4 ms·

A few hastily considered thoughts in response: I really don't understand why you think "convexity" is a new concept. Isn't this the way most if not all artisa

A few hastily considered thoughts in response: I really don't understand why you think "convexity" is a new concept. Isn't this the way most if not all artisan crafts have worked since antiquity? Master-apprentice and guilds. Either software creation is an engineering discipline or it is a craft. You seem to be cherry-picking. Also I believe that while your characterization of open-source projects as essentially ego trips may be accurate of some, it is so only of a minority and in any case will not withstand any serious feminist critique. I find that your assumption that only superior software engineers can judge the work of other engineers to be patently false and it sets up a false dichotomy. That is the same as saying that only other artists can judge artistic merit. It has no basis in fact or history. What software engineers simply fail to do is provide objective basis for their work because they don't want to lose any bargaining power over management; much the way alchemists conned their own patrons. I contribute to open source projects because it is the right thing to do. It is moral. Not to enhance my reputation in some false gift economy or portfolio waving. Perhaps I am "old school" but this is why reading Stallman was important to my development. Finally, one does not get satisfaction from simply seeing your cool stuff in use. For example, seeing an evil regime use your cool stuff to repress its citizenry would give me no satisfaction at all.

pagliara 14y ago
He never said convexity is a new concept. In the context of the industrial economy, convexity is a new problem because the past 100 years of industrial economic development has relied on concave work that is now being outsourced to computers.
- shawn-butler 14y ago
  Ok. That was not my interpretation of his long essay which includes no context of existing literature on the subject the absence of which leads me to believe the author believes the term or concept is novel. Your interpretation is equally fair and interesting but seems off-topic because isn't the efficient creation of correct source code almost exclusively the domain of human work? I probably misunderstand you.
  - michaelochurch 14y ago
    I would say that I've "independently discovered" convexity. I'm the first (to my knowledge) to use it to describe human performance at work. I'm far from the first to have the idea. Traders talk about gamma (yep, one of those Greeks) which is the second derivative of an asset's sensitivity to an underlying price. If your portfolio has positive gamma with respect to an underlying, that means you benefit from (zero-mean) volatility in it. What I realized at some point was that labor tends to be split between "fun" work that has extreme inequality (e.g. writing novels) and volatility, and boring, commodity work. I started reading up on the economics of labor and S-curves (logistics) kept coming up and I realized what was going on. An industrial enterprise is, in truth, much like an artificial neural net, because it's a bunch of logistic input/output relationships feeding into each other. It's a simple matter of calculus to show that the old industrial optimization (maximum yield on capital, within bounded risk constraints) favors concave work, and that explains a lot about why businessmen have favored it. But now it's all being automated, because machines are even better at it (more precision) than we are. That leaves us doing convex work, but there are some serious social problems in a world where (a) concave work is increasingly being gobbled up by machines, but (b) there's no basic income for people who can't supply convex work (again, the only stuff that needs humans) to the market at a living wage. See: http://michaelochurch.wordpress.com/2013/04/03/gervais-macleod-21-why-does-work-suck/ http://michaelochurch.wordpress.com/2013/04/03/gervais-macle... We're in the beginning stages of seeing Convexity on the larger scale, as people struggle to get the corporate jobs that our parents' generation took for granted.
michaelochurch 14y ago
I really don't understand why you think "convexity" is a new concept. Convexity's not new. What is new is the inability for humans to compete with machines on concave labor. Convexity and concavity come (in theory, anyway) from the logistic form: L(x; ...) = V + A/(1+exp(-S(x-D))) It grows exponentially (convex) over x << D, becomes linear around x = D with an inflection point at (D, V + A/2), and then turns concave at x >> D, leveling off to meet a horizontal asymptote at x = V + A. V is a vertical offset, A is maximum potential/saturation, S is precision, and D is difficulty. Those are constants per function; x is the input (performance). L computes the economic payoff. Generally, values of x tend to be fairly narrowly distributed based on "the state of the art" and if the typical x << D, we have "hard" (concave) labor. If the typical x >> D, then we have concave/easy stuff. Since input is abstract here, it's typical to say the "typical" x is at 0. What management has been doing for 200 years is increasing precision (S) and pushing D to the left... making labor concave/easier and precise. That is a good thing. That's exactly what you should do if you're managing a concave job. Incidentally, if you have a convex job, increasing S is not what you want to do because it increases your failure rate. Now, concave work is (from an economic perspective) great because the ratio of the payoff to the risk (first derivative of L) is high, and that means that in the "knapsack problem" of trying to maximize value under risk constraints, you want to be doing as much of that as you can. Right. No argument there. With machines, S → ∞ (step function) which you never see with humans, and not only that but the "step" level is usually quite low (in cost; cents per hour of electricity) once it's properly programmed. So what this means is that, for a given business enterprise, we can often give most or all of the concave work over to machines and there's still room in our "knapsack" for high-yield (convex) work. For convex work, value-per-risk is nearly constant and thus the traditional industrial optimization heuristic (value-per-risk) becomes irrelevant. Convex is risk-intrinsic, meaning that it is the risk (rather than unpleasantness) that explains the value. Thus, the convex work is what's left for us. If it's concave, there's no need for humans to do. The good news? When work is convex, demand for it is effectively limitless. (Limits may exist; the S-curve may level off and turn concave. But if the work's convex, that means the state of the art isn't anywhere near there.) Also I believe that while your characterization of open-source projects as essentially ego trips may be accurate of some Oh no, that's is not what I am saying at all. What is egotistical about doing great work and sharing it for free? I am very much pro-OSS and would never say that. I find that your assumption that only superior software engineers can judge the work of other engineers to be patently false and it sets up a false dichotomy. We're just going to disagree on this one, because I think that there are attributes of work (especially presentation) that are easy to assess, and others that are hard, and, on the whole, management is about as equipped to assess code-quality issues as I am to go into a chip factory and start spouting off theories about how I think electrical engineering should work. (I have no EE experience.) Imagine me going into an EE shop and saying, "No more use of imaginary numbers! I don't quite understand how something not real can be useful! If I catch you taking a square root of a negative number, you're fired!" Well, that's how management often comes across on software. What software engineers simply fail to do is provide objective basis for their work because they don't want to lose any bargaining power over management Sorry, but that's nonsense. If anything, we lack bargaining power over management, because most of us are terrible negotiators who are socially marginal and awkward, and our inability as a group to demand what we're worth is extremely detrimental. I contribute to open source projects because it is the right thing to do. It is moral. Not to enhance my reputation in some false gift economy or portfolio waving. Perhaps I am "old school" but this is why reading Stallman was important to my development. Finally, one does not get satisfaction from simply seeing your cool stuff in use. For example, seeing an evil regime use your cool stuff to repress its citizenry would give me no satisfaction at all. That's awesome, and there's no disagreement coming from this corner.

