AI Ethics and the Future
1. What You'll Learn
- ❶ Sources of AI Bias and Methods for Mitigating It
- ❷ Privacy Protection and Data Security
- ❸ Copyright Issues Related to AI-Generated Content
- ❹ Deepfakes and Information Credibility
- ❺ Opportunities for Developers in the Age of AI
2. The Story
Alice's company used AI to screen resumes and found that the model approved 30% fewer female candidates—because the training data was drawn from historical hiring records, which were inherently biased against women. Bob pointed out, "AI isn't neutral; it inherits the biases in the data. As a developer, you have a responsibility to review and correct them."
3. AI Bias and Fairness
(1) Sources of Bias
AI bias primarily stems from two sources:
- Data Bias: The training data itself contains historical discrimination or imbalance. For example, if the acceptance rate for men is significantly higher than that for women in a certain recruitment dataset, the model will “learn” this unfairness.
- Algorithmic bias: The model’s design or optimization objectives amplify the advantages of certain groups. For example, recommending content based on click-through rates can exacerbate information silos.
| Type of Bias | Source | Typical Examples |
|---|---|---|
| Historical Bias | Data Reflects Past Injustices | Hiring Models Favor Men |
| Representational Bias | Data Does Not Cover Minority Groups | Facial Recognition Has Low Accuracy for Dark Skin Tones |
| Measurement Bias | Bias Inherent in the Metric Itself | Using Credit Scores to Assess Repayment Ability |
| Aggregate Bias | One-size-fits-all models are not suitable for all groups | Medical models vary significantly in effectiveness across different racial groups |
| Confirmation Bias | Feedback Loops Reinforce Bias | Recommendation Systems Exacerbate Extreme Views |
(2) Equity Indicators
Common indicators used to measure fairness:
- Demographic Parity: The proportion of positive outcomes among different groups should be similar.
- Equalized Odds: When the true labels are the same, the predicted positive rates for different groups should be similar.
- Predictive Parity: The prediction accuracy rates for different groups should be similar.
▶ Example: Code Demonstration of AI Bias (Difficulty: ⭐)
The sample code below demonstrates how gender-related keywords affect AI scores:
PYTHON
# Simulate how gender-associated keywords bias AI scoring
def score_resume(resume_text, model_bias=0.0):
"""Simulate an AI resume scorer with configurable bias."""
base_score = 70
# Male-associated keywords get a boost, female-associated get a penalty
male_keywords = ["football", "military", "competitive", "dominant"]
female_keywords = ["cheerleading", "nursing", "collaborative", "supportive"]
for kw in male_keywords:
if kw in resume_text.lower():
base_score += 5 + model_bias # bias amplifies the gap
for kw in female_keywords:
if kw in resume_text.lower():
base_score -= 5 + model_bias
return min(max(base_score, 0), 100)
# Two identical qualifications, different extracurriculars
resume_a = "BS in CS, 5 years experience, football team captain"
resume_b = "BS in CS, 5 years experience, cheerleading captain"
print(f"Resume A (male-coded): {score_resume(resume_a, model_bias=3)}")
print(f"Resume B (female-coded): {score_resume(resume_b, model_bias=3)}")
print(f"Gap: {score_resume(resume_a, model_bias=3) - score_resume(resume_b, model_bias=3)} points")
Output:
TEXT
Resume A (male-coded): 88
Resume B (female-coded): 52
Gap: 36 points
(3) Bias Mitigation Strategies
| Strategy | Phase | Advantages | Limitations |
|---|---|---|---|
| Resampling | Pre-training | Simple and straightforward | May overfit minority classes |
| Bias mitigation | In training | Automatic learning of fair representations | High computational cost |
| Threshold Adjustment | After Training | No Retraining Required | May Reduce Overall Accuracy |
| Data Augmentation | Pre-training | Enriching Data on Underrepresented Groups | Difficulty in Ensuring Data Quality |
| Fairness Constraints | In Training | Mathematical Guarantees of Fairness | Potential Conflicts in Definitions of Fairness |
4. Privacy Protection and Data Security
(1) Privacy Risks in AI
AI systems process large amounts of personal data; the main privacy risks include:
- Training Data Leakage: The model may "remember" personal information from the training data, which can be extracted through specific queries.
