||
for stakeholder in self.stakeholders
|
||||
}
|
||||
}
|
||||
|
||||
def _evaluate_ecological_value(self, route, context):
|
||||
"""评估生态价值"""
|
||||
# 连通的源地质量
|
||||
source_quality = self._connected_source_quality(route, context)
|
||||
|
||||
# 廊道宽度
|
||||
width_score = min(route['width'] / 100, 1.0)
|
||||
|
||||
# 栖息地适宜性
|
||||
habitat_score = self._habitat_suitability(route, context)
|
||||
|
||||
return (source_quality + width_score + habitat_score) / 3
|
||||
|
||||
def _evaluate_community_impact(self, route, context):
|
||||
"""评估社区影响"""
|
||||
# 正面：休闲价值
|
||||
recreational_value = self._recreational_potential(route)
|
||||
|
||||
# 负面：拆迁、限制使用
|
||||
negative_impact = self._negative_community_impact(route, context)
|
||||
|
||||
return max(0, recreational_value - negative_impact)
|
||||
|
||||
def _evaluate_economic_cost(self, route, context):
|
||||
"""评估经济成本（分数越高表示成本越可接受）"""
|
||||
# 土地获取成本
|
||||
land_cost = route.get('land_cost', 0)
|
||||
|
||||
# 建设成本
|
||||
construction_cost = route.get('construction_cost', 0)
|
||||
|
||||
# 归一化：成本越低分数越高
|
||||
max_cost = context.get('max_budget', float('inf'))
|
||||
total_cost = land_cost + construction_cost
|
||||
|
||||
if total_cost > max_cost:
|
||||
return 0 # 超预算
|
||||
else:
|
||||
return 1 - (total_cost / max_cost) * 0.5
|
||||
|
||||
def _evaluate_cultural_impact(self, route, context):
|
||||
"""评估文化影响"""
|
||||
# 是否涉及文化遗产
|
||||
cultural_sites = route.get('cultural_sites', [])
|
||||
if cultural_sites:
|
||||
return 0.3 # 低分，需要特别处理
|
||||
|
||||
# 是否支持传统文化活动
|
||||
traditional_use = route.get('supports_traditional_use', False)
|
||||
if traditional_use:
|
||||
return 1.0
|
||||
|
||||
return 0.7 # 中性
|
||||
|
||||
# ... 其他辅助方法 ...
|
||||
|
||||
if __name__ == "__main__":
|
||||
print("=== 伦理导向的廊道选址 ===\n")
|
||||
|
||||
selector = EthicalCorridorSelector()
|
||||
|
||||
# 示例路线
|
||||
route = {
|
||||
'width': 80,
|
||||
'land_cost': 500000,
|
||||
'construction_cost': 1000000,
|
||||
'cultural_sites': [],
|
||||
'supports_traditional_use': True
|
||||
}
|
||||
|
||||
context = {
|
||||
'max_budget': 2000000,
|
||||
'ecological_data': {},
|
||||
'community_data': {}
|
||||
}
|
||||
|
||||
result = selector.evaluate_corridor_route(route, context)
|
||||
|
||||
if result['acceptable']:
|
||||
print(f"路线可接受，综合评分: {result['total_score']:.2f}")
|
||||
print("\n各利益相关方评分:")
|
||||
for stakeholder, info in result['breakdown'].items():
|
||||
print(f" {stakeholder}: {info['score']:.2f} "
|
||||
f"(权重: {info['weight']}, 关注: {info['concern']})")
|
||||
else:
|
||||
print(f"路线不可接受: {result['reason']}")
|
||||
```
|
||||
|
||||
### 案例3：透明度和可审计性
|
||||
|
||||
```python
|
||||
"""
|
||||
决策日志系统
|
||||
"""
|
||||
import json
|
||||
from datetime import datetime
|
||||
from typing import Dict, Any, List
|
||||
|
||||
class DecisionLog:
|
||||
"""决策日志系统"""
|
||||
|
||||
def __init__(self, project_id: str):
|
||||
self.project_id = project_id
|
||||
self.entries: List[Dict] = []
|
||||
|
||||
def log_decision(self, decision_type: str, decision: Any,
|
||||
rationale: str, alternatives: List[Dict],
|
||||
metadata: Dict = None):
|
||||
"""记录决策"""
|
||||
