hacker42

I liked this interview with Judea Pearl on free will: https://youtube.com/watch?v=sg7Oq4suH_E

The summary is basically that to our best knowledge there is no free will because decisions are caused by neural activities which in turn are caused by sensory input and noise (but unlikely by quantum noise). However, humans have evolved a strong sense of agency because that's simply an efficient way to reason about machines that produce actions in response to the entirety of their sensory input (especially regarding parent child relationships and mutual behavior correction). This neuroarchitectural bias is essentially an illusion of free will that is so firmly wired into our brains that we cannot escape it. It is also the reason why the idea of a God comes so intuitively to many of us: an invisible actor which can be used to explain inexplicable chains of causation and can serve as a very effective metaphor for behavioral error correction (as a proxy for actual social repercussions, and hence relieved from all the complicated and hence fallible power relations to actual social error correction instances).

I'm wondering to which extent this effect can be attributed to illusion of competence, i.e. with age some meta learning program in our brains that is concerned with managing novel data simply becomes lazy because (1) adults really are able to handle most of the situations they are confronted with and (2) they are also expected to be able to deal with any situation. That might simply lead to a detoriation of the sort of novelty detection. Adults tend to automatically explain away things they don't understand (possibly to save face) while the young examine them. With increasing age these kinds of automatisms might simply override or detoriate learning strategies that are required to learn quickly.

But motivated by greed and (geo)politics, rarely by religion.

The last sentence in my previous comment was not very clear. What I meant to say is that there are many non-Muslim African countries which have suffered from Western exploitation, yet we don't witness the kind of religiously motivated violence as in Islamic countries with comparable histories. This is an indicator for different dispositions caused by different sets of cultural memes.

See also 100+ Concepts To Add To Your Cognitive Toolkit (including all comments): https://news.ycombinator.com/item?id=10819355

Back then the inhibition level for violence was much lower than it is today, even in the Middle East. Also the Christian culture went through various changes, i.e. the French revolution and enlightenment, which have never occurred in Islamic cultures. Even the kind of enlightenment in the Golden Age were deeply motivated by religion and were hence inherently unstable and also limited.

I am talking about the cultural difference today, not in the past, about aggression and harmful traditions occurring in reasonably stable and educated middle classes. Note, that I am not trying to whitewash the past of Christianity; both sets of beliefs are bad in that they have or had the potential to produce fundamentalism and badly informed decisions, but I am just arguing for a qualitative difference between Islam and Christianity (especially today's Christianity and Islam), i.e. one is much worse than the other. Today's Christianity does not produce anything as Islam does in other troubled regions such as African countries, even though these people would be in their 'right' to take vengeance for e.g. centuries of slavery, right?

That's just an appeal to nature. Just because something was an evolutionary necessity (and that is not even clear in this case) it does not follow that that thing should actually be desirable for us.

A plausible explanation of aging is for example the evolutionary shadow: From the perspective of your genes you don't matter (much) after you've successfully passed on you genes to the next generation. Thus there is no evolutionary pressure for longevity. The programmed cell decay of aging might be a way of multicellular organisms to fight cancer.

There is a difference in the percentages and arguably in the degree of accepted violence and actually lived misogyny. There is definitely a difference in the extremes (Islam is currently the only ideology to produce suicide bombers in such large quantities). There is also a difference in that the Bible is regarded to consist of interpretations of God's words, whereas the Quran is regarded to contain a direct transcript of Allah's words.

Why not call the flaws what they are? The diffusion of values seems to be a very slow process, if it works at all given the ever growing Muslim population. The percentage of third generation Muslims in Western countries who still don't really oppose ideas of violence and misogyny in the Quran and who would prefer to return to a social order of the lifetime of Muhammed is astonishingly high (~30%), and Muslim fanatics in the Middle East are also often highly educated. They know about Western values, but they don't value them. Why should they, given that our view is all wrong to them?

It boils down to the simple question whether value diffusion is faster than they breed and escape possible upcoming droughts and whether there really is potential for a major cultural clash. Some people say it's obvious in one way or the other. I really don't know. I am also not convinced by anything I'm writing, I am just trying to get some confirmation or disconfirmation of these sort of ideas.

I think addressing cultural flaws more directly could make such social programs more effective.

Sure, the praying alone would not create a huge amount of cognitive load, but it is just symptomatic of an overly binding religion. It must be terribly confusing as a Muslim to see other cultures vastly outperforming them even though these cultures don't follow the word of Allah, e.g. by not throwing people off of cliffs for having wild gay sex and not stoning women for adultery. These are literally the kinds of rules, fears and confusions that shape Islamic cultures.

