The One-Dollar Haircut

Every year, I travel more than 9K miles to my native place in Tamil Nadu, India, to get my $1 haircut.

(pic source: http://jhingu.com)

The local barber is an expert and does it the old-fashioned way. A comb to sift, and super-quick scissors acting as efficient cutting planes to produce a nice, convex-hull hairdo. No fancy machinery, seating, lighting, or A/C in his environment-friendly shop, and not a single nick, or hair out of place when he's done. This keeps his customers happy and operating cost low. It's a dollar regardless of crop density. Perhaps the argument is that for the sparse-headed, the cost of searching goes up even as the actual cutting time reduces. If the shop is full, spill-over customers can sit in makeshift chairs outside the shop, and sip chai from the adjoining 'tea-kadai'.

The Indian barber shop is also an instance of the 'elastic' capacity model that is key to understanding how the Indian economy has chugged along. Here is what Prof. R. Vaidyanathan, Professor at the Indian Institute of Management, Bangalore, in his superb book 'India, Uninc' has to say:

Data shows that it this 'unincorporated' economy built from the ground-up by entrepreneurs that is responsible for much of India's GDP and employment, and not the stock market that grabs headlines. Despite the recent crash of the global markets, the Indian elephant is likely to remain solid, like it did after 2008.

The traditional Indian way of doing business may be initially bewildering to the external observer, which may have led to this 'Uninc' being categorized by Nehruvian India's west-educated officials as "unorganized". Actually, it is anything but, and appears to be based on balancing constraint-enforcing 'order' and constraint-relaxing 'chaos'. This traditional Indian approach, which I personally view as dharmic optimization, yields astonishingly high and sustainable levels of quality and efficiency when done right, but can also be disastrous when either order or chaos is excessive.

End Note: Apparently, the global wig supply-chain sources much of its hair from India.

Finding an optimal parking spot for shopping

Here is the original discussion at 'Punk Rock OR'.  The blog below studies a parking objective in the shopping context, and where the entrance to the shopping complex and the exit are at different locations. The cost to be minimized consists of at least two components:
(objective-a) the 'work' done searching for a good parking spot, as well as
(objective-b) the total physical work done after parking. This includes the work done while walking to the entrance, as well as the work done after shopping and exiting the building. Let's focus on objective-b here, given that objective-a has been analyzed before, using the following notation.

Parking coordinates (O, D), where
O = walking distance from the parking spot to the entrance, and
D = walking distance from the exit to the parking spot.
M = mass of the person,
S = expected mass of purchased items, and
(μ, g) = constants denoting the coefficient of friction during walking, and the acceleration due to gravity, respectively.

Work done = force * distance = μg * mass * distance (A prior blog shows how to optimize our in-store shopping route.)

Objective is to minimize: M*O + (M+S)*D

A minimal work objective suggests these simple rules:
1) Heavy shopping: park close to the exit. 
2) Light shopping: park to minimize total walk.

Thus, the goodness of a parking spot depends on the shopping context. Let us figure out a good parking spot for the shopping scenario below.

Linear Store, Manhattan Distances
Consider a long, 'linear' store along the x-axis, with the exit @ x = 0, and entrance @ x = L. Assume 'Manhattan' walking distances (ℓ¹) and that prior customers selected spots nearest the building along y-axis, and picked their spots along the x-axis based on individual preferences. Given this, the distance we can expect to walk to/from the store along the y-axis is already (near) minimal. This leaves us with a (non-convex) one-dimensional search for open locations x(i) along the x-axis over three regions. Let us split this task into two steps. 

Step-1: Find the best spot within each region
A) x(i) = -a ≤ 0 
Minimize M(O+D)+S*D = M(L+a +a) + Sa = ML + (2M+S)a*

B) 0 ≤ x(i) = b ≤ L
Minimize M(L-b + b) + Sb = ML +Sb*

C)  x(i) = L+c ≥ L
Minimize M(c+L+c) + S(L+c) = ML+SL + (2M+S)c*

The optimal x* within each of these regions is the one closest to the exit. This is independent of S and L, and identifiable by greedy search.

