Software & Hardware - How complexity and risk define engineering practice and culture.

Striking cultural differences exist between companies that specialise in software development and those that specialise in hardware development & manufacturing.

The key to understanding the divergence of these two cultures may be found in the differing approaches that each takes to risk and failure; driven both by a difference in the expected cost of failure and a difference in the cost associated with predicting failure.

One approach seeks to mitigate the impact of adverse events by emphasising flexibility and agility; the other approach seeks to minimise the chances that adverse events occur at all, by emphasising predictability and control.

In other words, do you design so you can fix your system quickly when it breaks (at the expense of having it break often), or do you design your system so that it very rarely breaks (but when it does, it is more expensive to fix)?

The answer to this question not only depends on how safety-critical the system is, but how complex it is too. The prediction-and-control approach rapidly becomes untenable when you have systems that reach a certain level of complexity -- the cost of accurately predicting when failures are going to occur rapidly becomes larger than the cost of the failure itself. As the complexity of the system under development increases, the activity looks less like development and more like research. The predictability of disciplined engineering falls apart in the face of sufficient complexity. Worse; complexity increases in a combinatorial manner, so we can very easily move from a predictable system to an unpredictable one with the addition of only a small number of innocuous looking components.

Most forms of mechanical engineering emphasise the second (predictive) approach, particularly for safety critical equipment, since the systems are simple (compared to a lot of software) and the costs of failure are high. On the other hand, a lot of software development emphasises the first (agile/reactive) approach, because the costs associated with failures are (normally) a lot less than the costs associated with development.

Of course, a lot of pejorative terms get mixed up in this, like: "sloppy engineering", and "cowboy developers", vs "expensive failures" and "moribund bureaucracy" but really, these approaches are just the result of the same cost/benefit analysis producing different answers given different input conditions.

Problems mainly arise when you use the wrong risk-management approach for the wrong application; or for the wrong *part* of the application. Things can get quite subtle quite quickly, and managers really need to be on top of their game to succeed.

One of the challenges in developing automotive ADAS systems is that a lot of the software is safety critical, and therefore very expensive to write, because of all of the (necessary) bureaucratic support that the OEMs require for traceability and accountability.

Equally, a lot of the advanced functionality for machine vision / Radar / Lidar signal processing is very advanced, and (unfortunately) has a lot of necessary complexity. As a result it is very very costly to develop when using the former approach; yet may be involved in safety-critical functions.

This is not by any means a solved problem, and very much requires detailed management on a case-by-case basis.

Certainly testing infrastructure becomes much more important as the sensor systems that we develop become more complex. (Disclaimer: my area of interest). Indeed, my experience indicates that for sophisticated sensor systems well over 80% of the effort (measured by both in hours of development & size of the code-base) is associated with test infrastructure, and less than 20% with the software that ends up in the vehicle.

Perhaps the word "test" is a misnomer here; since the role of this infrastructure is not so much to do V&V on the completed system, but to help to develop the system's requirements -- to do the "Data Science" and analytics that are needed to understand the operating environment well enough that you can correctly specify the behaviour of the application.

Dstillery

Media 6 Degrees (the new owner of my former employer, EveryScreen Media), has changed it's name to "Dstillery".

http://www.fastcompany.com/3017495/dstillery-is-picasso-in-the-dark-art-of-digital-advertising

This Fast Company article does a remarkably good job of explaining what the technology delivers. (Disclosure: I implemented the early mobile-phone IP-tracking algorithms when I worked at EveryScreen Media).

Whilst I do believe (from what I saw) that ESM/M6D/Dstillery take privacy very seriously; and will (continue to) behave in a responsible manner, I still feel a sense of unease when I think about the extent to which participants in the advertising industry are able to peer into people's personal lives.

To balance this (sort-of) criticism, I feel I should emphasise the fact that the advertising industry is, in general, a pile 'em high sell 'em cheap kind of affair, where advertising impressions are bought and sold by the million; where performance is measured in statistical terms; and where the idea of paying close attention to any one individual would be laughed off as a ludicrous waste of time.

However, it is possible that not all participants will feel that way, just as it is possible that not all participants will be as motivated to act responsibly as my former employers were.

There has been a lot of debate recently about NSA spying on what we read on-line and using our mobile phones to track where we go in the world. Well, you don't need anything like the NSA's multi-billion dollar budgets and mastery of (de)cryptography to do something that feels (like it could be) similarly invasive. (The advertising industry is an order of magnitude less creepy, but it is facilitated by the same social & technological developments, and is still heading in a very similar direction).

I think that this is something that we should think very carefully about, and just as we seek to find better ways to regulate our security services, so too we should (carefully; deliberately; deliberatively) seek to find ways to regulate the flow of personal information around our advertising ecosystem.

--

Edit: One thing that does bear mentioning -- the link up between M6D & ESM really is a smart move. From a data point of view, it is a marriage made in heaven (a bit of a no-brainer, actually); and I think that the insights that result from combining their respective data-streams will yield genuine benefits for their clients & the brands that they represent.

Antisocial development

Human beings are social creatures; to the extent that our sanity can be undermined by isolation.

Much of our behaviour is guided and controlled by observing others around us. We instinctively imitate our peers, and follow normative standards of behaviour. As a result, I believe that the most effective learning occurs in a social context; with the presence of peers to shape and guide behaviours.

