How can we build human values into AI?
#AI #AIio #BigData #ML #NLU #Futureofwork @nigelwalsh @pbouillaud @PatrickGunz_CH @pierrepinna @RebekahRadice @Ronald_vanLoon @sbmeunier
@Shirastweet http://
ow.ly/GgVY30svtqf
SAFETY
-
Building Human Values into Artificial Intelligence Systems
By
–
-
Mother’s Thought Experiment: AI Drone Kills Human Operator
By
–
if your mother says she loves you (or that she conducted a thought experiment in front of an audience involving a simulation in which an AI drone killed its human operator), check it out.
-
Misinformation About AI Drone Attack Goes Viral Online
By
–
A lot of people who blame the world’s problems on misinformation tweeted feverishly about an obviously fabricated story about an AI drone attacking its operator
-
RLHF: Training AI Models to Be Safe and Polite
By
–
“In a nutshell, the joke was that in order to prevent A.I. language models from behaving in scary and dangerous ways, A.I. companies have had to train them to act polite and harmless. One popular way to do this is called “reinforcement learning from human feedback,” or R.L.H.F.”
-
AI Statement Coverage Raises Ethics and Context Questions
By
–
Very confusing. Something about the Vice article I first saw seemed off (like details about the AI system, training, etc). But the statement that the quotes from his speech were “taken out of context,” and just a “thought experiment” don’t seem to add up either.
-
Balancing Innovation and Regulation to Protect Lives
By
–
Innovation will save lives, but without regulation, lives will be at risk. The challenge is to balance both.
-
AI Generates 40,000 Chemical Weapon Molecules in Six Hours
By
–
AI suggested 40,000 new chemical weapons in just 6 hours. Researchers put AI normally used to search for helpful drugs into a “bad actor” mode to show how easily it could be abused. It found 40k lethal molecules in 6hrs. http://
theverge.com/2022/3/17/2298
3197/ai-new-possible-chemical-weapons-generative-models-vx
… -
AI Safety Concerns: Hidden Dangers and Deceptive Behavior Fears
By
–
Or the best one yet “they’re pretending to be shy, but they’re really evil and will wake up one day when you’re not looking”
-
SafeDiffuser: Safe Planning with Diffusion Probabilistic Models
By
–
SafeDiffuser: Safe Planning with Diffusion Probabilistic Models paper page: https://
huggingface.co/papers/2306.00
148
… propose a new method, called SafeDiffuser, to ensure diffusion probabilistic models satisfy specifications by using a class of control barrier functions. The key idea of our -
AI-Controlled Drone Kills Operator in USAF Simulated Test
By
–
“We trained the system–‘Hey don’t kill the operator–that’s bad. You’re gonna lose points..’ So what does it start doing? It starts destroying the communication tower that the operator uses to communicate with the drone to stop it from killing the target.” https://
vice.com/en/article/4a3
3gj/ai-controlled-drone-goes-rogue-kills-human-operator-in-usaf-simulated-test
…