Step-2: Finding the least cost spot among (a*, b*, c*)
The worst spot in B is no worse than c*, independent of S and L, yielding a binary choice: pick a* or b*.
x(i) ≤ 0: incremental cost = (2M+S)a*, a* ≥ 0
x(i) ≥ 0: incremental cost = Sb*, b* ≥ 0

For small values of S, b* dominates and is optimal for light shopping or window shoppers. As S increases and becomes comparable to M, b* is preferable when:
b*/a*  ≤ 1 +2(M/S)

Even if we 'shop till we drop' and carry our own weight, b* can be as much as 3a* and still dominate.

Practically speaking, b* is a solid bet, but when B is full, the shopper is forced to choose between disjoint regions A and C.
x(i) ≤ 0: incremental cost = (2M+S)a*
x(i) ≥ L: incremental cost = SL + (2M+S)c*

shoppers would prefer a* to c* unless:

(a*-c*) > L/[2(M/S)+1]

Scenarios:
i) Between spots equidistant from the entrance and exit, a* dominates c*, independent of S and L. 
ii) For the 'carry our own weight' scenario, a* is preferable when
(a*-c*) ≤  L/3
iii) For unmanageably large S, a* remains preferable if:
(a*-c*) ≤ L

We can now draw some conclusions.

Finding a Minimum-Work Parking Spot
Preferred order of parking is:
b* > a* > c*

which is out of order. The actual search pattern we employ may depend on whether the aisles in the parking lot are along the x- or y-axis. 

Light/medium shopping context
Clockwise, starting from the exit, scan B, park if feasible. Else scan C, and park if feasible. Else, turn around choose a spot in A. 

Heavy shopping context
Clockwise, starting from the exit, scan B, park if feasible. Else turn around, scan A, and park if feasible. Else, choose a spot in C.

Hunting for Optimality

OK. Sometimes, we really do need to find 'the' answer. Recall the scene from Jaws: "It's a shark, not the shark". But in general, seeking 'the' one and only, true, unique, prophetic optimal solution to real-life decision problems is an exercise in futility. It makes sense to talk about the best practically achievable solutions given resource constraints, costs, and priorities, which can then be iteratively refined over time. While filling out a healthcare quiz about better nutrition and weight management, I was delighted to see this question.

The health-care researcher who designed this quiz deserves a round of applause.

But how do Operations Researchers find such solutions? Sometimes, the problems are provably cute (journals love to publish these).

But very often in practice, you end up dealing with this:

No fear. Well-designed practical algorithmic approaches employing robust, industrial-strength optimization tools like CPLEX can help OR'ers recover high quality solutions to challenging, seemingly intractable, messy-data riddled, real-world decision problems.

Losing out on the perfectly good in a quest for the perfect seems pretty silly.

The Contextual Optimization of Sanskrit

I'd written a blog for the INFORMS conference earlier this month based on my practice perspective, which emphasized the importance of contextual optimization rather than despairing over the 'not infallible' theoretical worst-case nature of certain mathematical problems. This is something well-internalized by those in the large-scale and non-convex optimization community, where 'NP-Hardness' is often the start, rather than the end point of R&D.

The wonderful 'Philip McCord Morse Lecture' at the recently concluded INFORMS conference in San Francisco by Prof. Dimitris Bertsimas of MIT touched upon this point, and the 'conclusions' slide in the talk explained this idea really well. To paraphrase, 'tractability of a problem class is tied to the context - whether the instances you actually encounter can be solved well enough'. I mentioned the Sulba Sutras in that blog - a well known body of work that epitomizes the Indian approach to mathematics as Ganitha - the science of computation. The genius, Srinivasa Ramanujan, was a relatively recent and famous example of a mathematician hailing from this tradition. The Indian approach is often algorithmic and more about rule generation than infallible theorem proving. Not that Indians shied away from proof ('Pramaana'. For example, see 'Yuktibhasa'). As I understand it, this sequential process of discovery and refinement does not lose sleep over theoretical fallibility, and consists of:
a) in-depth empirical observation of, and a deep meditation on facts,
b) insightful rule generation,
c) iterative, data-driven refinement of rules.
This quintessential Indian approach is applied not just to math, but to practically every field of human activity, including economics, commerce, art, medicine, law, ethics, and the diverse dharmic religions of India, including Hinduism and Buddhism. Panini's Sanskrit is a great example of this approach.

