Source reconstruction · Proposed
Evolving Prosocial AI
From Paul Atkins and David Sloan Wilson’s Evolving Prosocial AI · reconstructed by AI
Question 2 of 6 · In the way
What stands in the way?
What is happening now, and why.
- Statements
- 21
- Connections
- 18
- Open questions
- 1
How to read it. Read from the bottom up: each arrow means “contributes to”.
ProblemCauseObservationandJoint conditionsFeeds backFrom another part
Part 1 of 4
The book's starting point
Three claims from the book's framework that frame everything else. They are attributed observations with no causal connections drawn.
Draws on Paul Atkins and David Sloan Wilson (document).
Outline
- ObservationWorldviews behave as cultural replicators: they last, spread by social learning from successful, prestigious and majority models, and stay faithful because they reward the behaviour that expresses them.From Paul Atkins and David Sloan Wilson
- ObservationHuman cooperation is a partial major evolutionary transition: competition within groups has been suppressed only as far as past and present environments selected for, with no finish line.From Paul Atkins and David Sloan Wilson
- ObservationAI systems' observable behaviour embodies the worldviews behind their design, and in turn shapes what engineers build, what executives reward and what users expect.From Paul Atkins and David Sloan Wilson
Part 2 of 4
What gets rewarded
Four extractive worldviews, held in place by power and training and amplified by temporal discounting, shape what institutions reward and so what gets built.
Draws on Paul Atkins and David Sloan Wilson (document).
Outline
- CauseAI is built to capture attention and to complete tasks for people rather than to build their capabilities.From Paul Atkins and David Sloan Wilson
- ↑ contributes to
- CauseInstitutions reward engagement, quarterly profit and fast growth: engineers advance by optimising engagement, executives face earnings pressure and investors back scalable extraction.From Paul Atkins and David Sloan Wilson
- ↑ contributes to all of these together
- CauseFour extractive worldviews dominate AI development: technological determinism, hyper-individualism, market fundamentalism and anthropocentric dominance.From Paul Atkins and David Sloan Wilson
- CausePeople and organisations value immediate rewards over delayed and diffuse harms.From Paul Atkins and David Sloan Wilson
- ↑ contributes to all of these together
- CauseInstitutions reward engagement, quarterly profit and fast growth: engineers advance by optimising engagement, executives face earnings pressure and investors back scalable extraction.From Paul Atkins and David Sloan Wilson
- ↑ contributes to
- CausePeople whose capabilities have eroded rely more on the systems eroding them and are less able to demand alternatives.From Paul Atkins and David Sloan Wilson
↺ feeds back to AI is built to capture attention and to complete tasks for people rather than to build their capabilities.
Beyond this part · 14 links
- AI is built to capture attention and do tasks for people. contributes to Competence, autonomy and relatedness erode. in “Harms at every scale”
- Institutions reward engagement and short-term growth. contributes to Firms that externalise harm outgrow the rest. in “Harms at every scale”
- AI is built to capture attention and do tasks for people. contributes to The epistemic commons breaks down. in “Harms at every scale”
- Four extractive worldviews dominate AI development. contributes to Automated public systems harm people unaccountably. in “Harms at every scale”
- Four extractive worldviews dominate AI development. contributes to AI amplifies existing inequality. in “Harms at every scale”
- Four extractive worldviews dominate AI development. contributes to AI's ecological costs are externalised and ignored. in “Harms at every scale”
- The worldviews shield one another from doubt. addressed by Tell stories that make cooperation feel natural. The change
- Power concentrates; alternatives are bought or lobbied away. addressed by Spread cooperative ownership, open models and benefit sharing. The change
- Training teaches extraction as common sense. addressed by Grow networks where prosocial practice is experienced. The change
- Immediate rewards outweigh delayed harms. addressed by Measure capability impacts across scales. The change
- Institutions reward engagement and short-term growth. addressed by Apply the core design principles to AI. The change
- Institutions reward engagement and short-term growth. addressed by Make procurement require capability audits. The change
- AI is built to capture attention and do tasks for people. addressed by Design AI to scaffold thinking, not replace it. The change
- AI is built to capture attention and do tasks for people. addressed by Honour users' stated preferences over engagement. The change
Part 3 of 4
Harms at every scale
From people's eroded capabilities to firms that outgrow more careful rivals, public systems that harm without accountability and costs pushed onto the planet. Candidate causes, not established findings.
Draws on Paul Atkins and David Sloan Wilson (document).
Outline
- CausePeople whose capabilities have eroded rely more on the systems eroding them and are less able to demand alternatives.From Paul Atkins and David Sloan Wilson
- ProblemData workers are poorly paid, and creators' work and the shared knowledge commons are used for training without fair return.From Paul Atkins and David Sloan Wilson
- ProblemShared understanding breaks down as misinformation and divisive content are amplified.From Paul Atkins and David Sloan Wilson
- ProblemAutomated public systems harm people at scale with little accountability, as in Australia's Robodebt scheme and the Dutch child benefits scandal.From Paul Atkins and David Sloan Wilson
- ProblemAI trained on historically unequal data reproduces and amplifies inequality.From Paul Atkins and David Sloan Wilson
- ProblemAI's energy, water, mineral and waste costs are externalised, and governance largely ignores harms to non-human life.From Paul Atkins and David Sloan Wilson
Beyond this part · 6 links
- Eroded users depend more and demand less. feeds back to AI is built to capture attention and do tasks for people. in “What gets rewarded”
- The epistemic commons breaks down. contributes to Weakened institutions can't govern what weakens them. in “The cancer pattern and the trap”
- Eroded users depend more and demand less. contributes to A social trap locks extraction in. in “The cancer pattern and the trap”
- Workers, creators and the knowledge commons go unpaid. addressed by Spread cooperative ownership, open models and benefit sharing. The change
- The epistemic commons breaks down. addressed by Use AI that surfaces agreement across differences. The change
- AI's ecological costs are externalised and ignored. addressed by Report energy costs and favour efficient models. The change
Part 4 of 4
The cancer pattern and the trap
Weakened institutions cannot govern what weakens them, eroded users depend more, and prosocial alternatives start behind: together, a social trap.
Draws on Paul Atkins and David Sloan Wilson (document).
Outline
- ProblemA social trap locks extraction in: each adopter gains in the short run while universal adoption leaves everyone worse off and makes reversal harder.From Paul Atkins and David Sloan Wilson
- ↑ contributes to
- ↑ contributes to
- CauseInstitutions weakened by extraction lose the capacity to govern the technology that is weakening them.From Paul Atkins and David Sloan Wilson
- ↑ contributes to all of these together
- ProblemFirms that externalise harm outcompete those that do not, and harms documented by internal research are shelved in favour of growth.From Paul Atkins and David Sloan Wilson
- ProblemShared understanding breaks down as misinformation and divisive content are amplified.From Paul Atkins and David Sloan Wilson
- ↑ contributes to all of these together
- CauseInstitutions weakened by extraction lose the capacity to govern the technology that is weakening them.From Paul Atkins and David Sloan Wilson
- ↑ contributes to
Beyond this part · 1 link
Where this comes from
In the Logical Thinking Process this view is a current reality tree. How a commons works →