pagliara 14y ago

He never said convexity is a new concept. In the context of the industrial economy, convexity is a new problem because the past 100 years of industrial economic development has relied on concave work that is now being outsourced to computers.

shawn-butler 14y ago
Ok. That was not my interpretation of his long essay which includes no context of existing literature on the subject the absence of which leads me to believe the author believes the term or concept is novel. Your interpretation is equally fair and interesting but seems off-topic because isn't the efficient creation of correct source code almost exclusively the domain of human work? I probably misunderstand you.
- michaelochurch 14y ago
  I would say that I've "independently discovered" convexity. I'm the first (to my knowledge) to use it to describe human performance at work. I'm far from the first to have the idea. Traders talk about gamma (yep, one of those Greeks) which is the second derivative of an asset's sensitivity to an underlying price. If your portfolio has positive gamma with respect to an underlying, that means you benefit from (zero-mean) volatility in it. What I realized at some point was that labor tends to be split between "fun" work that has extreme inequality (e.g. writing novels) and volatility, and boring, commodity work. I started reading up on the economics of labor and S-curves (logistics) kept coming up and I realized what was going on. An industrial enterprise is, in truth, much like an artificial neural net, because it's a bunch of logistic input/output relationships feeding into each other. It's a simple matter of calculus to show that the old industrial optimization (maximum yield on capital, within bounded risk constraints) favors concave work, and that explains a lot about why businessmen have favored it. But now it's all being automated, because machines are even better at it (more precision) than we are. That leaves us doing convex work, but there are some serious social problems in a world where (a) concave work is increasingly being gobbled up by machines, but (b) there's no basic income for people who can't supply convex work (again, the only stuff that needs humans) to the market at a living wage. See: http://michaelochurch.wordpress.com/2013/04/03/gervais-macleod-21-why-does-work-suck/ http://michaelochurch.wordpress.com/2013/04/03/gervais-macle... We're in the beginning stages of seeing Convexity on the larger scale, as people struggle to get the corporate jobs that our parents' generation took for granted.

