Great power must be accompanied by great responsibility, which remains largely absent in Silicon Valley. The Valley's fantastical vision of a techno-utopian future where robots do all the menial jobs, Universal Basic Income is firmly in place, and everyone can freely pursue their creative endeavors, is just that: a fantasy. Between that fantasy of tomorrow and the reality of today is a gap into which real families, with real mortgages to pay and real mouths to feed, are falling into.Berkeley CS students must help address sexism in their field
In Gender Trouble, a classic text in gender theory, Judith Butler writes, "Gender is performative... There is only the deed, no doer behind the deed". But how does one understand a doer and a deed? One option is to gorge oneself on the rest of Butler's thick, dense book. But a computer science student has a shortcut to understanding the concepts.
Try not to laugh: A doer is a piece of data, whereas a deed is an execution of a function. ... Therefore, if one understands the halting problem, one has a significant grasp on the idea that gender is performative: Someone's gender is not a piece of data that can be copied, modified, and tossed around, but an arbitrary program about which we do not know if it even halts or not.
This is not an accident: Computer science, at its core, is a series of elegant ideas that, in a way, form the blueprint of thought. It is no accident that computer science has allowed the compression of so much labor in the world. CS 61A is useful not just for landing that sweet internship at Google; it can be used, in fact, to actually make the world a better place -- not by selling your soul to a Silicon Valley corporation intent on increasing economic disparity and disseminating racist manifestos, but by understanding and extending texts of central social justice importance such as Gender Trouble.
Various moral conundrums plague population ethics: The Non-Identity Problem, The Procreation Asymmetry, The Repugnant Conclusion, and more. I argue that the aforementioned moral conundrums have a structure neatly accounted for, and solved by, some ideas in computability theory. I introduce a mathematical model based on computability theory and show how previous arguments pertaining to these conundrums fit into the model. This paper proceeds as follows. First, I do a very brief survey of the history of computability theory in moral philosophy. Second, I follow various papers, and show how their arguments fit into, or don't fit into, our model. Third, I discuss the implications of our model to the question why the human race should or should not continue to exist. Finally, I show that our model ineluctably leads us to a Confucian moral principle.The Poetry of Computer Science, the Computer Science of Poetry
Using computer science to talk about moral philosophy is a sort of perversion. In a sense, all philosophy is a sort of perversion. But as a smartypants once said, the purpose of philosophy is the dissolution of philosophy. I know at least a dozen grandmothers and grandfathers, most of them selling fish at a street market, who know everything this book can say and more. The audience I have in mind are the cynics, the highly educated, the "rationalists", who have retreated to their enclave, who refuse to believe anything that cannot be proven, who endorse things like utilitarianism, behaviorism, and The Bell Curve. I believe I can change their minds because they are rational, and rationality is an admirable ontological property. Rationality, for all its faults, does one job very well: when proven wrong, it clips off, however much it hurts, that irrational cancerous outgrowth, the misapplication of ego. What this book has tried to do is to show that the Modern Scientific World View, and its moral philosophy, which purports to be based on rationality, is utterly irrational. I tried to show this using something every "rationalist" would agree as a method for achieving rational truth: theoretical computer science.Paper 3: Uncomputability and the Categorical Imperative
When Kant said we must not treat humans as mere means, but ends in themselves, he was saying that humans are arbitrary programs, not specific programs. Therefore, we cannot use them as mere means, and must treat them as ends in themselves. And the sense just used in cannot and must is not that of mere moral indoctrination. It is the authority of mathematics. So Williams has been answered: we legislate to the moral sentiments by right of mathematical fact.Culture, Computation, Morality
I point to a deep and unjustly ignored relation between culture and computation. I first establish interpretations of Piaget's and Vygotsky's theories of child development with the language of theoretical computer science. Using these interpretations, I argue that the two different possible routes to Piagetian disequilibrium -- a tendency to overaccommodate, and a tendency to overassimilate -- are equivalent to the two distinct cultural tendencies, collectivistism and individualism. I argue that this simple characterization of overaccommodation versus overassimilation provides a satisfying explanation as to why the two cultural tendencies differ in the way they empirically do. All such notions are grounded on a firm mathematical framework for those who prefer the computable, and grounded on my personal history for those who prefer the uncomputable.The Value Function of Human-Compatible AI
It seems that complexity theory has a surprising relationship with morality. Generally, if a decision renders the environment seemingly more complex than another decision, this decision is considered moral.How to Do Things With Metaphors
A Markov Chain algorithm can "learn" by being fed some text, and then it can generate some string of words according to the aforementioned assumptions. To give a feel of what this does, I have written a simple Markov Chain algorithm, fed it some of my own poetry, and asked it to spit some back out:Hobbes and Xunzi on Human Nature: the Fixed and the ChangingThey seep and leap and feel and all other
mouths seemed occupied with drooling moist. Besides her bed, a bed
full of lava. Why won’t you melt inside of me? Your components are nice,
and they looked at the sky and their faces grew
red, and they sang in despair a lullaby rhythm. The
rain, rolling by like guilty trains.
The object is the fixed, the existent; the configuration is the changing, the variable. (TLP 2.0271, Wittgenstein)
Koltun et al. present a novel machine learning architecture in Learning to Act By Predicting the Future [DK16]. In this report, we test the flexibility and general applicability of Koltun's architecture. To this end, we:Identifying Semantic Components From Cross-Language Variation, Structured Lexical Sources, and Corpora: a Review of Recent Literature
(1) Reimplement the code in Keras [Cho+15] and test it on VizDoom [Kem+16] to reproduce results;
(2) Apply the model to a substantially simpler environment, the game of Flappy Bird;
(3) Test the offline capabilities of Koltuns model, training an offline agent with a set of observations and actions from an expert agent.
Distributional semantics makes a compelling case of how meaning arises from statistical distributions of data. So far, most accounts of distributional semantics have focused on English words and using vectors to represent those words. However, that landscape is quickly changing. Recent efforts on distributional semantics may be divided into at least four clusters: using more complicated mathematics to represent words and compositions of words, leveraging high quality structured knowledge bases, combining multiple languages, and going beyond corpus training data to visual, auditory, and olfactory training data, to giving the machine multiple modalities, so to speak. What emerges is an exciting hodgepodge of ideas that look ripe to advance further and intertwine together. In what follows, I have listed papers of interest with a brief summary of their methods and, if particularly promising, an extended discussion of their formalizations and limitations.