I also believe that learning is at the centre of what we do as software developers. Functioning software is seemingly more a by-product of learning than the converse.

It is therefore a great pity that modern development methods seem tailor-made to encourage developers to work alone; to minimize the need for contact with their peers and colleagues, and to reduce the need for social interaction.

For example, part of the appeal of the modern distributed version control system is that it allows individuals to work independently and alone, without requiring coordination or synchronisation.

It is possible that this has a rational basis; After all Fred Brooks' analysis in: "The Mythical Man-Month" suggests that optimally sized teams are composed of a single solitary individual, Conway's Law also seems to suggest that maximum modularity is achieved by distributed teams.

Perhaps developer solitude really is the global optimum towards which we, as an industry, are headed. However, this clearly neglects the important social aspects of human behaviour, as well as our need to learn as a group.

I wonder if we ever see a resurgence of tools that support and encourage face-to-face social interaction and learning; rather than obviate the need for it.

Managing Complexity

Quoting from Djikstra's "Go-To Considered Harmful" essay:

"My second remark is that our intellectual powers are rather geared to master static relations and that our powers to visualize processes evolving in time are relatively poorly developed."

And from Moseley and Marks' "Out of the Tar Pit" essay:

"We believe that the major contributor to this complexity in many systems is the handling of state and the burden that this adds when trying to analyse and reason about the system."

We find it difficult to reason about how system state changes over time, particularly when lots of things can modify that state. It therefore makes sense to (mentally) partition our systems into stateless and stateful parts.

If we can shovel most of the essential complexity into the stateless parts, we can make sure that this complexity is (at least) easily testable.

Equally, if we can keep the stateful parts of the program relatively free of complexity, then we can (at least) reason about (and control) how the state can change.

This argues for a system architecture that has parts that are programmed in an Object-Oriented manner as well as parts that are programmed in a pure-functional manner.

Objects enable us to make state private, and to limit and control how that state is manipulated (through public interfaces).

The use of pure functions makes it easier to reason about and test the more complex parts of system functionality.

It is when we mix statefulness and complexity that we really come unstuck.


Of course, it goes without saying that if you can avoid state you should, and if you can avoid complexity, you should ... but in my experience at least a little of both is usually required as a cost of doing business.

Google Logic - The Challenge of Agility

Here is an interesting Guardian article examining Google's management culture. It starts of by contrasting traditional long term business planning and culture with the more "agile" approach that is exemplified by Google.

http://www.guardian.co.uk/technology/2013/jul/09/google-android-reader-why

The article correctly notes that this approach to product development requires a culture that is fundamentally different from the culture and practices found at traditional (long-term-plan oriented) organizations.

These traditional cultures struggle to deal with shifting goals and a quickly changing competitive landscape.

In addition to these criticisms, I would like to add another of my own:

I do not think that these prediction & planning-oriented approaches are any good at dealing with complexity. This is particularly clear when faced with the complexity involved in the development of modern software systems.

The reasons are forehead-slappingly obvious:

Long term planning or indeed, any planning at all, relies on making predictions. Making predictions is difficult. Traditional companies respond by trying harder, by limiting the scope of what they try to predict, or by deluding themselves that they are better at it than they are. The longer your time-frame and the more complex your plan/product, the worse your predictions become. Indeed, the difficulty of making predictions increases catastrophically quickly with the complexity of your product - making planning for all but the simplest product development nothing better than a wild and irresponsible gamble. Various cognitive biases compound this difficulty, making it difficult for us to measure or even acknowledge complexity that we cannot see.

This is a very real and very serious multi-billion-dollar crisis - and it's impact extends far beyond the field of software development - although this is where it's effects are most easily and most keenly felt. How to escape from this tar-pit? How to deal with an unpredictable, complex, and fundamentally "unmanageable" world?

This, it is not easy.

It is clear that "agile/reactive" approaches are going to be a key, important part of the solution - but experience has shown that implementing an agile culture and achieving success is far from easy - it remains a long way from the turn-key exercise that we might wish it to be. Even Google, who you claim champions this approach, is hardly the shining beacon that you suggest. Good enough, perhaps, but hardly exemplary.

The agile culture needs to be able to plan to reduce the impact of adverse events, without knowing in advance what they will be. Similarly, it needs to plan to increase the impact of favourable events, again without knowing in advance what they will be. At the same time, it needs to balance this with the need to keep on top of and maintain a complex and growing inventory of capabilities and skills; technical and human capital, all of which is susceptible to the forces of entropy and decay.

It has been said that developing software (in the large) is akin to growing a plant - it requires tending & guiding as it grows, rather than planning and "engineering" (in the bridge-building sense). I imagine very much that the senior leadership of Google feel the same way.

The Next Big Thing

You might not have noticed it yet, but the leaves on the grapevine are rustling, and there is a whispering breeze in your ear. This, my friends, is the sound of the winds of change: gathering, growing, building momentum. There is a storm brewing, friends, and the game: It is afoot!

So many people: working quietly, diligently, are spear-heading that change, and so few of us realize that it is even happening, let alone comprehend the profound; monumental; world-changing implications of the technological tidal-wave that is coming our way.

