Depends mainly on how we both learn best. For me, when it comes to learning a new language that tends to be finding a well-defined, small (but larger than toy) project and implementing it, and having someone to rubber-duck with (over IM/IRC/email is fine) when I hit conceptual walls. I’m certainly up for tackling something that would help out MIRI.
LM7805
Karma: 130
if some post or comment is confusing to you, don’t skip it and leave it to people who might have the right background. Ask for clarifications.
I am hereby publicly committing to doing this at least once per day for the next week.
Modafinil eliminates the feeling of being tired, but not the body’s need for sleep. Being in sleep deficit weakens the immune system, and I’ve seen long-term modafinil use without sleep recovery end in pneumonia. So, if I take modafinil to work on a project or stay alert for a long drive, and I finish before the modafinil wears off, I’ll go to sleep anyway, because even if my brain doesn’t care whether it gets sleep or not, my body needs it. With stimulants like caffeine, getting to sleep before the stimulant wears off is difficult and leads to less sleep.
Most social-science studies are designed to elicit answers in such a way that the participant doesn’t realize what question is actually being asked. For example, when William Labov studied the distribution of rhoticity in spoken English, he asked people innocuous questions whose answers contained the sound /r/ in (phonetic) environments where rhoticity can occur. He’d go into a multi-story department store, look at the map, and ask an employee something along the lines of “Where can I find towels?” so that the person would answer “Those are on the fourth floor.” Similarly, the wi-fi study wasn’t looking at usability any more than Labov was interested in towels; they were really eliciting “willingness to do something dangerous” as a proxy for (lack of) risk awareness. As long as the measure is wearing unique clothing, participants shouldn’t be able to recognize it.
One class of questions you didn’t bring up has to do with perceptions of risk. There was a poster at this year’s USENIX Security about a Mechanical Turk experiment that purported to be a Starbucks study evaluating the usability of a new method of accessing the wi-fi at Starbucks locations: click here to install this new root certificate! (Nearly 3⁄4 of participants did so.) I can’t find the poster online, but this short paper accompanied it at SOUPS.
I’ve TAed a class like the Programming Languages class you described. It was half Haskell, half Prolog. By the end of the semester, most of my students were functionally literate in both languages, but I did not get the impression that the students I later encountered in other classes had internalized the functional or logical/declarative paradigms particularly well—e.g., I would expect most of them to struggle with Clojure. I’d strongly recommend following up on that class with SICP, as sketerpot suggested, and maybe broadening your experience with Prolog. In a decade of professional software engineering I’ve only run into a handful of situations where logic programming was the best tool for the job, but knowing how to work in that paradigm made a huge difference, and it’s getting more common.
My local hackerspace, and broadly the US and European hacker communities. This is mainly because information security is my primary focus, but I find myself happier interacting with hackers because in general they tend not only to be highly outcome-oriented (i.e., inherently consequentialist), but also pragmatic about it: as the saying goes, there’s no arguing with a root shell. (Modulo bikeshedding, but this seems to be more of a failure mode of subgroups that don’t strive to avoid that problem.) The hacker community is also where I learned to think of communities in terms of design patterns; it’s one of the few groups I’ve encountered so far that puts effort into that sort of community self-evaluation. Mostly it helps me because it’s a place where I feel welcome, where other people see value in the goals I want to achieve and are working toward compatible goals. I’d encourage any instrumental rationalist with an interest in software engineering, and especially security, to visit a hackerspace or attend a hacker conference.
Until recently I was also involved in the “liberation technology” activism community, but ultimately found it toxic and left. I’m still too close to that situation to evaluate it fairly, but a lot of the toxicity had to do with identity politics and status games getting in the way of accomplishing anything of lasting value. (I’m also dissatisfied with the degree to which activism in general fixates on removing existing structures rather than replacing them with better ones, but again, too close to evaluate fairly.)
Also, presuming that the talk Andreas Bogk has proposed for 30c3 is accepted, you’ll want to see it—it’s a huge pragmatic leap forward. (I apologize for not being at liberty to go into any more detail than that. The talk will be livestreamed and recorded, FWIW.)
Eh, cryptocurrency has been a thing since the 80s (see Chaum’s Blind signatures for untraceable payments, 1983), but it was patent-encumbered for a long time. If you look at the people involved, they’re all cypherpunks of one stripe or another, so I’d place cryptocurrency as a subset of hackers at least initially. (The community has certainly expanded since the emergence of bitcoin.)
Toby Ord has some elderly C code (see Appendix II) that he used in his societal iterated prisoner’s dilemma tournaments. You’d have to modify it for your purposes, but it’s a small codebase.