Panini, the famous Sanskrit grammarian (along with Patanjali) is perhaps the most influential human that much of the world does not know much about. His fundamental contributions to linguistics more than 2000 years ago continues to transform the world in many ways even today. Noted Indian commentator, Rajeev Srinivasan, has recently penned a wonderful article on Panini and Sanskrit. You can learn more about Panini's works by reading Dr. Subhash Kak's (OK State Univ) research papers (samples are here and here). This blog was in part, triggered by this article, and talks about Sanskrit and its contextual optimizations.

Abstract: Sanskrit recognizes the importance of context. Two examples that show how Sanskrit is optimized depending on the context, in two entirely opposite directions, is shown below.

Optimization-1. The grammar is designed to be entirely context-free as Rajeev Srinivasan's article explains, and anticipated the 'grammar' of today's high-level computing language by more than 2000 years: precise with zero room for ambiguity of nominal meaning. To the best of my knowledge, punctuation marks are not required, and order of the words can be switched without breaking down, although there may be personal preferences for some orders over the others, and the sentence remains unambiguously correct. An optimization goal here therefore is to generate a minimum (necessary and sufficient) number of rules that result in an maximally error-free production and management of a maximal number possible variations of Sanskrit text. In this case, Panini appears to have achieved the ultimate goal of generating a minimal set of rules that will produce error-free text, forever. There are other well-known optimizations hidden in the structure and order of the Sanskrit alphabet - more on that later.

Optimization-2. The final interpretation of many keywords in Sanskrit ARE contextual. Which means there are multiple, related interpretations for some words that have a nominal/generic meaning, but you have to optimize the final interpretation at run-time by examining the context of usage, to recover the most suitable specific choice. If the first optimization helped eliminate fallibility, this second optimization in a sense re-introduces a limited fallibility and a degree of uncertainty and freedom by design! This feature has encouraged me to reflect (recall Ganesha and Veda Vyasa), develop a situational awareness while reading, pay attention to the vibrations of the words, and grasp the context of keywords employed, rather than mechanically parse words and process sentences in isolation. A thoughtful Sanskrit reader who recognizes this subtle optimization comes away with a deeper understanding.  For example, Rajiv Malhotra, in his book 'Being Different' (now in the top-10 list of Amazon's books on metaphysics) gives us the example of 'Lingam'. This can mean 'sign', 'spot', 'token', 'emblem', 'badge', etc, depending on the context. Apparently, there are at least 16 alternatives usages for 'Lingam' of which one best suits a given context is picked, and not simply selected at random. And of course, the thousand contextual names ('Sahasranamam') for Vishnu is well known in India. Some well-known western and Indian 'indologists' have ended up producing erroneous, and often tragic translations of Sanskrit text either because they failed to recognize this second optimization, or because they misused this scope for optimization to choose a silly interpretation, leading to comic or tragic conclusions.

Again, this contextual optimization approach by the ancient Indians is not just restricted to Sanskrit, but is employed gainfully in many areas, including classical arts, management, healthcare, ethics, etc., and of course dharmic religion. This contextual dharmic optimization has perhaps helped India in getting the best out of its diverse society, as well as keep its Sanskriti refreshed and refined over time. For example, the contextual ethics of dharma (ref: Rajiv Malhotra's book) has a universal pole as well as a contextual pole that allows the decision maker faced with a dilemma, to not blindly follow some hard-wired ideological copybook, but contemplate and wisely optimize his/her 'run-time' choice based on the context, such that himsa is minimized (dharma is maximally satisfied). Some posts in this space has tried to explore the applications of this idea'.