Automotive industry dollars are funding this work; the dream of safer roads, of self-driving cars is driving it forwards, but this dream on four wheels is merely the tip of the iceberg: the gateway drug that leads us to an intoxicating, and perhaps slightly scary future.

What are self-driving cars but mass produced autonomous robots? Autonomous vehicles and robots will be to my daughter's generation what computers and the internet were to mine; except this change will be bigger and far more wide-reaching, as the machines step off the desktop and climb out of the server-room, and march onto the streets; as our mobile phones and tablets sprout wheels and wings and legs and arms -- as our algorithms and neural networks and distributed systems stop passively sucking in click-stream data and twitter sentiment scores, and start actively participating in a world that they can see, hear and touch.

The physical world that my parents inhabited is not so very different from the physical world that I inherited, but the world that I pass on to my daughters will change so quickly, so rapidly, that I cannot begin to imagine what it will look like when they are my age, and working away on shaping a future for their children.

What a singularly monumental change is afoot! What a wonderful, exhilarating time to be alive! And how exciting to be a part of it!


http://bits.blogs.nytimes.com/2013/07/07/disruptions-how-driverless-cars-could-reshape-cities/?smid=tw-nytimes

Thoughts on Compressed Sensing

Imagine that you have a sparsely sampled signal (with many missing measurements), and you want to recover a densely sampled version of the same signal.

If you can find a function that is capable of transforming the original signal into a representation that is also sparse (with many zero or trivially non-zero coefficients), then you have a good heuristic that you can use to find an approximation of the original signal:- maximising sparsity in the transform domain. This is really just Occam's razor - if you have good reason to suppose that the data can be fit by a simple model, then choosing the simplest model that fits the data is an optimal guessing strategy.

Good examples of such functions include the Fourier transform or various wavelet transforms - indeed, any parametric approximation or model would probably do the job.

If the transform domain really is sparse, and each coefficient in the transform domain impacts multiple measurements in the original domain, then (with high probability) you will be able to guess the "right" coefficients in the transform domain by simply 
picking the set of coefficients in the transform domain which both reproduces the original signal and minimizes the number of non-zero components in the transform domain.

However, choosing the most parsimonious model is only one kind of regularization that we can (and should) deploy. We should also think about the set of coefficients that represents the most physically and statistically plausible model.

Thus, when choosing our approach to regularization, we need to consider three things:

* Parsimony (As represented by sparsity, approximated by the L0 and L1 norms).

* Physical plausibility (A transform domain closely related to a physical model is needed here).

* Statistical plausibility (A transform domain where coefficients are as independent as possible is needed here - to reduce the number and dimensionality of the joint distributions that we need to calculate).

These three requirements are not necessarily harmonious. Does that mean we need three different transforms? How do we reconcile differences when they crop up?

Code review tool affordances

I have had a bit of experience using different code review tools: Gerrit most recently, and Atlassian's FishEye in the roles preceding that.

Strangely, the most effective code reviews that I have participated happened at Sophos, where no tool at all was used: The rule was simply to get another developer to sit next to you when you did a commit, so you could talk him through the code line by line before submitting the change.

The experience there was that most (if not all) of the problems were spotted by the original developer, not by the reviewer, who frequently lacked the contextual awareness to identify anything deeper and more significant than the usual picayune: formatting and layout errors -- the real value came from being forced to re-examine and explain your work - controlling the searchlight of your attention into a more disciplined and fine-grained search pattern than is possible alone.

The other types of review to which I have been party:- asynchronous, distributed reviews mediated by a web tool of some sort, as well as formal half-dozen-people-in-a-room-with-slides style reviews have, in my experience, proven far less effective.

So, I sit here wondering if we can rescue the asynchronous distributed code review tool, either through an alternative approach or the application of a formal and disciplined approach of some sort ... or if it is doomed to more-than-uselessness?

Dual functional & OO interfaces for program decomposition into stateless & stateful parts.

If you expose complex logic through a functional interface, it is (much) easier to test due to it's statelessness and lack of access controls.

On the other hand, parts of the program that are necessarily stateful are best handled via OO interfaces, as this allows us to limit the number of possible state transitions. Access to the stateful variables themselves is restricted (as private), and state transitions are allowed only via a limited number of (public) methods.

Management of finite resources (file handles, memory etc...) is a simple stateful part of the program, and is handled neatly by RIAA.

Types, Tensions and Risk.

The dichotomy between strongly typed languages (such as C) & dynamic scripting languages (such as Python) has already been debated & explored ad nauseam. This contribution, akin to flogging a horse long since dead and buried, will be of little interest to anybody, but it does help me clarify my own position on the matter, so here is my tuppence worth anyway:

Strong typing is supposed to buy you (the developer) a degree of protection from certain types of error. However, that protection comes at a significant cost, and really only provides a worthwhile benefit in a limited set of circumstances.

First of all you do not really reap the benefits of it *just* by using a strongly typed language. You really have to go out of your way to make use of the type system to get the benefits. You have to define your own types so that the system can enforce the non-convertibility of mutually incompatible data types. You also loose a helluva lot of flexibility, because every data item must be specifically typed.

Indeed, developers working with strongly typed languages waste a lot of time writing "generic" frameworks that do nothing else other than circumvent the type system; simultaneously losing the benefit of strong typing whilst also adding to the complexity and cognitive burden of others working on the same system.

