The Digital Sociology of
Generative AI
Five Speculative Propositions
What is generative artificial intelligence?
- Generative artificial intelligence is a broad category of machine learning
systems coming to be embedded in consumer-facing and business to
business products which generate one or more forms of cultural content.
- Chat-GPT by OpenAI built on the GPT-3.5 has received most attention but
this text-to-text generator one of a range of developments, including text-to-
image, text-to-video, text-to-voice and automatic generation of computer
code.
- Object of significant hype with ‘generative AI’ replacing ‘the metaverse’ and
crypto as the next big thing for the tech sector. Important to recognise hype
as a material factor rather than discursive froth which organises the diffusion
and institutionalisation of a technology e.g. from ‘big data’ overturning
everything through to social data science and computational social science.
- Current hype sparked by well planned comms strategy by Open AI which
turned Chat-GPT in a viral sensation, filling social media with screenshots of
conversations with the eerily informative chatbot.
Proposition 1: diffusion will be extremely fast and messy
- Huge reduction of staff underway in tech (40,474 tech jobs cut from 151 firm in Jan) and
plummeting stock prices (e.g. Amazon: 51%, Tesla: 68%, Meta: 66%) in the last year.
- Expected post-pandemic age of big tech hasn’t materialised with firms now confronting
high interest rate environment with declining consumer spending: their promises of future
growth becomes less attractive when returns on other classes of investments are growing.
- This is the context in which ‘generative AI’ is seized upon as a vehicle for exciting markets.
For example OpenAI is seemingly operating with a valuation of $29 billion follow ChatGPT’s
success and BuzzFeed’s stock price jumped from $1 to $4 when it announced ambitious
plans to incorporate generative AI into its publishing.
- Other direction: Alphabet lost $100 billion in market value after a promotional video for its
Bard competitor to Chat-GPT was found to contain inaccurate information.
- This environment creates a rush towards diffusion e.g. rapid incorporation into Office 365
and Bing, launch of Microsoft Designer, incorporating Bard into Google. But this also means
we’ll see a return to move fast and break things with ethical oversight constrained and
staffing reduced.
Proposition 2: it will intensify concentration of tech firms
- The arms race for ever bigger models (described by ex-Google AI ethics lead
Timnit Gebru as a corporate pissing contest) intensifies resource demands.
- GPT-3 was trained on one of the top 5 supercomputers in the world. With
each iteration of large language models the need for data, energy and money
grow.
- If this is a development trajectory in which each iteration depends on the one
which went before, it will generate an isomorphism amongst tech giants in
order to remain competitive in uncertain conditions.
- Their products and investments will resemble each other while other firms will
be left behind. This will be a sector dominated by the big three cloud
providers (AWS, Azure, Google Cloud) which constitute an ecosystem for
developers and consultants who apply their generative AI tools in specialised
conditions.
Proposition 3: its implications for work will be variegated
- Expectations of work futures (including within education) need to be grounded
in analysis of current organisation of work.
- Lazega distinguishes between collegial (collaborative relations between
formally equal peers undertaking non-routine work) and bureaucratic
(hierarchical relations structured around routinised work) modes of
organisation.
- For collegial pockets generative AI will be creatively enriching, offering new
opportunity for human/machine hybrid creation eg eliminating preparatory
tasks, new modes of creativity in AI prompting, surfing cultural abundance
- For bureaucratic pockets generative AI will intensify rationalisation for those
who remain employed (expecting them to produce more in same amount of
time) and will imperil the jobs of producers of routine content.
Proposition 4: the nature of description will change
- Incorporation of generative AI into search engines involved a shift from providing links
to websites which might answer questions to directly providing answers (see also
Google snippets)
- BuzzFeed CEO described this a shift from curated timelines to created content: it’s a
radical deepening of algorithmic mediation
- The nature of description already contested in ‘post-truth’ politics: representations of
the world seen as partial fronts for sectional interests. Widespread suspicion of ‘facts’.