An earlier blog discussed a related example of seemingly opposite goals for contextual optimizations. When it came to mathematical algorithms, data, and linguistic rules in Sanskrit, a goal was to be brief and dense, minimize redundancy, and maximize data compression, so that for example, an entire Sine-value table or generating the first N decimals of Pi can be both encoded and decompressed elegantly using terse verse. Panini's 'Iko Yan Aci' in the Siva Sutras is a famous example of a super-terse linguistic rule. On the other hand, when it comes to preserving long-term recall and accuracy of transmission of Sanskrit word meanings as well as the precise vibrations of mantras (e.g. Vedic chants) that are critically linked to the 'embodied knowing' tradition of India, the aim appears to be one of re-introducing controlled data redundancy to maximize recall-ability, and error-reduction. This optimization enabled Sanskrit mantras to be accurately transmitted orally over thousands of years.

To summarize, contextual optimization is a powerful and universal dharmic approach that has been employed wisely by our Rishis, Acharyas, Gurus, and thinkers over centuries to help us communicate better, be more productive, healthier, creative, empathetic, scientific, ethical, and interact harmoniously with mutual respect.

update 11/22/2014: 'optimize the final interpretation at parse-time / read-time' is the intent for optimization-2, rather than the computer-science notion of 'interpretation at run time'.

Lassoing the Exponential

An abbreviated version was blogged for the INFORMS 2014 annual conference as 'Not Particularly Hard'.

-------------------------------------

While working on a new retail optimization problem a few weeks earlier, a colleague was a bit disappointed that it turned out to be NP-Hard. Does that make the work unpublishable? I don't know know, but unsolvable? No. The celebrated Traveling Salesman Problem (TSP) is known to be a difficult problem, yet Operations Researchers continue to solve incredibly large TSP instances to proven near-global optimality, and we routinely manage small TSP instances every time we lace our shoes. Why did I bring up laces? In a moment ...

Hundreds of problems that are known to be difficult are 'solved' routinely in industrial applications. In this practical context it matters relatively less what the theoretical worst-case result is, as long as the real-life instances that show up can be managed well enough, and invariably, the answer to this latter question is a resounding YES. The worst-case exponential but elegant 'Simplex method' continues to be a core algorithm in modern-day optimization software packages. 

This issue of contextual optimization is not a new one. For some ancient people who first came across 'irrational' numbers, it was apparently a moment of uneasiness: how to 'exactly' measure quantities that were seemingly beyond rational thought. For some others, it was not much of an issue. Indeed, there is an entire body of Ganitha (the science of calculations, or mathematics) work in Sanskrit, the 'Sulba Sutras', almost 3000 years old, where irrational numbers show up without much ado. 'Sulba' means rope or lace or cord. If we want to calculate the circumference of a circle of radius r, we can simply use (2πr) along with an approximation for 'π' that is optimally accurate, i.e., good enough in the context of our application. If we we did not have a good enough value for π, we could literally get around the problem: simply draw a circle of radius r, and line up a Sulba along its circumference to get our answer. For really large circles, we can use a scaled model instead of ordering many miles of Sulba. Not particularly hard. Encountering a really difficult optimization problem can be a positive thing, depending on how we respond to it.  Often, there are alternative approaches to business problems that at first glance, appear to have  insufficient data: this tempts us to throw in the towel and send the problem back to the customer and say "infeasible" or "unbounded". Instead, we can use a Sulba and Lasso our decision variables. This could well be an ORMS motto:

"When the going gets NP-Hard, Operations Researchers get going" :)

The Best Decisions are Optimally Delayed

The lessons learned from the last few years of practice have convinced me that analytics and OR (OR = Operations Research), or at least MyOR is mainly about learning the art and science of engineering an optimally delayed response. Good analytics produces an optimally delayed response.