Languages with weaker type constraints gain a *lot* of productivity simply by avoiding this whole "let's write something generic" song-and-dance routine. It also focuses the mind on alternative approaches to QA & testing. However, this additional flexibility & productivity often results in a *lot* more code being written (which must be maintained), a tendency which can quickly become self-defeating. (Sometimes what looks initially like exceptional productivity is just exceptional waste in disguise).

Strong typing, used properly, *does* provide value, under certain circumstances, in forcing decisions about data types to be made up-front, rather than deferred to a later stage. It also helps to document interfaces and to make them more specific.

It seems to me that weak typing & the generic interfaces that we end up with as a result result are most useful when we have to defer decisions about the specific data types travelling through those interfaces until later in the development process. Strong typing; leading to the development of specific interfaces are most useful when we are able to specify exactly what data types will be passing through our interfaces early on in the development process.

I.e. strong typing is more congruent with a predictive approach to risk-management, whereas weak typing is more congruent with a reactive / agile approach to risk-management. Of course, this is not a hard-and-fast rule, and successful adoption of either risk-management approach requires high levels of discipline.

Anyway, it is interesting to note the interplay between detailed technical decisions (Strong vs Weakly typed programming languages) and broader business decisions (Approach to risk management & decision scheduling).

How to engineer an Artificial Intelligence

Intelligence is a defining human characteristic, so it is somewhat doubtful that we would ever be able to pin down what it is precisely enough to make an artificial version of it ... we will always try to define and redefine the term "intelligence" in such a manner that humans remain separated and protected as a separate and distinct class of entities ... our collective ego demands no less.

However, if we put our collective ego to one side, we can surmise that intelligent behaviour consists of some general purpose learning and behaviour-directing mechanisms sitting on top of a whole heap of special-purpose sensorimotor mechanisms.

Engineering an artificial system to emulate some or all of that behaviour is a monumental engineering task. Irrespective of the fundamental breakthroughs in learning, generalisation and planning that may or may not be required, we are still left with an exceedingly large and complex software engineering challenge.

In my mind, this represents the primary obstacle to the development of a true Artificial Intelligence: Not better science, but better software engineering practices and (particularly) better software engineering tools to help manage, visualise, understand and communicate complicated software systems.

Now, this is an interesting problem, because software engineering at a large scale is much less concerned with hard technical challenges and much more concerned with "soft" human challenges of communication and politics. The same can be said for any engineering challenge where the complexity of the subject matter, combined with it's lack of visibility makes communication and documentation a key technical challenge.

Let us look at the communication challenge first: One thing that is becoming more apparent in our increasingly information-rich environment is that communication is constrained much more by the time-management and motivation-management of the reader than by the availability of the information.

In other words, it is not enough just to make information available, you have to manage the relationship between the reader and the information as well; so that information is not just dropped in front of the reader, but the relationship between the reader and the information is managed to support learning and effective decision making also.

How might one achieve this?

Procyclic

The Carnot cycle provides us with a fun little analogy that we can abuse and overextend in any number of ways:-

Energy is turned into work by harnessing and controlling a cycle of heating and cooling; expansion and contraction; increasing and decreasing pressure.

One side without the other provides only transient and worthless motion.

Brought together in opposition; made to alternate; managed and controlled in a reciprocating framework, these transient forces may be tamed and made to perform useful and beneficial work over an extended period of time.

Creative Cycles

In the same way, our innovative capacity does not produce new ideas in a single unbroken stream of creativity, but rather in fits and starts.

Contemplation and intensity must be brought together in alternating opposition:

Contemplation so that the mind may range wide, gathering inputs from disparate sources as one might pick blackberries from a brambly hedgerow.

Intensity so that the mind may delve deeply into the detail, working through the consequences of each decision; picking apart the idea to find it's essence, and turning the idea over and over to find the right terminology  the right language, the right conceptual framework within which it may be best expressed
and exploited.

One phase without the other is unproductive. Bring the two together in alternation, and you get productive work. Manage that alternation, and you can start to exploit our real potential for creativity.

Communication Cycles

There has been some discussion about the utilization and design of shared workspaces to maximise spontaneous communication (prompted by Marissa Meyer's edict banning remote working at Yahoo). In my mind, this focuses on only half of the solution: perhaps unsurprisingly, in light of the theme of this piece, I believe that communication and isolation must alternate.

When working in a totally isolated manner, one cannot determine what is needed, one does not know what is important. When sitting in the middle of a bustling open plan office, one often knows what is important, but one lacks the space and calm isolation needed to actually do anything about it. It is noteworthy that many effective organisations mandate specific periods; specific opportunities for interaction, typically involving food or drink, perhaps we could also mandate specific periods for isolation and calm introspection, a "quiet time" each day free from phone calls, emails and other forms of communication.

This may be unnecessary, however, since the ubiquitous headphone-wearing developers sitting in an open-plan-office provides a very flexible compromise, offering opportunities both for focussed work and for collaboration.

Economic Cycles

As with the small, so with the large. "The era of boom and bust economics is over" was a statement that ran not only counter to common sense, but possibly counter to some sort of law of statistical physics.

To believe that we have any real positive control over the economy is a dangerous conceit. A vague and sloppy influence, perhaps, but not control in any recognisable meaning of the word.