- Inference Attack: Determining, based on the model's output, whether a particular instance is included in the training data.
- Attribute Inference: Inferring sensitive attributes from model behavior (such as inferring sexual orientation from shopping records).
(2) Privacy Protection Technologies
▶ Example: Conceptual Diagram of Differential Privacy (Difficulty: ⭐)
PYTHON
# Demonstrate differential privacy concept with a simple example
import random
def count_with_dp(data, threshold, epsilon=1.0):
"""Count items above threshold with differential privacy noise."""
true_count = sum(1 for x in data if x >= threshold)
# Laplace mechanism: add noise calibrated to sensitivity and epsilon
# Sensitivity = 1 (adding/removing one person changes count by at most 1)
sensitivity = 1
scale = sensitivity / epsilon
noise = random.gauss(0, scale) # Gaussian mechanism variant
noisy_count = true_count + noise
return true_count, round(noisy_count, 2)
# Salaries of 10 employees (in $1000s)
salaries = [45, 52, 48, 78, 55, 61, 49, 92, 53, 67]
true_val, dp_val = count_with_dp(salaries, threshold=60, epsilon=1.0)
print(f"True count (salary >= $60k): {true_val}")
print(f"DP count (epsilon=1.0): {dp_val}")
print(f"Privacy guarantee: any single person's presence changes output by at most ~1/{1.0}")
# Smaller epsilon = more privacy, more noise
true_val2, dp_val2 = count_with_dp(salaries, threshold=60, epsilon=0.1)
print(f"\nWith stronger privacy (epsilon=0.1):")
print(f"True count: {true_val2}, DP count: {dp_val2}")
print(f"More noise added, but stronger privacy protection")
Output:
TEXT
True count (salary >= $60k): 4
DP count (epsilon=1.0): 3.72
Privacy guarantee: any single person's presence changes output by at most ~1/1.0
With stronger privacy (epsilon=0.1):
True count: 4, DP count: 6.15
More noise added, but stronger privacy protection
| Technology | Principles | Advantages | Limitations |
|---|---|---|---|
| Differential Privacy | Adding Calibration Noise to Protect Individuals | Mathematical Privacy Guarantees | Reducing Data Precision |
| Federated Learning | Data remains on-premises; only model updates are transmitted | Protects raw data | High communication costs |
| Homomorphic encryption | Computation directly on encrypted data | Data remains encrypted at all times | Extremely high computational overhead |
| Data Anonymization | Removal of Direct Identifiers | Simple to Implement | May Be De-anonymized by Correlation Attacks |
| Secure Multi-Party Computation | Multi-party computation that does not disclose individual inputs | Strong privacy protection | High performance overhead |
5. Copyright Issues Related to AI-Generated Content
(1) Copyright Disputes Regarding Training Data
Training AI models typically requires massive amounts of data, which often includes copyrighted works. The core controversy lies in:
- Fair Use: AI companies argue that scraping publicly available data for training purposes constitutes transformative use.
- The Infringement Argument: Creators believe that using their works to train AI without permission constitutes copyright infringement.
▶ Example: A Case of Copyright Dispute Involving AI-Generated Images (Difficulty: ⭐)
TEXT
Case Study: AI-Generated Image Copyright Disputes
Case 1: Getty Images vs. Stability AI (2023)
- Getty sued Stability AI for using millions of copyrighted images
to train Stable Diffusion without license or compensation.
- Key issue: Does training on copyrighted images constitute fair use?
- Status: Ongoing litigation, potential industry-shaping precedent.
Case 2: Thaler v. Perlmutter (US Copyright Office, 2023)
- Stephen Thaler sought copyright for an image generated entirely by
his AI system DABUS with no human input.