It seems they are missing an important factor, namely 4) the personal, cultural and economical impairment due to the religion of Islam. Nobody has time to thrive when they have to pray five times a day, when almost every thought and action is dictated by an old book and when half of their human potential is hidden in cloth bags.

The evidence is that this works in the first place.

tl;dr I think you need a small set of tempo-spatially grounded meanings (though not necessarily agent-related) and you can bootstrap everything from that using only textual knowledge.

Here is the reason why I don't believe in that: First of all, it seems that mind space is huge, i.e. there are very different programs that can lead to general intelligence, many of which will be very different from humans (this along is evidenced by the fact how strongly human characters and intellects vary, e.g. highly functioning mental conditions). A lot in machine learning points to that possibility. There are for example many ways to get supervision signals e.g. reconstruction error, prediction error, adversaries, intrinsic motivation (reducing the number of bits required for a representation), compression, sparseness etc.

We basically just need a system that comes up with efficient representations of the world such that it can reason about it, i.e. that it can tell you which hypotheses about the world are likely true given some data. This computation allows you to make predictions and predictions are really at the heart of intelligence. If you can follow a hypothetical trajectory of generated, hallucinated or simulated samples of reality, i.e. samples that likely correspond to what actually happens in the world (and in the agent's own brain), then you can actually perform actions that are targeted at some purpose (e.g. maximizing reward signals). However, there are many sources of data that essentially give you the same information. Whether you create a representation by directly interacting with the world or just watch many examples of how the world generally evolves over time and how different entities interact with one another, you essentially get the same idea about how the world works (except in the first case you also learn a motor system). I think the anthropomorphism is really misplaced here, because computer systems are not dependent on actually performing in the real world. Since computers have near unlimited, noiseless memory and have super fast access to that memory, they can actually plan interactions by careful reasoning on the fly instead of needing to learn e.g. motor skills for manipulating objects, eating and handwriting before one can get anywhere near the performance of computers wrt. access to reliable external memory. A computer system also does not have hormone and neuromodulator levels that need to be met for healthy development (e.g. dopamine), therefore the intuition that deprivation of interaction with the world prevents learning is extremely misleading.

I agree that learning to reason about the world likely does not require experience with motor control and proprioception (i.e. literally how babies do it), though I do think that you either need at least some sort of tempo-spatial experience (e.g. visual). Tempo-spatial representations are just extremely hard to convey by text only. You might get the idea of closeness by saying 'close is when two words are close in a sequence of words' and ordering by saying 'this word comes after that word', but I think it would be very difficult to extrapolate that concept to more than one dimension, and dimensions that actually have not just an ordering but also a metric (just think about our inability to reason about just the fourth dimension). You need rich representations of our 3+1 dimensional world to be able to reason about it and text only gives you perhaps "0.5" dimensions (because it lacks a metric, i.e. it does not convey durations in terms of the ticks of the recurrent network). But I doubt, too, that interaction with the world is necessary, in fact I think it would be rather easy for an AI to simply write motor programs in a programming language given unrestricted and noiseless memory once it has learned to reason about tempo-spatial patterns from just observing them and identifying them with our language-coded shared concept space. It is not constrained to real-time performance of actions as humans are, therefore it can take the much easier way of programming any interaction with the world as needed, on the fly. Our shared concept space likely sufficiently conveys our general (common sense) knowledge how these patterns are known to interact and evolve over time once rudimentary tempo-spatial representations are in place.

That reminded my of this blog article by Scott Aaronson: http://www.scottaaronson.com/writings/bignumbers.html

It's called random optimization or random search depending on whether you sample from a normal or uniform distribution for the random direction. MC typically refers to any algorithm that computes approximated solutions using random numbers (as opposed to Las Vegas algorithms which use random numbers to always compute the correct solution). So, yes, RO, RS and gradient descent are MC local optimization algorithms.

There is evidence that language is fairly smooth though. For example, we can extract e.g. the gender vector from a word embedding space that is learned by a recurrent neural network. That seems to hint at the possibility that words, sentences and concepts live in smooth, high-dimensional manifold that makes them learnable for us in the first place (because in that case they can be learned by small local improvements which seems to be required for biological plausibility). That is also the reason why we have often many words for the same or similar meanings and, conversely, why formal grammars have failed at modeling language.