entry = {
|
||||
'timestamp': datetime.now().isoformat(),
|
||||
'project_id': self.project_id,
|
||||
'decision_type': decision_type,
|
||||
'decision': decision,
|
||||
'rationale': rationale,
|
||||
'alternatives': alternatives,
|
||||
'metadata': metadata or {}
|
||||
}
|
||||
self.entries.append(entry)
|
||||
|
||||
def log_data_source(self, data_type: str, source: str,
|
||||
quality: Dict, limitations: List[str]):
|
||||
"""记录数据来源"""
|
||||
entry = {
|
||||
'timestamp': datetime.now().isoformat(),
|
||||
'project_id': self.project_id,
|
||||
'type': 'data_source',
|
||||
'data_type': data_type,
|
||||
'source': source,
|
||||
'quality': quality,
|
||||
'limitations': limitations
|
||||
}
|
||||
self.entries.append(entry)
|
||||
|
||||
def log_model_info(self, model_name: str, version: str,
|
||||
training_data: Dict, limitations: List[str]):
|
||||
"""记录模型信息"""
|
||||
entry = {
|
||||
'timestamp': datetime.now().isoformat(),
|
||||
'project_id': self.project_id,
|
||||
'type': 'model_info',
|
||||
'model_name': model_name,
|
||||
'version': version,
|
||||
'training_data': training_data,
|
||||
'limitations': limitations
|
||||
}
|
||||
self.entries.append(entry)
|
||||
|
||||
def export_audit_report(self) -> str:
|
||||
"""导出审计报告"""
|
||||
report = {
|
||||
'project_id': self.project_id,
|
||||
'export_time': datetime.now().isoformat(),
|
||||
'entries': self.entries,
|
||||
'summary': self._generate_summary()
|
||||
}
|
||||
return json.dumps(report, indent=2, ensure_ascii=False)
|
||||
|
||||
def _generate_summary(self) -> Dict:
|
||||
"""生成摘要"""
|
||||
summary = {
|
||||
'total_entries': len(self.entries),
|
||||
'decision_types': {},
|
||||
'data_sources': [],
|
||||
'models_used': []
|
||||
}
|
||||
|
||||
for entry in self.entries:
|
||||
if entry.get('type') == 'data_source':
|
||||
summary['data_sources'].append(entry['data_type'])
|
||||
elif entry.get('type') == 'model_info':
|
||||
summary['models_used'].append(entry['model_name'])
|
||||
elif 'decision_type' in entry:
|
||||
dt = entry['decision_type']
|
||||
summary['decision_types'][dt] = \
|
||||
summary['decision_types'].get(dt, 0) + 1
|
||||
|
||||
return summary
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 反思与延伸
|
||||
|
||||
### 思考问题
|
||||
|
||||
1. **价值权衡**：当生态目标和社会目标冲突时，应该如何权衡？谁有权决定？
|
||||
|
||||
2. **偏见识别**：你的空间分析可能隐含哪些偏见？如何检测？
|
||||
|
||||
3. **透明度边界**：哪些决策细节必须公开？哪些可以保密？
|
||||
|
||||
4. **长期责任**：AI辅助的决策出现问题后，如何追溯和纠正？
|
||||
|
||||
### 实践练习
|
||||
|
||||
1. **伦理审计**：对你做过的一个空间项目进行伦理审计
|
||||
|
||||
2. **利益相关者地图**：绘制项目的利益相关者及其关注点
|
||||
|
||||
3. **透明度检查**：为你的分析流程建立可审计的文档体系
|
||||
|
||||
### 延伸阅读
|
||||
|
||||
- **"Weapons of Math Destruction"** (Cathy O'Neil) - 算法的社会影响
|
||||
- **"The Alignment Problem"** (Brian Christian) - AI对齐问题
|
||||
- **"Spatial Justice"** (Edward Soja) - 空间正义理论
|
||||
- UN-Habitat's ethics guidelines for spatial planning
|
||||
|
||||
---
|
||||
|
||||
## 关键要点
|
||||
|
||||
1. **空间决策具有深刻的伦理维度**，影响资源分配和社会正义
|
||||
2. **AI可能放大既有偏见**，需要主动的公平性设计
|
||||
3. **可解释性是负责任AI的基础**，决策过程应可审查
|
||||
4. **建立清晰的问责机制**，明确各方责任
|
||||
5. **伦理思考应该贯穿整个项目生命周期**，而非事后补充
|
||||
Reference in New Issue
Block a user