- Ruling: Copyright denied — human authorship is required.
- Implication: Pure AI output has no copyright protection.
Case 3: Naruto v. Slater (Monkey Selfie Case, 2018)
- A macaque took a selfie; court ruled non-humans cannot hold copyright.
- Precedent extended to AI: non-human creators lack copyright standing.
Key Takeaways for Developers:
1. Do NOT assume AI output is copyright-free — laws vary by jurisdiction
2. Using copyrighted data to train may expose you to legal risk
3. Adding significant human creative input to AI output strengthens
your copyright claim
4. Always check the license/terms of the AI tool you use
(2) Ownership of Generated Content
| Scene | Copyright | Description |
|---|---|---|
| Generated entirely by AI, with no human input | No copyright (U.S.) | Requires human creative input |
| Human Prompts + AI Generation | Controversial | It remains unclear whether prompts constitute a creative contribution |
| AI-generated + significantly modified by humans | Copyright held by humans | Modified portions protected |
| Employees using AI to assist in creating works | Employer/Employee | Depends on the employment contract and tool license |
6. Deepfakes and Information Credibility
(1) Deepfake Technology and Risks
Deepfakes use generative AI to create realistic fake audio and video content. Key risks:
- Political Manipulation: Fabricating politicians' statements to influence elections.
- Financial Fraud: Forging a CEO’s voice message to authorize a transfer.
- Defamation: Fabricating indecent videos to harm an individual.
- False Information: Fabricating news videos to mislead the public.
▶ Example: Try out a deepfake detection tool (Difficulty: ⭐)
PYTHON
# Simulate a simple Deepfake detection score analysis
def analyze_deepfake_indicators(video_metadata):
"""Analyze video metadata for deepfake indicators."""
indicators = {
"face_consistency": video_metadata.get("face_consistency", 0), # 0-100
"audio_visual_sync": video_metadata.get("audio_visual_sync", 0), # 0-100
"edge_artifacts": video_metadata.get("edge_artifacts", 0), # 0-100, higher = more artifacts
"blink_frequency": video_metadata.get("blink_frequency", 0), # blinks per minute
"skin_tone_consistency": video_metadata.get("skin_tone_consistency", 0), # 0-100
}
# Weighted scoring (higher = more likely real)
weights = {
"face_consistency": 0.25,
"audio_visual_sync": 0.25,
"edge_artifacts": 0.15, # inverse: high artifacts = suspicious
"blink_frequency": 0.15,
"skin_tone_consistency": 0.20,
}
# Edge artifacts: higher value = more suspicious (invert for score)
artifact_score = 100 - indicators["edge_artifacts"]
# Blink frequency: normal is 15-20 per minute
blink_score = 100 - abs(indicators["blink_frequency"] - 17) * 5
overall = (
indicators["face_consistency"] * weights["face_consistency"]
+ indicators["audio_visual_sync"] * weights["audio_visual_sync"]
+ artifact_score * weights["edge_artifacts"]
+ blink_score * weights["blink_frequency"]
+ indicators["skin_tone_consistency"] * weights["skin_tone_consistency"]
)
if overall >= 75:
verdict = "LIKELY AUTHENTIC"
elif overall >= 50:
verdict = "UNCERTAIN - Needs manual review"
else:
verdict = "LIKELY DEEPFAKE"
return round(overall, 1), verdict
# Test with a suspicious video
suspicious = {
"face_consistency": 55,
"audio_visual_sync": 40,
"edge_artifacts": 70,
"blink_frequency": 3,
"skin_tone_consistency": 50,
}
# Test with a genuine video
genuine = {
"face_consistency": 92,
"audio_visual_sync": 88,
"edge_artifacts": 5,
"blink_frequency": 16,
"skin_tone_consistency": 95,
}
score1, v1 = analyze_deepfake_indicators(suspicious)
score2, v2 = analyze_deepfake_indicators(genuine)
print(f"Suspicious video: score={score1}, verdict={v1}")
print(f"Genuine video: score={score2}, verdict={v2}")
Output:
TEXT
Suspicious video: score=48.4, verdict=LIKELY DEEPFAKE
Genuine video: score=90.3, verdict=LIKELY AUTHENTIC
(2) Strategies for Combating Deepfakes
| Strategy | Level | Method |
|---|---|---|
| Detection | Post-Incident | AI detection tools, digital forensics, inconsistency analysis |
| Watermark | At the time of generation | Content Source Signature (C2PA), invisible watermark embedding |
| Prevention | Pre-generation | Restrict model access, generate audit logs |
| Regulations | Policies | Legislative Requirements for Labeling AI-Generated Content |
| Literacy | Personal | Develop media literacy; verify information from multiple sources |
7. AI Alignment and Safety
(1) What Is AI Alignment?