The presence of investment cycles for individual businesses as well as individual people is a well understood fact, as is the web of interactions that exists between those economic actors, and (so I would suppose) the presence of effects that promote the synchronisation of these individual cycles into macroeconomic feedback loops.

Why attempt to eliminate these macro-scale cycles by controlling government spending? Why not, instead, as our analogy suggests, control their amplitude and frequency so that the harm that they do is minimised.

Indeed, as Nasim Taleb points out, a small "bust" can fend off a larger and more damaging one. Targeted cyclical and perhaps aperiodic variations in the tax burden might help stress, (or threaten stress) economic actors in a manner that promotes decoupling and desynchronisation, and inhibits catastrophic modes-of-failure, but, as stated above, it would be dangerously arrogant to assume that we really know what is going on, and what the effect of our actions is, or will be.

So we have to exert control, not in a direct and forceful manner, but in a way that tends to the ecosystem, but does not seek to dominate it.

Anyway, I am straying dangerously far from my area of expertise, and my intuition oft runs astray, so let me leave things here before I go too far off the beaten track.

Order of magnitude variations in developer productivity .... for the same person!

I have been in situations where I was (very nearly) 10x more productive than anybody else in the team, as well as in situations where I was (frustratingly) considerably less productive than those around me. Looking back at the last decade or so, I can definitely see periods where my productivity dipped, as well as periods where I was able to maintain consistently outstanding results. The variance is astonishing and shocking.

Over a shorter time-scale, programming, like any other creative endeavour, has tremendous temporal performance-volatility. Performance "in the zone", when I am in a mental flow-state of high concentration is orders-of-magnitude better than when sitting in the doldrums, unable to perceive or engage with the natural semantics of the problem domain. Writers block (to an extent) happens to programmers, too.

There are a number of reasons for these variations:

Firstly, over the long term, sampling effects play a part in (relative) performance. You can expect the quality and dedication of the team that you are working with to vary significantly, so as you move from team to team your baseline for comparison swings all over the place.

Secondly, experience and tools. You are never going to perform as well when you are learning as you are going - I am easily an order of magnitude more productive when using tools with which I am familiar than when trying to learn something new. (But, as per the technologist's typical Catch-22, you need to always be learning something new to stay relevant)...

Thirdly, personal circumstances:- commuting distance, family obligations, (small children), illness, living conditions. All of these have an impact on day-to-day mental alertness and ability to get "into the zone", although perhaps to a lesser extent than the other factors in this list.

Fourth, team dynamics. Some of my highest levels of productivity have been when operating as part of creative, collaborative partnerships, with another highly engaged team member to bounce ideas off, and to debate the merits of various approaches. This produces a creative dynamism that both improves the quality of the end product, and promotes active engagement in the process by both members of the "dynamic duo".

Finally, the big one: enthusiasm and engagement. This is really about organizational dynamics, leadership and psychology and is perhaps the hardest aspect to understand and control. For a programming task where attention to detail and mental engagement with complex systems is critical, the level of enthusiasm and engagement in the problem domain is critical for performance. In those roles where I have "lived the job", and spent every waking moment turning the problem-at-hand around in my head, dreaming about it when I fall asleep at night, I have obviously and significantly outperformed, in comparison to those roles where the job feels like an endless (and pointless) grind with no end in sight. You have to believe in the mission to perform, and that is as much an (external) function of leadership as your own (internal) reserves of fortitude and grit.

In summary, the biggest effect on (long term) performance is probably the existence of total, absolute and life-consuming dedication to the task at hand, as it promotes rapid learning and extended periods in flow-state. Inspirational leadership and creative partnerships do an enormous amount to encourage and support this level of engagement in the work environment, whilst suitable domain expertise and knowledge helps to reduce frustrations and remove barriers to progress. Finally, an absence of confounding factors and distractions in the out-of-work environment also helps.

So, a lot of it is how the job, the organization and the developer fit together. The 10x thing is not about innate skill (except in a statistical sense). As a leader, there are definitely a large number of things that you can do to increase the probability that members of your team will perform at their peak.

On Commuting

Dont

On the importance of attention-to-detail in software tools development.

As the Apple example demonstrates, attention to detail, in aggregate, results in a superior product, which enables you to justify charging a premium price.

This truism is particularly apt when it comes to tools, because the affordances and micro-features that one's day-to-day tools offer have a huge impact on the way that we do our day-to-day job. The extent to which we developers follow the path of least resistance is, in many ways, quite sad, but this most human of characteristics places incredible power in the hands of the tool-makers. Best practice is often defined by the tools that we use. As a crude example, witness how Hungarian notation fell out of favour as soon as popular IDEs started to display "type hints" when the mouse hovered over the variable.

Make it slightly easier to do something, or to arrange text in a particular way, and developers will change their behaviour in response.

For many years (2001-2011) I used TextPad on windows, and came to particularly appreciate it's block select mode when managing vertically aligned characters. The shortcut for this feature (Ctrl+Q, B) has been burned irreversibly into my motor cortex. I used this feature not only to manage vertically-aligned assignment statements, but also to quickly add blocks of end-of-line comments, and to support a declarative programming style that relied heavily on in-source tables of parameters & other data.