- Generative AI radically lowers the costs involved in reproducing an established
domain of fact by enabling the immediate production of plausible textual descriptions
related to it.
- It simultaneously obscures the role of cultural producers in factual production so
they become part of the infrastructure for the AI rather than active cultural agents.
- This will increase the quantity of factual descriptions in social circulation while
detaching them from identifiable agents who can be objects of critique.
Proposition 5: epistemic chaos of platform capitalism will grow
- Information will be mediated/arbitrated by these systems in the near future. How will
this be received? ‘Algorithmic awe’ (Cave) on the one hand and deep suspicion on
the other
- Existing tendency towards paranoia (rather than criticality) in platform capitalism
likely to grow: a vague sense of having our strings pulled without being able to
specify who, what, when, why etc
- Interplay between credulity and paranoia intensified by propensity of GPT towards
‘hallucination’ or simply being wrong because it is trained on the ‘social average’
(Horning) - these systems often wrong but hard to fact check or know when they are
wrong
- Accelerated production of social media content facilitated by generative AI
(cost:output ratio) including deliberate misinformation saturating platforms which have
cut back their trust/security teams as advertising revenues collapse
What does this mean for education?
- Our students will be working in environments saturated by generative AI and
we need to prepare them for this: what are better or worse uses of generative
AI? What critical literacy is needed to guard against epistemic risks?
- There are emerging skills which will be significant to them: creating prompts
for AI systems, curating/wrangling AI content, critical AI literacy instructors
- The reliable reiteration of established descriptive facts will decline as markers
of knowledge and expertise. New faculties like evaluation of novelty,
improvisation, creative expression and aesthetic particularism will become
more important.
- To what extent is learning and assessment supporting the development of
these skills? How do make institutional space for assessment reform rather
than arms race between generative AI and detection software?
- How do we approach the vast ethical/political issues posed by generative AI
as platform capitalism i.e. prompts are raw material on which the system
learnings, system outputs likely to be fuel for future learning.

The Digital Sociology of Generative AI (1).pptx

  • 1.
    The Digital Sociologyof Generative AI Five Speculative Propositions
  • 2.
    What is generativeartificial intelligence? - Generative artificial intelligence is a broad category of machine learning systems coming to be embedded in consumer-facing and business to business products which generate one or more forms of cultural content. - Chat-GPT by OpenAI built on the GPT-3.5 has received most attention but this text-to-text generator one of a range of developments, including text-to- image, text-to-video, text-to-voice and automatic generation of computer code. - Object of significant hype with ‘generative AI’ replacing ‘the metaverse’ and crypto as the next big thing for the tech sector. Important to recognise hype as a material factor rather than discursive froth which organises the diffusion and institutionalisation of a technology e.g. from ‘big data’ overturning everything through to social data science and computational social science. - Current hype sparked by well planned comms strategy by Open AI which turned Chat-GPT in a viral sensation, filling social media with screenshots of conversations with the eerily informative chatbot.
  • 4.
    Proposition 1: diffusionwill be extremely fast and messy - Huge reduction of staff underway in tech (40,474 tech jobs cut from 151 firm in Jan) and plummeting stock prices (e.g. Amazon: 51%, Tesla: 68%, Meta: 66%) in the last year. - Expected post-pandemic age of big tech hasn’t materialised with firms now confronting high interest rate environment with declining consumer spending: their promises of future growth becomes less attractive when returns on other classes of investments are growing. - This is the context in which ‘generative AI’ is seized upon as a vehicle for exciting markets. For example OpenAI is seemingly operating with a valuation of $29 billion follow ChatGPT’s success and BuzzFeed’s stock price jumped from $1 to $4 when it announced ambitious plans to incorporate generative AI into its publishing. - Other direction: Alphabet lost $100 billion in market value after a promotional video for its Bard competitor to Chat-GPT was found to contain inaccurate information. - This environment creates a rush towards diffusion e.g. rapid incorporation into Office 365 and Bing, launch of Microsoft Designer, incorporating Bard into Google. But this also means we’ll see a return to move fast and break things with ethical oversight constrained and staffing reduced.