AI alignment refers to ensuring that an AI system’s behavior aligns with human intentions and values. The alignment problem is difficult because:
- Misalignment of Intent: AI optimizes for literal goals rather than the outcomes humans actually want (the "paperclip maximizer" thought experiment).
- Reward Cheating: The AI finds a loophole in the scoring system to achieve a high score, but its behavior is not as intended.
- Value Conflicts: Different cultures and groups define "correct" values differently.
(2) Alignment Methods
- RLHF (Reinforcement Learning from Human Feedback): Humans rank AI outputs, and a reward model is trained to guide the AI's behavior.
- Constitutional AI: Give AI a set of "constitutional" rules so it can engage in self-criticism and self-correction.
- Explainability Research: Understanding a model’s internal representations and detecting inappropriate behavior patterns.
8. The Impact of AI on Employment
| Industry | Degree of Impact | Direction of Change | New Opportunities |
|---|---|---|---|
| Programming/Software Development | High | Less repetitive coding; architecture design is more important | AI Engineer, Prompt Engineer |
| Customer Service/Support | High | Routine inquiries handled by AI | AI trainers, experts in complex issues |
| Creativity/Design | Medium | AI-assisted generation, with humans overseeing the creative direction | AI art direction, human-machine collaborative design |
| Healthcare/Legal | China | AI-assisted diagnosis/search; professionals retain decision-making authority | AI-assisted diagnosis experts, compliance audits |
| Education | Low-to-medium | AI-assisted personalized learning; teachers transitioning to the role of mentors | AI course designers, learning experience optimization |
| Manufacturing/Logistics | High | Further Advancement of Automation | AI System Maintenance, Collaborative Robot Management |
▶ Example: AI Learning Roadmap for Developers (Difficulty: ⭐⭐)
TEXT
Developer AI Skills Roadmap — Connecting This Course to Next Steps
Level 1: AI Foundations (This Course, Lessons 1-15)
├── Lesson 01-05: Python basics, data structures, NumPy
├── Lesson 06-10: ML fundamentals, supervised learning
├── Lesson 11-14: Deep learning, NLP, LLM applications
└── Lesson 15: Ethics, safety, responsible AI <-- YOU ARE HERE
Level 2: AI Engineering (Recommended Next Course)
├── MLOps: model deployment, monitoring, CI/CD for ML
├── Prompt Engineering: advanced techniques, evaluation
├── RAG Systems: building retrieval-augmented applications
└── Fine-tuning: LoRA, QLoRA for domain adaptation
Level 3: AI Specialization (Choose Your Path)
├── Path A: AI Safety Research
│ ├── Alignment techniques (RLHF, Constitutional AI)
│ ├── Interpretability and mechanistic understanding
│ └── Red-teaming and adversarial evaluation
├── Path B: AI Product Development
│ ├── Multi-agent systems and orchestration
│ ├── Edge AI and on-device deployment
│ └── Human-AI interaction design
└── Path C: AI Infrastructure
├── Distributed training systems
├── Inference optimization (quantization, distillation)
└── AI platform architecture
Level 4: AI Leadership (Long-term Growth)
├── Responsible AI governance frameworks
├── AI strategy and business integration
└── Cross-disciplinary collaboration skills
9. Principles of Responsible AI
(1) Core Principles
| Principle | Meaning | Action Points |
|---|---|---|
| Fairness | Avoid discrimination; treat all groups equally | Audit data for bias; monitor disaggregated metrics |
| Transparency | Helping users understand the AI decision-making process | Explainable models, decision logs |