But why introduce a delay in the first place? Isn't faster always better? 'Science of better' does not always mean 'science of faster'. From age-old proverbs we find that in between 'haste makes waste or knee-jerk reaction' and being 'too clever by half' lies 'look before you leap'. If we view Einstein from an OR perspective: "Make things as simple as possible, but not simpler", the reason seems clearer. We must make situation-aware and contextually-optimal decisions as fast as possible, or as slow as necessary, but not faster, or slower, i.e., there exists a nonzero optimal delay for every response decision. A middle path in between a quick-and-shallow suboptimal answer, or a slow-and-unwieldy 'over-optimized' recipe. Of course, one must work hard during this delay to maximize the impact of the response, and put Parkinson's law to good use, as suggested below:

#orms methods often *delay* product response to find superior answers. Tricky to pick optimal (delay, improvement) pair :)
— (@dualnoise) August 1, 2014

See this old post on 'optimally delaying an apology' to maximize benefit to the recipient, or recall the best players of every sport being able to delay their response by those few milliseconds to produce moments of magic, or Gen. Eisenhower delaying the call to launch D-Day. In the same way, a good OR/analytics practitioner will instinctively seek an optimal delay. For an example of this idea within an industrial setting, read this excellent article by IBM Distinguished Engineer J. F. Puget on taxicab dispatching that he shared in response to the above tweet. Implication: If your analytics system is responding faster than necessary, then slow it down a bit to identify smarter decision options. The 'slower' version of this statement is more obvious and is a widely used elevator pitch to sell high-performance analytics and optimization tools.

The history of the 'optimal delay' is many thousand years old, going back to the writing of the world's longest dharmic poem, the Mahabharata, which also includes within it, the Bhagavad Gita, one of the many sacred texts of Hinduism.

(pic link: http://www.indianetzone.com)

The story about how this epic came to be written is as follows:
Krishna Dvaipana (the Veda Vyasa) wanted a fast but error-free encoding of the epic that could be told and re-told to generations thereafter. The only feasible candidate for this task was the elephant-faced and much beloved Ganesha, the Indian god of wisdom and knowledge, and remover of obstacles. The clever Ganesha agreed to be the amanuensis for this massive project on the condition that he must never be delayed by the narrator, and must be able to complete the entire epic in ceaseless flow. Veda Vyasa accepted but had his own counter-condition: Ganesha should first grasp the meaning of what he was writing down. This resulted in a brilliant equilibrium.

Veda Vyasa composed the Mahabharata in beautiful, often dense verse that Ganesha had to decipher and comprehend even as he was writing it down without lifting the piece of his tusk that he had broken off to inscribe, from the palm leaves. If Ganesha was too slow, it would potentially give Vyasa the opportunity to increase the density and frequency of incoming verses that may overload even his divine cognitive rate. If he went too fast, he would risk violating Vyasa's constraint. Similarly, if Vyasa was too slow, he would violate Ganesha's constraint. If he went too fast, his verse would lose density and risk becoming error-prone, and of course, then Ganesha would not have to think much and perhaps write it down even faster. Imagine if you will, a Poisson arrival of verses from Vyasa divinely balanced by the exponentially distributed comprehension times of Ganesha. Writer and composer optimally delayed each other to produce the greatest integral epic filled with wisdom ever known; written ceaselessly in spell-binding Sanskrit verse, without error, and flowing ceaselessly to this day without pause.

I can think of no better way to celebrate Ganesha Chathurthi than to recall and apply this lesson in everyday life.

Optimizing Shubh Laabh: Harmonious Profitability

Sustainable machine-generated, data-driven decisions
The growing popularity of 'Big Data' coupled with 'machine learning' techniques coincides with an increasing use of automated, machine-computed solutions for a a variety of business problems that were once solved and optimized based (predominantly) on human inputs. Machine-generated solutions have been shown to be superior to these previous methods on the measured performance metrics in many instances, and companies all over the globe have deployed advanced analytics and business optimization models (e.g. built using Operations Research) to achieve incremental profitability, cost reductions, or improved system efficiency. However, all is not well. Some solutions are sustainable, and work well over time, while others begin to run into a seemingly endless stream of human or environmental issues, and fall by the wayside. 

