Hanson often takes his turn to speak this way: he summarizes Yudkowsky's last argument, in a way that at least superficially does not seem unfair or tendentious, then explains why he doesn't find it compelling, then explains why his own position is more compelling.
Yudkowsky seems to respond to Hanson's points without summarizing them first.
I find Hanson to be hugely more effective in the recording. Is it because of this? I was less sympathetic to Yudkowsky's point of view before I started listening, so it's hard for me to tell if this is an illusion.
Modifiers like "new" or "old" in titles or filenames quickly become unhelpful. It's generally better to use dates (in this case, something like "Jun 2011" would do the trick).
Can someone who agrees with Yudkowsky do an extended summary of Yudkowsky's position and arguments in text, the way Hanson summarized his position and arguments in text?
72:00 - Hanson cites the AAAI white paper which rails against alarmism.
That seems more likely to be a tussle over funding. If the panic-stricken screechers make enough noise, it might adversely affect their member's funding. Something rather like that has happened before:
...It was rumoured in some of the UK national press of the time that Margaret Thatcher watched Professor Fredkin being interviewed on a late night TV science programme. Fredkin explained that superintelligent machines were destined to surpass the human race in intelligence quite soon, and
In both of these debates, the change in "margin of victory" between people was smaller than the number of people who voted the first time but not the second. In the debate with Yudkowsky, Hanson's margin of victory went from -5 to 1, and 20 voters dropped out. In the debate with Caplan the margin went from 32 to 5--actually getting smaller--and 17 voters dropped out. I'm not sure if we can even determine a "winner" in terms of audience popularity, with that many disappearing votes. Is it normal in debates for large numbers of audience members to not vote at the end?
My two cents:
Legg's Is there an Elegant Universal Theory of Prediction? is somewhat relevant to parts of this discussion.
How about a LW poll regarding this issue?
(Is there some new way to make one, since the site redesign, or are we still at vote-up-down-karma-balance pattern?)
Hanson seems to agree that if we get human-level agents that are cheap to run, this gets us a local takeover. I don't think that having cheap chimp-level agents widely available at that time overturns the advantage of gaining access to cheap human-level agents. So if we grant that the capability of AIs gets increased gradually and publicly, all that a local group needs to take over the world is make the step from chimp-level state-of-the-art agents to human-level agents before any other group does that. If chimp-level agents are not that different from hum...
I don't have an intuition for what would happen if you ran a chimp-level intelligence very fast. The ratio Yudkowsky mentioned in the recording was 2500 years of human-in-skull thinking = 8 hours of human-in-laptop thinking. Is it completely obvious that 2500 years of chimp thinking would yield nothing interesting or dangerous?
Chimps haven't accomplished much in the last 2500 years but that's at least partly because they don't pass on insights between generations. Can we stipulate 2500 years of chimp memory, too?
Hanson has made a lot of comments recently about how intellligence is poorly defined and how we don't really know what it is - e.g. 77:30 and 83:00 minutes in. I think we do now have a pretty good idea about that - thanks to the Hutter/Legg work on universal intelligence. If Hanson was more familiar with this sort of material, I rather doubt he would say the kinds of things he is currently saying.
Whoever asked Robin about his opinion that social skills separated humans from chimpanzees "Can you envision a scenario where one of the computers acquired this 'Social Skill' then said to all the other computers "hey guys lets go have a revolution" " Love that comment
73:00 - this seems to be a mis-summary by Hanson. I am pretty sure that Norvig was saying that complex models were still useful - not that simpler ones didn't even exist.
The situation is similar to that with compression. If you can compress a bit that is still useful - and it is easier to do than compressing a lot.
It's not a brain in a box in a basement - and it's not one grand architectural insight - but I think the NSA shows how a secretive organisation can get ahead and stay ahead - if it is big and well funded enough. Otherwise, public collaboration tends to get ahead and stay ahead, along similar lines to those Robin mentions.
Google, Apple, Facebook etc. are less-extreme versions of this kind of thing, in that they keep trade secrets which give them advantages - and don't contribute all of these back to the global ecosystem. As a result they gradually stack u...
Hanson gets polite and respectful treatment for his emulation scenario. I am not convinced that is the right approach. Emulations first is a pretty crazy idea - and Hanson doesn't appear to have been advised about that a sufficiently large number of times yet.
Compared to the farming and industrial revolutions, intelligence explosion first-movers will quickly control a much larger fraction of their new world. He was pro, I was con.
The thesis seems pretty obviously true to me, though there is some issue over how much is "much".
Google or Facebook control a much larger fraction the world compared to farmers or industy folk from decades ago. Essentially technological progress promotes wealth inequality by providing the powerful with technology for keeping control of their wealth and power. So, we have more wealth inequality than ever - and will most likely have even more wealth inequality in the future.
Hanson's debating success is all the more impressive given that he was fighting with a handicap. Imagine how potent his debating would be if he was actually arguing for a correct position!
Given we survive long enough, we'll find a way to write a self-modifying program that has, or can develop, human-level intelligence.
How can I arrive at the belief that it is possible for an algorithm to improve itself in a way to achieve something sufficiently similar to human-level intelligence? That it is in principle possible is not a question here. But is it possible given limited resources? And if it is possible given limited resources, is it efficient enough to pose an existential risk?
The capacity for self-modification follows from 'artificial human intelligence,' but since we've just seen links to writers ignoring that fact I thought I'd state it explicitly.
Humans can learn, that is far from what is necessary to reach a level above your own, on your own. Also, how do you know that any given level of intelligence is capable of handling its own complexity effectively? Many humans are not capable of handling the complexity of the brain of a worm.
This necessarily gives the AI the potential for greater-than-human intelligence due to our known flaws.
That humans have a hard time to change their flaws might be an actual feature, a trade off between plasticity, efficiency and the necessity of goal-stability.
Given A, the intelligence would improve itself to the point where we could no longer predict its actions in any detail.
I don't think that is a reasonable assumption, see my post here. The short version: I don't think that intelligence can be applied to itself efficiently.
...the AI could escape from any box we put it in. (IIRC this excludes certain forms of encryption, but I see no remotely credible scenario in which we sufficiently encrypt every self-modifying AI forever.)
Well, even humans can persuade their guards to let them out. I agree.
...the AI could wipe out humanity if it 'wanted' to do so.
I think it is unlikely that most AI designs will not hold. I agree with the argument that any AGI that isn't made to care about humans won't care about humans. But I also think that the same argument applies for spatio-temporal scope boundaries and resource limits. Even if the AGI is not told to hold, e.g. compute as many digits of Pi as possible, I consider it an far-fetched assumption that any AGI intrinsically cares to take over the universe as fast as possible to compute as many digits of Pi as possible. Sure, if all of that are presuppositions then it will happen, but I don't see that most of all AGI designs are like that. Most that have the potential for superhuman intelligence, but who are given simple goals, will in my opinion just bob up and down as slowly as possible. This is an antiprediction, not a claim to the contrary. What makes you sure that it will be different?
Humans can learn, that is far from what is necessary to reach a level above your own, on your own.
Yes, you also need the ability to self-modify and the ability to take 20 or fail and keep going. But I just argued that the phrase "on your own" obscures the issue, because if one AGI has a chance to rewrite itself (and does not take over the world) then I see no realistic way to stop another from trying at some point.
Also, how do you know that any given level of intelligence is capable of handling its own complexity effectively?
I don't think ...
Link: overcomingbias.com/2011/07/debating-yudkowsky.html