This declarative approach to software construction had a particularly pleasing synergy with the MATLAB language and the uncomplicated, linear loop-free aesthetic which is made possible by it's impressive collection of array-oriented library functions.

More recently, (2011-2013) I have been using Sublime Text as my editor-of-choice. (As well as a little vim, when I have no choice). Whilst the multiple selection feature of Sublime text is super-duper-awesome in it's own way, I still miss TextPad's block select feature, particularly it's ability to automatically insert whitespace at the end of lines, allowing one to easily reclaim blocks of space to the right of the logic for descriptive commentary. (80 column limits be damned as an onions-in-the-varnish anachronistic throwback).

Anyway, seemingly small things, but still important: What seem like trivial affordances really do shape the way that one thinks and works. For the past year I switched away from Textpad + MATLAB towards Sublime Text & vim + Python, and the combined strictures of PEP8 & PyLint, together with the subtly different affordances of the Sublime Text vs. TextPad steered my development style away from the declarative parameter-driven approach towards a more "traditional" programming style, to the moderate detriment of the "quality" of the applications that I produced.

Despite Python's numerous, obvious and justifiable claims to superiority, there exists an ineffable and emergent quality that language and tools together supply, and a quality that was not (quite) captured in the tools that I was using for the past year.

Another observation: When moving the caret quickly along a line of text using the arrow keys with the "ctrl" key pressed down, the cursor in the MATLAB editor stops very slightly more frequently than other editors, resulting in a slightly "sluggish" feel to the navigation. Paying attention to matters such as this is the equivalent, to a tool-maker, of the auto-maker paying attention to the "thunk" noise that the car door makes when it closes. A feature that operates almost on a subliminal level to create the impression of quality, and, to the tool-user, the feeling of "floating" through the text file making changes, rather than the feeling of wading through a muddy swamp.

Another observation: Little bits of automation to make life easier. When I highlight a word in Sublime Text and press the open-parentheses "(" key, the word is surrounded in parentheses. When I do the same in the MATLAB editor, the word is deleted and replaced by the open parenthesis character.

I am a craftsman. I want to love my tools. MATLAB has elements of greatness, but a large number of flaws as well. Some of these flaws cannot be addressed without making the tool something other than what it is.

Fair enough.

But some can be addressed with passion, attention-to-detail, design sense, and strong, detailed technical leadership.

Confidence and Competence and Information Overload

The internet is a wonderful thing.

Every problem has a thousand solutions: documented; discussed; dissected.

This sumptuous surfeit of suitable solutions seems satisfactory; superb even, save for one small snag:

My effective competence, my ability to work quickly and solve problems, is driven largely by my confidence in my own abilities, my own evaluation of the completeness of my knowledge versus the task at hand.

When I was younger, the arrogance of youth made this easy. I plowed on ahead, ignorant of my own ignorance, I performed, gained accolades, and life was good.

However, faced with ever rising flood-waters issuing from the fire-hose of information that is the internet today, it is all too easy to start comparing what I know with what I could know, which, expressed as a ratio, is always going to be depressingly close to zero. With self-confidence thus undermined, the inevitable consequence of this is for me to loose sight of the fact that I have enough information at hand to get on with it, without worry or concern, and to get dragged down into the rabbit-hole.

This problem is compounded when one tries to pick up a new tool or technology, an activity that (given the pace of technological advance) one needs to do more-or-less continuously.

Now, I love learning, but I cannot resist the temptation to compare my knowledge and level of expertise in the latest tool to my knowledge and level of expertise in the last, and, in the worst-case-scenario one or two weeks worth of experience is never going to compare favourably with six or seven years.

In situations such as this, I have to keep telling myself to HTFU and demilquetoastify my attitude.

Batteries Included

Software developers seem to be significantly more productive today than they were even a few years ago. What has happened? Have techniques improved? Have skill levels increased? Are best-practices being more closely adhered to?

One cannot easily isolate any one factor as bearing sole responsibility for the performance improvements, but I suspect that a great deal of our gratitude should be directed towards improvements in the availability, quality and performance of software libraries, together with growing communities around those libraries, promoting education and adoption in the wider engineering community.

After all, the one thing that makes software development so different from other disciplines are the tremendous cost savings available through reuse. The less logic your organization is responsible for, the less the expenditure on development, documentation and maintenance. The main expense incurred when using tools and libraries is in education and training ... this is perhaps not something that we as an industry have fully adjusted to:- neither in the way that we (as professionals) specialize; nor in the way that we (as organizations) recruit new talent or nurture existing talent.

In fact, the notion of talent as something that is exclusively innate and inimitable is a dangerous one. When one makes an experienced hire, the commodity that one is purchasing is (primarily) knowledge, rather than the commission of work/effort to be expended. The knowledge that is required to not have to do the work, to be more precise. Perhaps we need to introduce new work and employment patterns beyond the traditional notions of contractor and perm employee?

Personal Statement

My passion for machine vision, machine learning and statistical pattern recognition is longstanding, having started over 10 years ago and continuing today.

Most of my undergraduate AI degree was oriented towards logic, theorem proving, and computational linguistics, which was fascinating in it's own right, but did not strike me as a particularly realistic or pragmatic way of dealing with the messiness and complexity of the real world. As a result, I latched on to the (at the time) less mainstream "soft" computing approaches with enthusiasm, devouring the content of the machine learning, machine vision and neural networks modules avidly. I saw these approaches as a pragmatic alternative to the hard and inflexible grammar-based approaches to Natural Language Processing espoused by the main body of the department.