michaelochurch 14y ago

I really don't understand why you think "convexity" is a new concept. Convexity's not new. What is new is the inability for humans to compete with machines on concave labor. Convexity and concavity come (in theory, anyway) from the logistic form: L(x; ...) = V + A/(1+exp(-S(x-D))) It grows exponentially (convex) over x << D, becomes linear around x = D with an inflection point at (D, V + A/2), and then turns concave at x >> D, leveling off to meet a horizontal asymptote at x = V + A. V is a vertical offset, A is maximum potential/saturation, S is precision, and D is difficulty. Those are constants per function; x is the input (performance). L computes the economic payoff. Generally, values of x tend to be fairly narrowly distributed based on "the state of the art" and if the typical x << D, we have "hard" (concave) labor. If the typical x >> D, then we have concave/easy stuff. Since input is abstract here, it's typical to say the "typical" x is at 0. What management has been doing for 200 years is increasing precision (S) and pushing D to the left... making labor concave/easier and precise. That is a good thing. That's exactly what you should do if you're managing a concave job. Incidentally, if you have a convex job, increasing S is not what you want to do because it increases your failure rate. Now, concave work is (from an economic perspective) great because the ratio of the payoff to the risk (first derivative of L) is high, and that means that in the "knapsack problem" of trying to maximize value under risk constraints, you want to be doing as much of that as you can. Right. No argument there. With machines, S → ∞ (step function) which you never see with humans, and not only that but the "step" level is usually quite low (in cost; cents per hour of electricity) once it's properly programmed. So what this means is that, for a given business enterprise, we can often give most or all of the concave work over to machines and there's still room in our "knapsack" for high-yield (convex) work. For convex work, value-per-risk is nearly constant and thus the traditional industrial optimization heuristic (value-per-risk) becomes irrelevant. Convex is risk-intrinsic, meaning that it is the risk (rather than unpleasantness) that explains the value. Thus, the convex work is what's left for us. If it's concave, there's no need for humans to do. The good news? When work is convex, demand for it is effectively limitless. (Limits may exist; the S-curve may level off and turn concave. But if the work's convex, that means the state of the art isn't anywhere near there.) Also I believe that while your characterization of open-source projects as essentially ego trips may be accurate of some Oh no, that's is not what I am saying at all. What is egotistical about doing great work and sharing it for free? I am very much pro-OSS and would never say that. I find that your assumption that only superior software engineers can judge the work of other engineers to be patently false and it sets up a false dichotomy. We're just going to disagree on this one, because I think that there are attributes of work (especially presentation) that are easy to assess, and others that are hard, and, on the whole, management is about as equipped to assess code-quality issues as I am to go into a chip factory and start spouting off theories about how I think electrical engineering should work. (I have no EE experience.) Imagine me going into an EE shop and saying, "No more use of imaginary numbers! I don't quite understand how something not real can be useful! If I catch you taking a square root of a negative number, you're fired!" Well, that's how management often comes across on software. What software engineers simply fail to do is provide objective basis for their work because they don't want to lose any bargaining power over management Sorry, but that's nonsense. If anything, we lack bargaining power over management, because most of us are terrible negotiators who are socially marginal and awkward, and our inability as a group to demand what we're worth is extremely detrimental. I contribute to open source projects because it is the right thing to do. It is moral. Not to enhance my reputation in some false gift economy or portfolio waving. Perhaps I am "old school" but this is why reading Stallman was important to my development. Finally, one does not get satisfaction from simply seeing your cool stuff in use. For example, seeing an evil regime use your cool stuff to repress its citizenry would give me no satisfaction at all. That's awesome, and there's no disagreement coming from this corner.