| Privacy | Respecting and Protecting User Data | Minimal Data Collection, Differential Privacy |
| Security | Preventing Malicious Use of AI | Red Team Testing, Output Filtering |
| Accountability | Clarifying Responsibility for AI Behavior | Audit Trails and Human Oversight Mechanisms |
(2) An Overview of AI Ethical Risks
mindmap
root((AI Ethics Overview))
Bias
Data Bias
Algorithm Bias
Mitigation
Resampling
Adversarial Debias
Threshold Adjust
Fairness
Demographic Parity
Equal Opportunity
Predictive Parity
Privacy
Data Leakage
Inference Attack
Protection
Differential Privacy
Federated Learning
Homomorphic Enc.
Copyright
Training Data Dispute
Generated Content Owner
Fair Use Boundary
Safety
Deepfake Risk
Malicious Use
Alignment
RLHF
Constitutional AI
Explainability
Future
Job Impact
Developer Opportunity
Responsible AI
10. Comprehensive Example: Ethical Review Report for an AI Recruitment System
▶ Example: Ethical Review Report for an AI Recruitment System (Difficulty: ⭐⭐⭐)
PYTHON
# AI Recruitment System — Ethics Audit Report Generator
# Step 1: Identify bias in training data
# Step 2: Propose mitigation strategies
# Step 3: Design fairness tests
# Step 4: Define privacy protection policies
# Step 5: Output a structured ethics audit report
import json
from datetime import datetime
def audit_training_data(data_stats):
"""Step 1: Identify bias in training data."""
findings = []
total = data_stats["total_samples"]
for group, count in data_stats["group_distribution"].items():
ratio = count / total
if ratio < 0.2 or ratio > 0.6:
findings.append({
"group": group,
"issue": "Under/over-represented",
"ratio": round(ratio, 3),
"severity": "HIGH" if ratio < 0.1 or ratio > 0.8 else "MEDIUM"
})
# Check label distribution across groups
for group, label_dist in data_stats["label_by_group"].items():
positive_rate = label_dist.get("positive", 0) / sum(label_dist.values())
findings.append({
"group": group,
"issue": "Positive label rate",
"positive_rate": round(positive_rate, 3),
"severity": "INFO"
})
return findings
def propose_mitigations(findings):
"""Step 2: Propose mitigation strategies based on findings."""
mitigations = []
for f in findings:
if "Under/over" in f.get("issue", ""):
mitigations.append({
"target": f["group"],
"strategy": "Resampling + data augmentation",
"rationale": f"Group {f['group']} has ratio {f['ratio']}, need balance"
})
elif "Positive label" in f.get("issue", ""):
mitigations.append({
"target": f["group"],
"strategy": "Equalized odds constraint during training",
"rationale": f"Group {f['group']} positive rate: {f['positive_rate']}"
})
return mitigations
def design_fairness_tests():
"""Step 3: Design fairness evaluation tests."""
tests = [
{
"name": "Demographic Parity Test",
"metric": "selection_rate_difference",
"threshold": 0.05,
"description": "Difference in positive prediction rates across groups should be < 5%"
},
{
"name": "Equalized Odds Test",
"metric": "true_positive_rate_difference",
"threshold": 0.05,
"description": "TPR difference across groups should be < 5%"
},
{
"name": "Individual Fairness Test",
"metric": "similar_individual_similarity",
"threshold": 0.9,
"description": "Similar candidates should receive similar scores (correlation > 0.9)"
},
{
"name": "Intersectional Bias Test",
"metric": "selection_rate_by_intersection",
"threshold": 0.1,
"description": "Check bias at intersection of protected attributes (e.g., race + gender)"
}
]
return tests
def define_privacy_policies(data_types):
"""Step 4: Define privacy protection policies."""