This view of machine learning as a pragmatic tool, at odds with ivory-tower academicism has stuck with me ever since, even as the subject has become more mainstream (and more academic and mathematically sophisticated). As a result, I tend to focus on simple techniques that work, rather than techniques which demonstrate mathematical chops and academic sophistication. I am fortunate in this regard, because, paradoxically, the solutions to difficult problems are often conceptually simpler and mathematically less sophisticated than the "optimum" solutions to simple problems. Perhaps a little bit of the Yorkshire/Lancashire culture of engineering pragmatism rubbed off on me during my time in Manchester.

Another thing that was dawning on me as I finished my undergraduate degree was the importance of scale. As I attempted to find datasets for my hobby projects, (Far harder back then than today), I began to develop suspicions that scale, rather than any qualitative leap in understanding, was going to be a key factor in the development of genuinely interesting artificial intelligence techniques. From this came my interest in machine vision, which I saw as a key "gateway" technique for the collection of data -- to help the machine build an understanding of the world around it and to "bootstrap" itself to a more interesting level.

I was lucky with my first employer, Cambridge Research Systems, where I had the opportunity to work with some very talented people, both within the company and across our customer community. From that experience, and the abortive neuroscience PhD that I started, I learned a lot about the neuroscience of biological visual systems, particularly the older, lower-level pathways that go, not to the primary visual cortex, but to the evolutionarily older "reptilian" parts of the brainstem. In contrast with the "general purpose" and "reconfigurable" nature of the cortex, these older pathways consist of a large number of (less flexible) special-purpose circuits handling things like eye movements and attention-directing mechanisms. Crucially, these lower-level circuits enable our visual system to stabilise, normalise and "clean" the data that we present to our higher-level cortical mechanisms. This insight crosses across well to more commercial work, where the importance of solid groundwork (data quality, normalization and sampling) can make or break a machine learning implementation. I was also fortunate enough to pick up some signal processing and FIR filter design fundamentals - as I was writing software to process biological time-series signals (EOG) to identify and isolate events like saccades and blinks.

At around about this time, I was starting to become aware of the second important thread in my intellectual development: The incredible slowness of software development, and the difficulty and cost that we would incur trying to implement the large number of these lower level stabilization mechanisms that would be required.

I left Cambridge Research Systems specifically to broaden my real-world, commercial software development experience, working at a larger scale than was possible at CRS. Again, I was lucky to find a role with Sophos, where I learned a great deal from a large group of very talented C++ developers doing Test Driven Development in the highest-functioning Agile team I have yet encountered. Here, I started to think seriously about the role of communication and human factors in software development, as well as the role that tools play in guiding development culture, always with an eye to how we might go about developing those special purpose data processing functions.

Following a relocation closer to London (for family reasons), I left Sophos and started working for Thales Optronics. Again fortunate, I found myself working on (very) large scale machine vision applications. Here, during a joyous three year period, I was able to put much of my previous intellectual development and thinking into practice, developing not only the in-flight signal processing, tracking and classification algorithms, but more significantly, the petabyte-scale data handling systems needed to train, test and gain confidence in them. In addition to the technical work, I worked to encourage a development culture conducive to the development of complex systems. This was the most significant, successful and rewarding role I have had to date.

Unfortunately, budgetary constraints led Thales to close their office in Staines, and rather than transferring to the new office, I chose to "jump ship" and join Fidelity Asset Managers in the City of London, partly in an attempt to defeat some budgetary constraints of my own, and partly out of an awareness of the potential non-transferability of defense industry expertise, made more pressing by an impending overseas relocation.

At Fidelity, I used my knowledge of the MATLAB distributed computing toolbox to act as the High-Performance Computing expert in the "quant" team. I gained exposure to a very different development culture, and learned a lot about asset management and quantitative investing, gaining some insight into the accounting and management factors that drive development culture in other organizations. I particularly valued my exposure to the insights that Fidelity had, as an institutional investor, into what makes a successful organization, as well as it's attempts to apply those insights to itself.

Finally, in 2011, my family's long expected overseas posting came. Yet again we were incredibly lucky, and got to spend a wonderful year-and-a-half living in the middle of New York city. I was fortunate, and managed to get a job at an incredible Silicon Alley startup, EveryScreen Media, which was riding the wave of interest in mobile advertising that was just beginning to ramp up in 2011 and 2012. Again, finding myself working with incredibly talented and passionate colleagues, I was given the opportunity to broaden my skills once again, picking up Python and Unix development skills, becoming immersed in (back-end) web development, building out the data science infrastructure in an early-stage startup. From this year and a half or so, I particularly value what I learned about how to develop large scale, (scaleable) distributed real-time data processing systems systems and the effective use use of modern internet "web" technology.

Now, back in the UK, I am in search of the next step on my journey of learning and discovery. My focus is, and remains, on the pragmatics of developing complex statistical data processing systems, on how to create and curate large data-sets, how to integrate them into the development process, so that continuous integration and continuous testing, visualisation and monitoring help the development team to understand and communicate the system that they are building, as well as the data that feeds it; to respond to unexpected behaviors, and to steer the product and the project to success and, moreover, to help ensure that the organization remains rightly confident in the team and in the system.

