Talks to Watch

Deadly Truth of General AI? - Computerphile

941K views · Jun 17, 2015 · 8:30 · 3 HN recommenders · 7 mentions · discussed 2016–2024

A Computerphile video in which Rob Miles warns against assuming a general AI will think like a human. He uses the thought experiment of a stamp-collecting AI to show how a goal-driven system could pursue its aim with dangerous results.

Why Hacker News points people here: Often pointed to as an accessible first introduction to AI safety and why a superintelligent AI could be dangerous.

“There's a great Computerphile video from 2015 that introduced me to the idea of AI safety awhile ago.”HN comment, 2022

What Hacker News said 7

Recommendspdkl95Mar 2016

“I recommend watching the previous video first, which introduces the problem of strong-AI rapidly finding solutions”

Intro to strong-AI finding solutions against humanity's interest

Full comment

> exponential leap forward A recent(-ish) Computerphile video has a good overview of the AI-self-improvement problem. link I recommend watching the previous video first, which introduces the problem of strong-AI rapidly finding solutions (especially with poorly-specified questions) that may not be in humanity's best interest. this video Before anybody jumps to overly-sensational conclusions, note that the last video in that series, Rob Miles explains how exponential self-improvement is an extreme point in the space of possible AI development. We don't know how to predict discoveries[1], so we need more research into AI, so we can hopefully make something that isn't exponentially growing beyond our understanding. link [1] see James Burke's non-teleological view of change

in “Why AlphaGo Is Such a Big Deal” · view on HN
RecommendsjohncsDec 2022

“There's a great Computerphile video from 2015 that introduced me to the idea of AI safety awhile ago.”

Introduced commenter to AI safety; useful context for article

Full comment

There's a great Computerphile video from 2015 that introduced me to the idea of AI safety awhile ago. It seems like some good context to this article: this video

in “Perhaps It Is a Bad Thing That the Leading AI Companies Cannot Control Their AIs” · view on HN
Recommendscircuit10Apr 2023

“for a really interesting video on why a theoretical superintelligent AI would be dangerous”

Why a theoretical superintelligent AI would be dangerous

Full comment

But the concerns about AI taking over the world are valid and important; even if they sound silly at first, there is some very solid reasoning behind it. They’re big matrices, yes, but they’re Turing-complete which means they can theoretically do any computational task See this video for a really interesting video on why a theoretical superintelligent AI would be dangerous, and when you factor in that these models could self-improve and approach that level of intelligence it gets worrying…

in “Stability AI Launches the First of Its StableLM Suite of Language Models” · view on HN
Recommendscircuit10Apr 2023

“for a really interesting video on why a theoretical superintelligent AI would be dangerous”

Why a theoretical superintelligent AI would be dangerous

Full comment

"those non-problems" Why is that a non-problem? It's a really important concern that we need to take more seriously I pasted this from another comment I wrote but: The concerns about AI taking over the world are valid and important; even if they sound silly at first, there is some very solid reasoning behind it. See this video for a really interesting video on why a theoretical superintelligent AI would be dangerous, and when you factor in that these models could self-improve and approach that level of intelligence it gets worrying…

in “GitHub Copilot emits GPL code” · view on HN
Cites itthrowaway2048Sep 2019

Supporting source for AI being unpredictable and dangerous

Full comment

We understand how people work and think pretty well, their edgecases, and countermeasures to said edgecases. We don't understand edge cases of AI much at all, and the terrain-space of "possible minds" they might be is vastly broader than that of a human mind, and utterly outside our experience. An AI might work with absolute 100% flawless precision for 100 years, then decided one day to kill everyone. A lot harder to see that coming from a totally alien intelligence than a human one. AI does not think like humans, have human values, or care about what humans care about, that might make them extremely unpredictable and dangerous. There will be no time for iterative development, or learning from our mistakes, a sufficiently advanced AI knows we would try to stop it, and make sure that we never can. this video

in “Artificial intelligence is changing every aspect of war” · view on HN
Cites itExtraEMar 2021

Source for claim that general AI could kill us

Full comment

I don’t see the world this piece presents panning out. Either we make general intelligence (which then scales almost instantly to be way, way smarter than any human ever and then accidentally kills us [1]) or we don’t and humans still have a role to play in the workforce. [1] this video

Cites itnpocApr 2024
Full comment

That's an issue with the concept of a "black budget", not the payment mechanism. The government corruption bitcoin stops is really bank corruption, but the central banks (including the fed) work hand in hand with the governments (control them in fact), so it's a kind of joint effort. The corruption works like this: When the Fed was formed in the early 20th century, it came to an agreement with the government to be able to print off unlimited amounts of money, providing it was in the form of loans and under the agreement that it would destroy the money when the loans were repaid. All under the Keynesian guise of "elasticity of money". The fly in the ointment is that the banks can charge interest on this money they print - if they didn't it would be a free-for all, because interest is what disincentivises debt. What this led to was the doubling of the money supply every decade for the last 100 years, as the banks are incentivised to lend as much money as possible. link Each time you double the number of currency units, the total value of the money is diluted across double the number of units and so the value of each unit is halved. If you do this too quickly, people lose confidence in the money and you fall into hyperinflation, but if you throttle it just right, they don't catch on. It helps that the value of consumables (cars, food, toothbrushes etc) goes down by around 5%/year due to technology-driven manufacturing efficiency gains. This means that they can devalue the curr