I model people constantly, but agency and the “PC vs. NPC” distinction don’t even come into it. There are classes of models, but they’re more like classes of computational automata: less or more complex, roughly scaling with the scope of my interactions with a person. For instance, it’s usually fine to model a grocery store cashier as a nondeterministic finite state machine; handing over groceries and paying are simple enough interactions that an NFSM suffices. Of course the cashier has just as much agency and free will as I do—but there’s a point of diminishing returns on how much effort I invest into forming a more comprehensive model, and since my time and willpower are limited, I prefer to spend that effort on people I spend more time with. Agency is always present in every model, but whether it affects the predictions a given model outputs depends on the complexity of the model.
The way I see it, the difference between an online algorithm and a self-training AI is that the latter would not only need such a heuristic—let’s call it “knowledge base evaluation”—it would also need to be able to evaluate the fitness of novel knowledge base evaluation heuristics.
On reflection, I’m now contradicting my original statement; the above is a stab toward an algorithmic notion of “self-training” that is orthogonal to how restricted an algorithm’s training input set is, or who is restricting it, or how. Using this half-formed notion, I observe that Google’s ranking algorithm is AFAIK not self-training, and is also subject to a severely restricted input set. I apologize for any confusion.
Hold on, is he trying to imply that prokaryotes aren’t competitive? Not only does all single-celled life compete, it competes at a much faster pace than multicellular life does.
The Feynman Lectures on Computation did this for me by grounding computability theory in physics.
I wouldn’t call Google’s search personalization “self-training” because the user is responsible for adding new data points to his or her own model; it’s the same online algorithm it’s always been, just tailored to billions of individual users rather than a set of billions of users. The set of links that a user has clicked on through Google searches is updated every time the user clicks a new link, and the algorithm uses this to tweak the ordering of presented search results, but AFAIK the algorithm has no way to evaluate whether the model update actually brought the ordering closer to the user’s preferred ordering unless the user tells it so by clicking on one of the results. It could compare the ordering it did present to the ordering it would have presented if some set of data points wasn’t in the model, but then it would have to have some heuristic for which points to drop for cross-validation.
The way I see it, the difference between an online algorithm and a self-training AI is that the latter would not only need such a heuristic—let’s call it “knowledge base evaluation”—it would also need to be able to evaluate the fitness of novel knowledge base evaluation heuristics. (I’m torn as to whether that goalpost should also include “can generate novel KBE heuristics”; I’ll have to think about that a while longer.) Even so, as long as the user dictates which points the algorithm can even consider adding to its KB, the user is acting as a gatekeeper on what knowledge the algorithm can acquire.
Are self-training narrow AIs even a going concern yet? DeepQA can update its knowledge base in situ, but must be instructed to do so. Extracting syntactic and semantic information from a corpus is the easy part; figuring out what that corpus should include is still an open problem, requiring significant human curation. I don’t think anyone’s solved the problem of how an AI should evaluate whether to update its knowledge base with a new piece of information or not. In the Watson case, an iterative process would be something like “add new information → re-evaluate on gold standard question set → decide whether to keep new information”, but Watson’s fitness function is tied to that question set. It’s not clear to me how an AI with a domain-specific fitness function would acquire any knowledge unrelated to improving the accuracy of its fitness function—though that says more about the fitness functions that humans have come up with so far than it does about AGI.
It’s certainly the case that an above-human general intelligence could copy the algorithms and models behind a narrow AI, but then, it could just as easily copy the algorithms and models that we use to target missiles. I don’t think the question “is targeting software narrow AI” is a useful one; targeting software is a tool, just as (e.g.) pharmaceutical candidate structure generation software is a tool, and an AGI that can recognize the utility of a tool should be expected to use it if its fitness function selects a course of action that includes that tool. Recognition of utility is still the hard part.
In the past I’ve manually edited my hosts file to point time-wasting sites to 127.0.0.1, which is a small obstacle at most—I can easily edit it back—but is enough of a speedbump to remind me “oh, right, I’m trying to avoid that right now.” This time I haven’t needed to; I managed to become so frustrated with the fact that my queue was saturated with short-term rewards that weren’t netting me anything in the long run that choosing not to visit the sites doesn’t seem to require much willpower.
This leads me to wonder whether I could eliminate other influences I want to avoid by making myself sufficiently frustrated with them, but that seems like it would be an unpleasant skill to pick up.
Have you read Pierce’s Types and Programming Languages? If so, would you say it provides sufficient foundation for this book?
All of these are dummy subjects. English does not allow a null anaphor in subject position; there are other languages that do. (“There”, in that last clause, was also a dummy pronoun.)
There’s a lot of content here, and replying to old posts is encouraged, so I think there should be plenty of material.