IAPP AIGP

Supervised learning

labeled data, used for grouping, ex. spam filters

Unsupervised learning

unlabeled data, used for pattern recognition, ex. banking data

Reinforcement learning

maximizing rewards, ex. autonomous vehicles

Semi supervised learning

mix of labeled and unlabeled data

Data lineage

data origin, how its transformed, where it moves over time. ensures AI integrity and representative/accurate data.

Components of AI definition

technology, autonomy, human involvement, output

OECD Framework

purpose: to classify AI systems and examine risks

1) People & Planet, 2) Economic Context, 3) Data & Input, 4) AI Model, 5) Tasks & Outputs

AI use cases

• recognition

• event detection

• forecasting

• personalization

• interaction support

• goal driven optimization

• recommendation

Dartmouth conference

1956 - formalized AI into academic field, where it was decided that human intelligence could be mimicked artificially with enough description, term “AI” created

1st AI summer (mid 1950s to mid 1970s)

immediately after Dartmouth conference, creation of LISP (1st AI programming language, John McCarthy) and ELIZA (1st NLP, Joseph Weizenbaum)

1st AI winter (mid 1970s to mid 1980s)

period of skepticism, funding cuts

2nd AI summer (mid 1980s - late 1980s)

rise of expert systems, increased computing capabilities, AI powered computers by Japanese gov (5th Gen Computer Systems)

2nd AI winter (late 1980s - late 1990s)

decreased interest due to cost of expert systems, end of Cold War

AI renaissance (late 1990s - 2011)

era of big data, IBM’s Deep Blue beats world chess champion, emergence of Internet

AI boom (2011 - present)

advancements in deep learning

Weak/narrow AI

perform narrow set of related tasks at high level of proficiency

Strong AI

strong generalization capabilities, closely mimic human intelligence, AGI

ASI

outperforming humans

Broad AI

capable of performing broad set of tasks, relies of group of AI systems (ex. Autonomous vehicles)

Robotics processing automation (RPA)

software robots that automate repetitive tasks

Expert systems components

knowledge base, inference engine, user interface

Implicit bias

discrimination/prejudice towards particular person or group

Sampling bias

data skews towards a subset

Temporal bias

model works now, may not work later

Group harms

mass surveillance, restriction of freedom of assembly/protest, deepening of inequities, harm to democratic process, spread of deepfakes, LAWS

Individual harms

civil liberties, rights, safety, economic opportunity

Economic risk

discriminatory hiring, loss of jobs

Organizational harms

negative brand impact, ESG ratings, share price drop, investors flight, campaigners

Pros of risk management framework

ensure requirements are met, auditability, accountability

OECD AI Principles

• Inclusive growth, sustainable development, wellbeing

• Human centered values and fairness

• Transparency and explainability

• Robustness, safety, and security

• Accountability