Software Runs the World: How Scared Should We Be That So Much of It Is So Bad?

How many of the business processes that software manages were better before the software took over? MOST companies I've seen inside were a hodge-podge of bad/missing/sloppy paperwork and poorly documented or even understood guidelines and policies. Software amplifies everything about a business, both the good and the bad. The downside is that because so much is networked today, something "going wrong" can have much larger consequences (and faster) than a wrong decision carried out on paper 30 years ago.

Holy crap, The Atlantic has gone down hill in the past few years. It feels like they try to become the next Huffington Post. A couple of years ago it felt like a serious competitor to magazines like The New Yorker or The Economist. Then they rebranded themselves from The Atlantic Monthly to The Atlantic and started their "social media revolution".

This is an incredibly shoddy article, it even closes with an absolutely brain-dead "what if"... the way average high school students close their essays.

> And perhaps the most mission-critical of all mission-critical applications are the ones that underpin the securities markets where a large share of the world's wealth is locked up.

Really? I don't even consider the stock market software as mission critical. I mean, Jane Street's software is written in OCaml...

Mission critical is software that controls lives. Airplane, spaceship, heart pacer, ...

So, here's the unpleasant, unspoken, often subliminal truth about lots of enterprise software. It's supposed to be about automation to increase efficiency. In reality, it's often largely about control. Software enforces a certain workflow and can be used to allow or disallow certain actions. It's a way to enforce the procedures in the 3 ring binder.

One recurring pattern I've seen in enterprise software projects is separation between users and the developers. Often, there is this game of "telephone" where the user's managers talk to a manager above them, then there might be another layer above before information can start heading back down to the devs. Often, one is strictly forbidden to talk to users, except in exceptional situations. This is because the project is largely about control, so restrictions on communication with the users is necessary, since many of the unstated goals have to do with the frustration of the user's wishes.

What's even worse is when this power of control is used in political infighting.

People run the World: How scared should we be that so many of them [insert pet peeve].

Mine is: ...will vote for war and more spending.

The article tries to make a point and then at the end evades the conclusion:

> The only real solution is to acknowledge that computer programs are going to fail and try to minimize the damage they can cause in advance. [the best way to do this is to not use the software in question at all]

No, the only real solution is to provide a meaningful warranty with software (rather than typical "no warranty" clauses). Regulation can enforce this and the price will have to be paid by the customer. We have put up with these crappy licenses for decades and the result is the buggy mess we have now and no way for customers to demand a working product. Mistakes will still be made, but they shouldn't harm the customer more than the vendor.

http://mil-embedded.com/articles/software-warranties-new-era...

Recently I have been using SPARK and it's a pretty different development mindset than the embedded C that I had been used to. It makes me wonder if other development areas could benefit from adopting some of this mindset and discipline.

For example, just to take a random example of Java or Python, what would it look like to create a strict subset of language features which allowed for formally provable code and to create a static analysis tool which could achieve it. I realize that this sort of thing isn't for many developers or many sorts of projects. However, I would like to see more folks taking the best parts of High Integrity Software development and bringing it out to the wider world.

Steve Yegge wrote about this recently: http://news.ycombinator.com/item?id=4365255

Humans are the resilience in any system. No system is ever perfect, and no system will ever be. I think we'll be fine.

It's "bad" because of three factors: (1) complexity, (2) inadequate motivation, (3) lack of understanding.

First, complexity. Bridges aren't doing several million operations per second. They either bear the weight or they don't. They handle a known temperature range. If it can function at -50 F and 120 F, then it can handle the typical 65 F day. With a bridge, people put a lot of work into solving a simple problem extremely well and reliably. Software can be developed this way (Unix philosophy) but most software isn't. Small programs are written to solve problems well, but often invisibly; large-program methodologies exist to get promotions for higher-ups.

There are few things in the world like software, which is pure logic. We don't have the tools to understand the complexity, and what tools we do have are often used to write more complex software, not understand the software we have (see: IDEs and their "four wheel drive problem" of getting people stuck in more inaccessible places.)

Second, a lot of "software engineers" aren't very good and don't have the incentives and leeway to get better. That requires lifelong learning and professionalism in the true sense of the word. We're not a profession. Many things define a profession, but some salient traits of professions are: (1) an ethical ruleset that supersedes managerial authority-- not only are you allowed to refuse your manager on ethical grounds, but you have to do so; "I was just following orders" is not an excuse-- and (2) an expectation and allowance that the professional will dedicate half (~20 hours per week) her working time to continuing education, networking, and other varieties of "off-meter" work that will never occur in a typical manager/subordinate context because they benefit the professional's long-term development rather than the manager's parochial goals, (3) a very high degree of autonomy in work regarding social-bullshit protocols, but with very strong restrictions on the things that matter (i.e. no one sets working hours or vacation limits over a research professor, but if he steals another's work or publishes results he knows to be false, it's career-ending). None of these apply to software engineering. (1) If you disagree with your boss about how things should be done on ethical grounds related to software quality or intellectual honesty, you don't have an appeal process. You just get fired. (2) Most software engineers, if they want to keep learning, have to do it on their own time; hence, the stagnation that sets in for all but the most energetic and ambitious. (3) Nope. Here, the software industry looks more like middle-middle class white-collar culture (show up at certain times, take orders) than upper-middle-class professionalism. The salaries are often professional level, but work conditions are merely white-collar.

In a truly professional environment, manager-as-SPOF is avoided like the plague and people have genuine incentives to do good work, not just enough work to appease the manager. The manager still has some power and influence, but the role is more like a graduate advisor than an overseer. Also, professionals are encouraged to build lifelong reputations and seek visibility. (In a typical software company, trying to do this gets you the smear of a careerist and a "socialite".) Professional environments motivate people to do their best work. Merely white-collar environments don't.

The third issue is that we simply don't have a good understanding of software. This may be derived from the two above: complexity and lack of professionalism. Or it may result from something else entirely.

We know that, theoretically speaking, "it's impossible to reason about code" (cf. Halting Problem). Well, my response invariably is that that's correct: it's impossible to reason about arbitrary code. But we shouldn't be writing arbitrary code. Ever. We should write simple code that doesn't rely on extremely high-level mathematical or conceptual insight (which we model as non-computational "genius", though I have no interest in getting sidetracked into the debate of whether we are or are not computers) to understand.

I think individual people are good at writing software. People don't usually set out to write "arbitrary code". There are many individual software engineers who (1) use only as much complexity as they can handle, and (2) act professionally in spite of the lack of incentives or requirements ensuring it. Which means that small software can be good. What I like about the Unix philosophy and small-program philosophy is that it enables islands of quality to exist, and eventually bridges between those islands, which can lead to generally decent systems even though it's GRAI (generally recognized as impossible) to have high quality in an entire codebase. Systems design is all about accepting the possibility of failure amid complexity. It's about using small components that do one thing really well and interchangeable parts, not single-program mudballs. Engineers get this. Managers, who conflate bigness (cf. interview questions like "what's the largest team you've ever managed?" and metrics like kLoC) with success, generally don't.

The problem is that, once a program has a certain number of hands pass over it, it turns to shit. Even if the original code was good, entropy sets in at some point. One bad apple spoils the barrel, and a "bad apple" doesn't have to be a bad programmer. It could be a skilled programmer who hasn't had a promotion in 4 years and no longer gives a shit. The reason why big-program methodologies generally produce legacy mudballs is that it's just impossible to prevent the "one bad apple" problem from corrupting the whole program.