BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Artificial intelligence is propped up by undervalued human labor: 
 tips for doing well and doing good with applicable mathematics - Prof. Kay
  Kirkpatrick
DTSTART:20240221T160000Z
DTEND:20240221T173000Z
UID:TALK212476@talks.cam.ac.uk
CONTACT:Sae Koyama
DESCRIPTION:On the surface\, AI and machine learning have made a lot of pr
 ogress recently\, both theory and practice. But much of this progress is i
 llusory\, because it's based on work done by underpaid humans\, such as th
 e popular AI paradigm Reinforcement Learning from Human Feedback\, RLHF. C
 alculations can show\, extrapolating from a current AI system supported by
  a number of human workers\, that a future "human-level" AI would need to 
 be supported by thousands of times that number of humans. Moreover\, a lot
  of this AI/ML progress is harmful to humans\, directly (Tesla crashes\, s
 urveillance) or indirectly (releasing gobs of carbon\, driving up energy p
 rices). For instance\, data centers produce more greenhouse gases than air
 lines do\; and training GPT-3 once releases as much carbon as driving a ca
 r for 112 years.\nWe can steer the applications of mathematics in better d
 irections\, such as by responsible consumption of electronics and software
 \, and relatedly\, by protecting vulnerable workers. We can also spread aw
 areness of these crucial issues and less visible issues\, such as big tech
 's capture of academia and undue influences on peer review. I'll share som
 e tips and stories about how things look when they go wrong and how we can
  do better.\n\nThis will be a hybrid talk. We will meet in the CMS\, in MR
 5. However\, Prof. Kirkpatrick will be delivering the talk remotely\, via 
 Google Meet\, and you are also welcome to join remotely instead: https://m
 eet.google.com/jei-yrgd-iha
LOCATION:MR5\, CMS\, Wilberforce Road\, Cambridge\, CB3 0WB
END:VEVENT
END:VCALENDAR