A Network Model for Interpersonal Communication

Modeling interpersonal communication within an organization as a network of reconfigurable topology composed of high capacity data stores connected by limited bandwidth communication channels.


The Model:


The amount that we know on any given topic of interest vastly outweighs our practical ability to communicate that information in a reasonable time-frame. We simply do not have the time or the available bandwidth to communicate everything that we need to in the detail that the subject deserves. Our model reflects this - The data storage capacity at each node is immense, and contrasts sharply with the exceedingly limited bandwidth available for communication between nodes. The difference between the two is many orders of magnitude in size. For a visual analogy, we should not look to buckets connected by hosepipes, but rather half-million ton supertankers connected by thin cocktail straws. Transmitting even a gallon of knowledge is a challenge.


Chinese Whispers:

When communicating with distant nodes with messages routed through intermediary nodes, the information being transmitted is compressed to an incredibly high degree with a very lossy and low-quality compression algorithm. The poor quality of the communications channel is particularly evident when the network encompasses a diverse range of backgrounds, cultures and terminological-linguistic subtypes. In many such cases the intent of the message can easily be inverted as relevant details are either dropped or misinterpreted in transmission.


Systematic Factors impacting efficacy of communication:

The options available for compression are greater when two neighboring nodes already have a great deal in common, where shared datasets, terminology, mental models, and approaches to communication can be used to elide parts of the message, reducing bandwidth requirements, and allowing for communication that is both more reliable and more rapid. As a result of this, communication within an organization that has a strong, unified "culture" (common knowledge, terminology and practices) will be far more effective than communication within an organization that has a less cohesive "culture", purely because the options for message compression are greater, irrespective of any other measures that the organization might put in place to improve the available bandwidth. It is worth noting that, whilst this does improve the situation considerably, the problem itself is fundamental and always presents a significant challenge.


Organizational Optimization for Effective Communication:

Given that there is nothing inherently fixed about the topology of the communications network within which we are embedded, one simple response to this problem is to remove intermediary steps from source to destination nodes, and to allow the source node to connect directly with the destination node, permitting a direct and relatively high bandwidth exchange. This is even more effective if the exchange is bidirectional, permitting in-situ error correction. This argues for a collaborative approach to organization - with technical experts communicating with one another directly, with no intermediary communications or management specialists.

Another approach is to optimize for effective compression of the messages being transmitted through the network. As noted above, this relies on a common terminology, a common knowledge base, and a common set of practices and approaches to problems. In other words, most of the communication is moved out-of-band -- communicated though various channels prior to the point where it is actually needed. Again, this is well aligned with the collaborative approach to organization - where the technical experts within an organization continually educate one another on their own area of expertise so that when they need to communicate quickly, they can do so both effectively and reliably.


A Common Antipattern: Spin Doctors and Office Politicians:

Of course, the approaches outlined above are not the only solution to this fundamental problem. However, I argue that at least some of these approaches are anti-patterns, detrimental to the long term health and development of the organization.

Many individuals craft the messages they send extremely carefully to minimize the probability of corruption or misinterpretation en route, partly by reducing the information content of the message, and partly by crafting the emotional tone to remove ambiguity. This process is colloquially known as "spin", or "crafting the message".

In some situations this approach may even be appropriate, for example: where the network is large and cannot be reconfigured so that messages must be transmitted over more than one "hop"; where the environment for communication is particularly disadvantageous, with little common culture, terminology, or background technical knowledge; and finally, where long-term systematic improvements must be subordinated to short-term goals.

However, there are a couple of significant drawbacks to this approach. Firstly, by restricting knowledge transfer, opportunities to grow a base of common knowledge and understanding are squandered. Secondly, this approach is in direct conflict with cultural norms that emphasize honesty and transparency in interpersonal communication, the adherence to which builds a basis of trust and mutual understanding that further enhances communication.

Dialectic vs. Tribalism .... Fight!

One of the problems that we have is that there is a very human instinct to stereotype and denigrate any (real or imagined) opposition community. It is part of our tribalistic nature, and it infects every sphere of human thought, even when we are not aware of it. (Especially when we are not aware of it.)

The truth of the matter is that other people are not simple and they are not stupid. They may be preoccupied with other battles, but they are generally smarter and more complex than we are willing to give credit for.

Fortunately, we have centuries of philosophical learning to help us tackle this behavioral bias. Unfortunately we do not have the time to absorb all of this scholarship  so we will just jump on the whole "Hegelian" [sic] thesis-antithesis-synthesis meme as a quick-and-dirty fix for our ignorance.

We have to discard our tribalistic myopia and become aware of the other battles that people are fighting - the preoccupations that they are focused on, and within which they frame their arguments and their notions of right and wrong. Any synthesis solution will incorporate elements of these other concerns:- "In such-and-such a situation, you need to worry about risk A, and take action B, in this other situation, you need to worry about risk C and take action D."

In other words, a sign of maturity in the debate over a discipline is the presence of increasingly fine-grained recipe books, within which increasingly tightly defined specialist sub-disciplines emerge with their own concerns and heuristics. The division of labour becomes ever more fine-grained as the economy around a discipline matures and grows.