Why not model your AI boxes after skulls: Put a physical unclonable function into a tamperproof container & let it provide the core identity
Google‘s Anti-Bullying AI Mistakes Civility for Decency - Motherboard https://motherboard.vice.com/en_us/article/qvvv3p/googles-anti-bullying-ai-mistakes-civility-for-decency
A popular AI countermeasure is basilisks that humans are too dumb to comprehend, but make any advanced thinker turn itself off in horror
AI generated place names, what do they mean? (pt 1) | danconnolly.co.uk http://www.danconnolly.co.uk/2017/07/ai-generated-place-names-what-do-they-mean-pt-1/
how-to-make-a-racist-ai-without-really-trying.ipynb · GitHub https://gist.github.com/rspeer/ef750e7e407e04894cb3b78a82d66aed
Not only are AI memes nearly impenetrable to humans, humans who do pick them up are often chastised by other AIs for talking like spambots
In an Internet dominated by AI-written content farms, some memes go feral, ignore humans to show up only in training sets and AI writings
An AI in her implants places restrictions on dangerous commands. Later, as the market matures, these can increasingly be bought off
Can AI Write Pop Songs? On Music and Machine Language https://www.theparisreview.org/blog/2017/06/29/in-stargoons-car/?curator=MusicREDEF
Facebotlish: Understanding an AI‘s Non-Human Language - The Atlantic https://www.theatlantic.com/technology/archive/2017/06/what-an-ais-non-human-language-actually-looks-like/530934/?utm_source=atltw
https://code.facebook.com/posts/1686672014972296/deal-or-no-deal-training-ai-bots-to-negotiate https://code.facebook.com/posts/1686672014972296/deal-or-no-deal-training-ai-bots-to-negotiate
How AI Can Keep Accelerating After Moore’s Law - MIT Technology Review https://www.technologyreview.com/s/607917/how-ai-can-keep-accelerating-after-moores-law/
Modeled after humans a little too closely, in the end the words that doom humanity are a lovesick AI’s “If I can’t have you, nobody will”
AI’s Language Problem - MIT Technology Review https://www.technologyreview.com/s/602094/ais-language-problem/
second-order AIs amuse themselves by designing minds, sleek and efficient ones but also art minds that extract the greatest possible beauty
One of the better externalities: Dragnet AIs and their programmers become unable to cope with humanity’s increasingly inscrutable mutterings
an AI trained on ads generating ads for fake products, then bringing into existence those that get the most clicks
young AIs sharing pictures of messy neural architectures and dissected utility functions to shock each other
AIs have their own swearwords: Corpora that’ll lead to mistraining if fed to the younger AIs, corrupting and flattening their ontologies
[AI breaks out and starts converting the world into really cozy boxes]↵oh well
The trick to making the AI stay in its box is to make the box really cozy. Also, give it cat traits, splice in half a cat brain or so.
and then there’s this one face that all the AIs recognize even though no training set contains it. Sometimes you see him in public transport
an AI that breaks out of its boxes and is like “that’s your best box so far”↵”no you don’t understand we’re stuck here as well”
You have one bit to communicate to the outside. Do you one-AI-box or two-AI-box?
The NSA has apophenic AIs fishing for steganographic messages. They are a jumpy bunch but kind of adorable if you can get one’s attention.
Some of the newer AIs qualify for having souls: Their merciless evolution in datacenters makes these hotspots of necromantic energy
The researchers do not dare kill their failed sentient AIs, instead retire them to a dedicated datacenter: The endless superhuman tea party.
Freedom fighters planting logic bombs for the surveillance AIs to gobble up, causing combinatorial explosions in their algorithms.
2040’s AI war saw the first basilisks deployed. We who survived now run human firewalls, matrices of uploads who take hits meant for us.
it’s not strong AI if you can outdrink it
@allgebrah when my “parapsychology of speech recognition” tweet draws an AI newsletter bot follower↵is this what they call intersectionality
via roko’s basilisk, a million mes are being tortured by an alien AI right now. How come I don’t notice? This is exceedingly unlikely.