  • 5.
    Proposition 2: itwill intensify concentration of tech firms - The arms race for ever bigger models (described by ex-Google AI ethics lead Timnit Gebru as a corporate pissing contest) intensifies resource demands. - GPT-3 was trained on one of the top 5 supercomputers in the world. With each iteration of large language models the need for data, energy and money grow. - If this is a development trajectory in which each iteration depends on the one which went before, it will generate an isomorphism amongst tech giants in order to remain competitive in uncertain conditions. - Their products and investments will resemble each other while other firms will be left behind. This will be a sector dominated by the big three cloud providers (AWS, Azure, Google Cloud) which constitute an ecosystem for developers and consultants who apply their generative AI tools in specialised conditions.
  • 6.
    Proposition 3: itsimplications for work will be variegated - Expectations of work futures (including within education) need to be grounded in analysis of current organisation of work. - Lazega distinguishes between collegial (collaborative relations between formally equal peers undertaking non-routine work) and bureaucratic (hierarchical relations structured around routinised work) modes of organisation. - For collegial pockets generative AI will be creatively enriching, offering new opportunity for human/machine hybrid creation eg eliminating preparatory tasks, new modes of creativity in AI prompting, surfing cultural abundance - For bureaucratic pockets generative AI will intensify rationalisation for those who remain employed (expecting them to produce more in same amount of time) and will imperil the jobs of producers of routine content.
  • 7.
    Proposition 4: thenature of description will change - Incorporation of generative AI into search engines involved a shift from providing links to websites which might answer questions to directly providing answers (see also Google snippets) - BuzzFeed CEO described this a shift from curated timelines to created content: it’s a radical deepening of algorithmic mediation - The nature of description already contested in ‘post-truth’ politics: representations of the world seen as partial fronts for sectional interests. Widespread suspicion of ‘facts’. - Generative AI radically lowers the costs involved in reproducing an established domain of fact by enabling the immediate production of plausible textual descriptions related to it. - It simultaneously obscures the role of cultural producers in factual production so they become part of the infrastructure for the AI rather than active cultural agents. - This will increase the quantity of factual descriptions in social circulation while detaching them from identifiable agents who can be objects of critique.
  • 8.
    Proposition 5: epistemicchaos of platform capitalism will grow - Information will be mediated/arbitrated by these systems in the near future. How will this be received? ‘Algorithmic awe’ (Cave) on the one hand and deep suspicion on the other - Existing tendency towards paranoia (rather than criticality) in platform capitalism likely to grow: a vague sense of having our strings pulled without being able to specify who, what, when, why etc - Interplay between credulity and paranoia intensified by propensity of GPT towards ‘hallucination’ or simply being wrong because it is trained on the ‘social average’ (Horning) - these systems often wrong but hard to fact check or know when they are wrong - Accelerated production of social media content facilitated by generative AI (cost:output ratio) including deliberate misinformation saturating platforms which have cut back their trust/security teams as advertising revenues collapse
  • 9.
    What does thismean for education? - Our students will be working in environments saturated by generative AI and we need to prepare them for this: what are better or worse uses of generative AI? What critical literacy is needed to guard against epistemic risks? - There are emerging skills which will be significant to them: creating prompts for AI systems, curating/wrangling AI content, critical AI literacy instructors - The reliable reiteration of established descriptive facts will decline as markers of knowledge and expertise. New faculties like evaluation of novelty, improvisation, creative expression and aesthetic particularism will become more important. - To what extent is learning and assessment supporting the development of these skills? How do make institutional space for assessment reform rather than arms race between generative AI and detection software? - How do we approach the vast ethical/political issues posed by generative AI as platform capitalism i.e. prompts are raw material on which the system learnings, system outputs likely to be fuel for future learning.