Gary Marcus, a renowned scientist and AI author, dives deep into the heated debate over a proposed six-month pause on advanced AI development. He discusses the implications of an indefinite moratorium on AGI, as advocated by Eliezer Yudkowsky, highlighting the urgency of addressing current ethical dilemmas rather than focusing solely on speculative risks. The conversation covers critiques of long-termism in AI, the complexities of managing its rapid advancements, and the pressing need for regulations that prioritize human interests amidst ongoing debates about safety and governance.
26:57
forum Ask episode
web_stories AI Snips
view_agenda Chapters
auto_awesome Transcript
info_circle Episode notes
insights INSIGHT
Intelligence Without Understanding
Eliezer Yudkowsky wasn't skeptical that evolutionary computation would eventually work.
He believed throwing enough computing power at it, combined with gradient descent and other factors, could lead to intelligence.
volunteer_activism ADVICE
Managing AI Risks
Consider the global competition and unstoppable progress in AI when thinking about moratoriums.
Focus on managing AI risks and developing safety measures instead of halting progress.
volunteer_activism ADVICE
Open Sourcing AI: A Catastrophe?
Avoid open-sourcing powerful AI models that are difficult to control or align.
Focus on building things you understand and can manage responsibly to avoid catastrophic outcomes.
Get the Snipd Podcast app to discover more snips from this episode
Support us! https://www.patreon.com/mlst
MLST Discord: https://discord.gg/aNPkGUQtc5
Send us a voice message which you want us to publish: https://podcasters.spotify.com/pod/show/machinelearningstreettalk/message
In a recent open letter, over 1500 individuals called for a six-month pause on the development of advanced AI systems, expressing concerns over the potential risks AI poses to society and humanity. However, there are issues with this approach, including global competition, unstoppable progress, potential benefits, and the need to manage risks instead of avoiding them.
Decision theorist Eliezer Yudkowsky took it a step further in a Time magazine article, calling for an indefinite and worldwide moratorium on Artificial General Intelligence (AGI) development, warning of potential catastrophe if AGI exceeds human intelligence. Yudkowsky urged for an immediate halt to all large AI training runs and the shutdown of major GPU clusters, calling for international cooperation to enforce these measures.
However, several counterarguments question the validity of Yudkowsky's concerns:
1. Hard limits on AGI
2. Dismissing AI extinction risk
3. Collective action problem
4. Misplaced focus on AI threats
While the potential risks of AGI cannot be ignored, it is essential to consider various arguments and potential solutions before making drastic decisions. As AI continues to advance, it is crucial for researchers, policymakers, and society as a whole to engage in open and honest discussions about the potential consequences and the best path forward. With a balanced approach to AGI development, we may be able to harness its power for the betterment of humanity while mitigating its risks.
Eliezer Yudkowsky: https://en.wikipedia.org/wiki/Eliezer_Yudkowsky
Connor Leahy: https://twitter.com/NPCollapse (we will release that interview soon)
Gary Marcus: http://garymarcus.com/index.html
Tim Scarfe is the innovation CTO of XRAI Glass: https://xrai.glass/
Gary clip filmed at AIUK https://ai-uk.turing.ac.uk/programme/ and our appreciation to them for giving us a press pass. Check out their conference next year!
WIRED clip from Gary came from here: https://www.youtube.com/watch?v=Puo3VkPkNZ4
Refs:
Statement from the listed authors of Stochastic Parrots on the “AI pause” letterTimnit Gebru, Emily M. Bender, Angelina McMillan-Major, Margaret Mitchell
https://www.dair-institute.org/blog/letter-statement-March2023
Eliezer Yudkowsky on Lex: https://www.youtube.com/watch?v=AaTRHFaaPG8
Pause Giant AI Experiments: An Open Letter
https://futureoflife.org/open-letter/pause-giant-ai-experiments/
Pausing AI Developments Isn't Enough. We Need to Shut it All Down (Eliezer Yudkowsky)
https://time.com/6266923/ai-eliezer-yudkowsky-open-letter-not-enough/
Machine Learning Street Talk (MLST) 