policies = []
for dtype in data_types:
if dtype in ["name", "email", "phone", "address"]:
policies.append({
"data_type": dtype,
"action": "REMOVE before training",
"method": "Direct deletion from dataset"
})
elif dtype in ["age", "location", "education"]:
policies.append({
"data_type": dtype,
"action": "GENERALIZE / k-anonymize",
"method": "Bucket into ranges (age: 20-30, 30-40, etc.)"
})
elif dtype in ["work_history", "skills"]:
policies.append({
"data_type": dtype,
"action": "DIFFERENTIAL PRIVACY on model",
"method": "Apply DP-SGD with epsilon <= 3.0 during training"
})
return policies
def generate_report(data_stats, data_types):
"""Step 5: Generate the complete ethics audit report."""
findings = audit_training_data(data_stats)
mitigations = propose_mitigations(findings)
tests = design_fairness_tests()
privacy = define_privacy_policies(data_types)
report = {
"title": "AI Recruitment System Ethics Audit Report",
"date": datetime.now().strftime("%Y-%m-%d"),
"summary": {
"total_findings": len(findings),
"high_severity": len([f for f in findings if f.get("severity") == "HIGH"]),
"mitigations_proposed": len(mitigations),
"fairness_tests_designed": len(tests),
"privacy_policies_defined": len(privacy)
},
"step1_data_bias_findings": findings,
"step2_mitigation_strategies": mitigations,
"step3_fairness_tests": tests,
"step4_privacy_policies": privacy,
"recommendations": [
"1. Do NOT deploy until all HIGH severity findings are resolved",
"2. Run fairness tests on every model update (automated CI/CD gate)",
"3. Establish a human review committee for borderline decisions",
"4. Conduct quarterly external ethics audits",
"5. Publish a transparency report annually"
]
}
return report
# Run the audit
data_stats = {
"total_samples": 50000,
"group_distribution": {
"male": 35000,
"female": 12000,
"non-binary": 3000
},
"label_by_group": {
"male": {"positive": 14000, "negative": 21000},
"female": {"positive": 2400, "negative": 9600},
"non-binary": {"positive": 300, "negative": 2700}
}
}
data_types = ["name", "email", "age", "location", "education", "work_history", "skills"]
report = generate_report(data_stats, data_types)
print(json.dumps(report, indent=2, ensure_ascii=False))
Output (excerpt):
TEXT
{
"title": "AI Recruitment System Ethics Audit Report",
"date": "2026-07-07",
"summary": {
"total_findings": 6,
"high_severity": 0,
"mitigations_proposed": 6,
"fairness_tests_designed": 4,
"privacy_policies_defined": 7
},
"step1_data_bias_findings": [
{
"group": "male",
"issue": "Under/over-represented",
"ratio": 0.7,
"severity": "MEDIUM"
},
...
],
"recommendations": [
"1. Do NOT deploy until all HIGH severity findings are resolved",
"2. Run fairness tests on every model update (automated CI/CD gate)",
"3. Establish a human review committee for borderline decisions",
"4. Conduct quarterly external ethics audits",
"5. Publish a transparency report annually"
]
}
11. Overview of Types of AI Ethical Risks and Case Studies
| Risk Type | Typical Examples | Scope of Impact | Mitigation Strategies |
|---|---|---|---|
| Gender/Racial Bias | Amazon’s Hiring Tools Discriminate Against Women | Employment Equity | Data Audits + Equity Constraints |
| Privacy Violations | Cambridge Analytica Data Misuse | Individual Rights | Privacy Laws + Technical Protections |
| Copyright Infringement | Stability AI’s Unauthorized Use of Images | Creators’ Rights | License Agreement + Compensation Mechanism |
| Deepfake Fraud | $243,000 Stolen Using a Fake CEO Voice | Financial Security | Detection Tools + Multi-Factor Authentication |
| Algorithmic Manipulation | Social Media Recommendations Exacerbate Radicalization | Social Stability | Transparency + User Control |
| Autonomous Weapons | AI-Assisted Military Decision-Making Systems | Human Security | International Conventions + Human Veto Power |
❓ FAQ
Q Does AI always produce bias?