Arguing from the other direction, neural networks have also already proven to deal with very sharp features. For example the value and policy networks in AlphaGo are able to pick up on subtle changes in the game position. The changes from the placement of single stones can be vast in Go and by no means this is only solved by the Monte Carlo tree search. Without MCTS, AlphaGo still wins in ~80% of the time against the best hand-crafted Go program. The value and policy networks have pretty much evolved a bit of boolean logic, simply from the gradient from the smoothness that results from averaging over a lot of training data.

I have a pet theory that the discovery of sharp features and boolean programs might heavily rely on noise. If the error surface becomes too discrete, we basically need to backup to pure random optimization (i.e. trying any direction by random chance and keep it, if it is better). That allows us to skip down the energy surface even without the presence of a gradient. Of course, such noise can also lead to forgetting, but it just seems that elsewhere the gradient will be non-zero again, so any mistakes will be correct by more learning (or it simply leads to further improvement if the step was into the right direction). Surely, our episodic memory helps in the absence of gradient information as well. If we encounter a complex, previously unknown Go strategy, for example, it will likely not smoothly improve all our Go playing abilities by a small amount. Instead, we store a discrete chaining of states and actions as an episodic memory which allows us to reuse that knowledge simply by recalling it at a later point in time.

I've never understood the desire for beauty in a theory of everything. Reality clearly isn't beautiful through and through, but at its corners it is awfully complex and arbitrary. If anything, reality is much like the outcomes of an evolutionary process, a mere hack of various mechanisms that happen to give rise to the relatively stable patterns we experience. Of course, fundamental research needs the optimism that we can compress all of that in ever smaller formulae, but I think there is no unambiguous evidence that we should actually expect that to be possible. The most striking piece of counter evidence is actually the kind of reality we experience in the first place. Imagine a universe in which its inhabitants would be able to figure out its fundamental principles. These inhabitants would likely immediately conquer all available space and use it to maximize the reward signals that evolution has equipped them with and thus transform everything into something completely different from the world we are experiencing. A universe with a reality like ours and that is fully understandable at the same time is thus (together with various other assumptions) a logical impossibility.

What are some examples what the macro function can be used for? In many cases the macros I need are so specific such that I define them on the fly in VIM (i.e. q [a-z] and then q to stop, finally @ [a-z] for playback). Perhaps I simply don't write enough for macros to take advantage of more permanent macros.

I would also be interested whether one takes notice of the oppressiveness of the Chinese government, e.g. the Great Firewall of China.

Also runaway climate change, super bugs, super viruses, grey goo, AI, GRBs, supervolcanoes…

In which ways is LLDB better than GDB?

It seems to me that Wolfram Research missed a huge opportunity here with regards to Mathematica Notebooks and the Worlfram Cloud. They didn't see or ignored the momentum that open source and the web can gain.

I'm sticking to Phoenix. Are there any advantages in using Lua script instead of JavaScript?

Slight addition using wider, shadowy gradients to conceal some aliasing issues:

    td.ind > img { opacity: 0 }
    td.ind {
        background: 
            linear-gradient(to left, rgba(246,246,240,1) 0px, rgba(246,246,240,1) 14px, rgba(246,246,240,0) 14px),
            repeating-linear-gradient(to right, rgba(0,0,0,0) 0px, rgba(0,0,0,0.03) 39px, rgba(0,0,0,0) 40px);
     }

Screenshot: https://i.imgur.com/f2WfHyp.png

Unfortunately it breaks on the 'threads' page.

Try this:

    background: url("data:image/svg+xml,<svg xmlns='http://www.w3.org/2000/svg' width='12' height='20'><polygon points='0,20 12,20 6,0' style='fill:rgb(179,179,179);stroke:none;' /></svg>") !important;

It would also be handy to have some visual guides for the indentation levels of nested comments as on Reddit. I've hacked my own indentation marks using a CSS injection for now, but this is clearly still not optimal:

    @-moz-document url-prefix(https://news.ycombinator.com/) {
      td { vertical-align: top }
      td > img { background: repeating-linear-gradient(to right, rgba(0,0,0,0), rgba(0,0,0,0) 38px, rgba(0,0,0,0.3) 38px, rgba(0,0,0,0.3) 40px); margin-top: 4px; height: 12px!important; }
      div.votearrow { padding-bottom: 3px; background: url('http://localhost/uparrow.png') !important; background-size: 12px 20px !important; width: 12px!important;}
    }

I've also increased the size of the voting arrow because it's just uncomfortably small. By Fitt's law [0], smaller buttons require more time and effort to click!

[0] https://en.wikipedia.org/wiki/Fitts%27s_law

That relies on noiseless, unbiased data, right? What if an expert gets ruled out by accident?

Memory usage is also annoyingly high in FF for me, but with 380 MB not as high.