IMPLICATIONS OF GENERATIVE AI (7 marks short answer, 6 marks multiple choice) Flashcards
Hallucination
inaccuracy in training output; AI can make stuff up that’s not entirely true
Opacity
Gen AI is not transparent; fails in circumstances difficult to predict in advance
Alignment
“Trying to make sure the behaviour of AI systems matches what we want and what we expect” Snowswell, 2023
Top-Down Alignment
designers explicitly specify values/ethical principles for AI
Bottom-Up Alignment
reverse-engineer human values from data & build AI systems aligned w/those values
Gen AI ⇒ Employment
Positives:
- Add more jobs than it takes away
- Increase productivity and income
- Increase quality and flexibility
Gen AI ⇒ Employment
Negatives:
- Reduce jobs
- Reduced mobility in workers w/out post-secondary education
- Struggles w/reemployment for older workers
- Amplify income and wealth inequality at individual/macro levels
- Threaten company survival
- Lower quality of work DUE TO HALLUCINATIONS
Gen AI ⇒ Education
Positives:
- Increase access to quality education/support democratization of knowledge
- Offers personalized, higher quality tutoring for every student (2 sigma improvement)
- Can capably respond to prompts across myriad of areas with 3Cs
- Offers teach assistant and teaching resources for every teacher
Gen AI ⇒ Education
Negatives:
- Increase educational divide
(Billions ppl without access to info/communication technology) - Decrease in teaching quality and learning DUE TO HALLUCINATIONS
- Increase in academic misconduct
- Decrease in deep learning and competency development
- Decrease in authentic research and authorship
Gen AI ⇒ Climate Change
Positives:
Accelerate solutions
- reduce 5-10% of global HGH emissions by 2030
- forecast energy production to reduce energy consumption
- Help transform energy efficiency in other carbon intensive industries
- Companies need to power AI cheaply, reliably and in environmentally friendly way
Gen AI ⇒ Climate Change
Negatives:
- Accelerate problems
- Significant energy required to train data and run models on live data
- Gen AI search uses 10x electricity as google search
- Water, biodiversity, carbon emissions concerns
Accuracy of Information
- Increases the potential for proliferation of misinformation and disinformation
- We believe what is false
- We don’t believe what is true
Representation and Bias in Information - how well gen ai represents bias
- Helps make data driven decision that reduce human bias
- Help identify patterns in inequality
- Help increase diversity in leadership if education access is increased
- Reinforce and perpetuate social bias and inequities
Information Transparency and Consent
- Raises ethical issues regarding consent, copyright and intellectual property
- Creatives may lose control over rights and compensation
- Raises questions regarding control over personal data and privacy
Cyber Security
- Increases ability to detect fraud, defend attacks
- Increase opportunities for selling IP services
- Increase cybersecurity and intellectual property risks
Early examples of Gen AI governance 1
EU AI Act (Passed in December of 2023 and implemented in 2025)
- AI systems must be safe, transparent, traceable, non-discriminatory and environmentally friendly
- AI systems should be overseen by ppl, rather than automation, to prevent harmful outcomes
- Includes different rules for different risk levels
Early examples of Gen AI governance 2
Canadian Voluntary Code of Conduct (introduced in October of 2023
- AI systems are accountable, safe, fair, equitable, transparent, monitored by humans, valid and robust
- Aligned with international Bletchley Declaration
- Part of larger Pan-Canadian AI strategy and Bill C-27 (Digital Charters Implementation Act)