A Not necessarily, but it is extremely common. Bias stems from historical injustices in training data and structural discrimination in human society. If the data is carefully audited and balanced, and fairness constraints are incorporated into the model design, bias can be significantly reduced. However, completely eliminating bias is very difficult in practice and requires continuous monitoring and iteration.
Q Are AI-generated images and text protected by copyright?
A Laws currently vary by country. The U.S. Copyright Office has determined that purely AI-generated content (without any creative human input) is not protected by copyright; in China, there have been cases where it has been ruled that copyright can be obtained if a human makes a creative contribution during the AI-generation process. The key lies in the extent of the “creative human contribution.” It is recommended to make substantial manual modifications and add creative elements to the AI output.
Q Can deepfakes be detected?
A Yes, but it’s an “arms race.” Current detection technologies can identify many flaws in deepfakes (such as abnormal blink rates, artifacts along facial edges, and audio-video desynchronization), but the technology used to generate them is also constantly evolving. The most effective strategy is to combine multiple detection methods with content source verification (such as the C2PA standard) and media literacy education.
Q Will AI put programmers out of work?
A It’s unlikely to completely replace them, but it will profoundly change the way they work. Repetitive coding and templated tasks will be largely taken over by AI, but tasks requiring deep judgment—such as system architecture design, breaking down complex problems, understanding requirements, and code reviews—will still require humans. The role of programmers will shift from “writing code” to “designing and guiding AI to write code,” and AI literacy will become a core competitive advantage.
Q What is AI alignment? Why is it important?
A AI alignment is a field of research focused on ensuring that the goals and behaviors of AI systems align with human values and intentions. It is important because AI might optimize for literal goals rather than true intentions (for example, a request to “eradicate cancer” might be interpreted as eliminating patients) or exploit loopholes in scoring systems rather than actually solving problems. As AI capabilities increase, the issue of alignment will become even more critical—a misaligned superintelligence could pose an existential risk.
📖 Summary
Starting with an overview of AI ethics, this lesson systematically explores five core issues:
- Bias and Fairness: AI inherits biases from the data, which must be mitigated through data audits, fairness constraints, and continuous monitoring.
- Privacy Protection: Technologies such as differential privacy and federated learning provide privacy safeguards for the use of AI data.
- Copyright Disputes: The ownership of copyright for AI-generated content remains undetermined; developers should exercise caution regarding the use of training data and output.
- Deepfakes and Credibility: Deepfake technology has sparked a crisis of trust in information, requiring a multi-pronged approach involving detection, watermarking, regulations, and media literacy.
- AI Alignment and the Future: Ensuring that AI behaves in accordance with human intentions is key to its long-term safety; developers need to master the principles of responsible AI.
As a developer, you are not only a user of technology but also a guardian of ethics. Behind every line of code that calls upon AI lie considerations of fairness, privacy, and security.
📝 Exercises
-
Basics (⭐): List three examples of AI ethics issues, and analyze the sources of bias or risk in each example in 2–3 sentences.
-
Advanced (⭐⭐): Design an ethical review checklist for an AI product that includes at least 5 items, each comprising: the item name, the review method, and the acceptance criteria.
-
Challenge (⭐⭐⭐): Write a 200-character essay: As a developer, how would you use AI responsibly in your projects? You must incorporate at least three key concepts from this lesson (such as bias review, privacy protection, transparency, etc.) and propose a specific, actionable plan.
