1 of 57

✨The Librarian’s Dividend✨�Pathways towards Information Ethics and Literacy �in the Age of Generative AI

Casey Fiesler | casey.prof

Image credit: OpenAI’s DALL-E and countless uncredited and uncompensated artists whose work helped train the model

Image credit: Lone Thomasky & Bits&Bäume / https://betterimagesofai.org / CC-BY 4.0

2 of 57

2

information disorder

is the breakdown of the information ecosystem caused by the spread of misinformation, malinformation, and disinformation

3 of 57

3

bit.ly/ai-ethics-news

4 of 57

4

5 of 57

5

6 of 57

6

7 of 57

7

8 of 57

8

artificial intelligence

is when a machine does something that typically requires human thinking

9 of 57

9

10 of 57

10

CAT

DOG

11 of 57

11

Machine learning finds patterns in and makes predictions based on training data.

12 of 57

12

Machine learning finds patterns in and makes predictions based on training data.

Generative AI generates new data based on what’s in its training data.

13 of 57

13

14 of 57

14

15 of 57

15

16 of 57

16

Machine learning finds patterns in and makes predictions based on training data.

Generative AI generates new data based on what’s in its training data.

A large language model is a probability distribution over sequences of words in its training data.

17 of 57

17

18 of 57

18

19 of 57

19

20 of 57

20

21 of 57

21

22 of 57

22

Search engines search for information.

Language models generate or make up information (language).

​

… and are designed to be statistically probable and linguistically fluent, not verifiably accurate.

23 of 57

23

24 of 57

24

CAT

DOG

25 of 57

25

@cfiesler

26 of 57

26

@cfiesler

27 of 57

27

28 of 57

28

29 of 57

29

30 of 57

30

31 of 57

31

32 of 57

32

@cfiesler

33 of 57

33

34 of 57

34

35 of 57

35

36 of 57

36

the liar’s dividend

is the benefit that liars receive from existing in a world in which it is unclear what is true and what is false (and therefore one can claim anything is false)

Chesney, B., & Citron, D. (2019). Deep fakes: A looming challenge for privacy, democracy, and national security. Calif. L. Rev., 107, 1753.

37 of 57

37

@cfiesler

38 of 57

38

@cfiesler

39 of 57

39

the librarian’s dividend

is the social and ethical benefit that comes from access to people and institutions who can help us (and help teach us how to) evaluate information

40 of 57

40

@cfiesler

41 of 57

41

AI Literacy

  • Technical knowledge (how does this work and what are its limitations)
  • Practical skills (how – and whether/when – to use AI tools)
  • Critical thinking (information literacy and mindful engagement)
  • Ethical awareness (understanding the values built into AI systems)
  • Societal impact (examination of the broader role and impacts of AI

​

Leo S. Lo. “AI Literacy: A Guide for Academic Libraries.”

42 of 57

42

43 of 57

43

Taxonomy of large language model risks:

​

  • Misinformation and disinformation
  • Fraud, scams, and targeted manipulation
  • Bias: unfair discrimination, harmful stereotypes, and exclusion
  • Hate speech and offensive language
  • Privacy risks
  • Human interaction risks
  • Socioeconomic and labor harms
  • Environmental harms

​

Weidinger, Laura, et al. "Taxonomy of risks posed by language models." Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency. 2022.

44 of 57

44

Privacy

45 of 57

45

Data Ownership

46 of 57

46

47 of 57

47

Socioeconomic & Labor Harms

48 of 57

48

49 of 57

49

50 of 57

50

51 of 57

51

Environmental Harms

52 of 57

52

Environmental Harms

53 of 57

53

@cfiesler

54 of 57

54

@cfiesler

“Machines are made to behave in wondrous ways, often sufficient to dazzle even the most experienced observer. But once a particular program is unmasked, once its inner workings are explained in language sufficiently plain to induce understanding, its magic crumbles away; it stands revealed as a mere collection of procedures, each quite comprehensible.”

​

- Joseph Weizenbaum, 1966

55 of 57

55

Things everyone should know about AI to be able to think about responsible use:

  • This is not magic, and here are the basics of how it works.
  • Because of how it works, it may provide blatantly incorrect or misleading information.
  • In part because of where the training data comes from, the outputs may create or perpetuate bias.
  • It is important to be aware of the harms that currently exist and could in the future, and to be critical of the business models and systems of power that underlie this technology.

56 of 57

56

57 of 57

Image credit: OpenAI’s DALL-E and countless uncredited and uncompensated artists whose work helped train the model

@cfiesler

@professorcasey

caseyfiesler.com | casey.prof

Professor Casey

Image credit: Lone Thomasky & Bits&Bäume / https://betterimagesofai.org / CC-BY 4.0