I see a shitload of people talking past each other with regard to GenAI.
For example, I’ve seen a few people say something like “if GenAI is so bad how come it can do this?” Which is a fallacy; the word “bad” here is not referring to its efficacy, but other concerns such as ethics, morality, or environmental factors.
So I’m trying to come up with a complete, descriptive (not prescriptive) list of the potential harms, both so I can reason about them in my head separately (and decide which I want to focus on) and so I can better point out when someone is moving the goalposts.
Here’s my current list. Let me know if you have any I missed.
(This list contains nested bullet points. If it does not render well for you, consider viewing it on the web.)
Edit: I have added several points from https://whynoai.info/, which is an extremely well-argued page IMO.
Categorising the harms of GenAI
- mental
- brain rot
- deskilling
- cognitive surrender
- AI psychosis
- environmental
- building new data centres
- water consumption
- energy, especially fossil fuels
- more e-waste
- ethical
- reliance on unethical labelling and cleaning process
- benefits the rich and harms the poor
- economic
- unsustainable business models
- potential financial crash
- consent
- scraping without consent
- copyright violations
- network-level abuse
- loss of credit
- loss of revenue (advertising, or other)
- driving creators away from open publishing
- individualism
- atomisation, weakening community
- hypocrisy (my use of GenAI is fine, but everything else is slop)
- scabbing (I don’t agree with GenAI, but I’ll use it because I need to eat)
- centralisation
- single source of knowledge
- enclosing the commons
- canonicalisation of systemic bias
- language policing
- slop
- destroying and polluting knowledge bases (the internet of slop)
- decline in quality
- mediocre output by design
- overconfident output, leading to mistakes
- overcomplicated or extremely verbose solutions
- proxying and interfering with communication
- spam
- abuse of a probabilistic tool as “good enough”
- high rate of abandonment
- destroying business models by driving customers away
- abuse
- deepfakes
- non-consensual sexual material
- automated surveillance, profiling, and stalking
- rejection of humanity
- the Singularity
- effective altruists
- TESCREAL
- fascism
@samir Good list but there's always more!
- centralising compute
- pricing individuals and even business out of local compute (eg RAM CPU/GPU)
- environmental destruction (eg forests in Finland in order to build data centres)
- noise pollution from data centres and other effects which blight localities and communities
If I wasn't so tired I think I could add to this.
1/2