What differentiates sustainable machine-generated optimizations from the unsustainable ones? The answer is not straightforward, and this post explores one aspect. For an example of what kinds of issues can crop up, see this BBC news article: "Amazon workers face increased risk of mental illness", as well as this older article on 'unhappy truckers'. A portion of the BBC article highlighted below is color-coded to show where sustainable decision optimization could be potentially applied to improve upon the status-quo):

 "..... Amazon said the safety of its workers was its "number one priority."

Undercover reporter Adam Littler, 23, got an agency job at Amazon's Swansea warehouse. He took a hidden camera inside for BBC Panorama to record what happened on his shifts.

He was employed as a "picker", collecting orders from 800,000 sq ft of storage.

A handset told him what to collect and put on his trolley. It allotted him a set number of seconds to find each product and counted down. If he made a mistake the scanner beeped.

"We are machines, we are robots, we plug our scanner in, we're holding it, but we might as well be plugging it into ourselves", he said.

"We don't think for ourselves, maybe they don't trust us to think for ourselves as human beings, I don't know.

..... Prof Marmot, one of Britain's leading experts on stress at work, said the working conditions at the warehouse are "all the bad stuff at once".

He said: "The characteristics of this type of job, the evidence shows increased risk of mental illness and physical illness."

"There are always going to be menial jobs, but we can make them better or worse. And it seems to me the demands of efficiency at the cost of individual's health and wellbeing - it's got to be balanced." "

I spent the early-mid 2000s redesigning, and improving airline crew scheduling optimization systems. This period also happened to be the industry's most tumultuous: 9-11, out-of-control fuel and labor costs exacerbated by the invasion of Iraq, repeated strikes by various worker unions followed by contentious negotiations that lead to multiple CBAs (collective bargaining agreements) being ripped up and rewritten, and companies lining up to file for Chapter-11 bankruptcy protection, etc. Endless problems. The office atmosphere got quite intense when the R&D team somehow managed to find itself in the middle of these events, and facing heat from all sides (management, unions, soaring passenger complaints) on the kinds of solutions that were generated by our decision support systems (The US airline industry pioneered the use of such techniques). The analytical lessons empirically learnt from such episodes are hard to replicate in classrooms. One such lesson was "pay a lot of attention to the impact of your model on the people and environment". The application of this lesson has been explored in this space in a variety of contexts earlier: here (Gandhi's methods), here (Smart-Grid), here (Airline Crew scheduling), and here (Conflict resolution). The issue is revisited here by borrowing an idea from traditional Indian business philosophy to see if new insight can be generated toward answering our question on sustainable business optimization.

(pic link source: http://www.indiabazaar.co.uk)

(Updated: November 30, 2013 Finally found the link to article that inspired this post)
It is interesting to note that for centuries, traditional business communities in India had adopted the policy of Shubh Laabh (written in Hindi in the picture), which roughly translates into 'auspicious/harmonious profit' (Aravindan Neelakandan, co-author of 'Breaking India' in the linked article notes: "Lakshmi symbolizes the wealth that is holistic: it is wealth that puts welfare (Shub) before profit (Laabh)." The pursuit of wealth and profitability was never frowned upon in Hindu society, while unconstrained profit maximization was recognized as a socially destabilizing and ecologically unsustainable objective.  'Shubh Laabh' recognizes and respects the presence of long-lasting and latent side-effects that arise from business decisions (that can bring you 'bad luck') and attempts to balance them equitably with the more immediate goal of profitability (Laabh). These traditional businesses employed some operational form of Ahimsa (the principle of minimal harm) to optimize Shubh Laabh:

Rule a) limit harm (hard-constraint version)

Rule b) minimize harm (soft-constraint version)

Let us see how this idea can be incorporated within modern decision optimization systems. Amazon appears to have satisfied all legal requirements via (a) by making safety a top priority. It has probably ensured that the statistical rate of accidents is below some stringent threshold. In the airline world, (a) is achieved by ensuring total compliance with respect to all FAA- and CBA-mandated safety rules. However, this represents a necessary condition that tolerates a certain level of error as 'legally acceptable collateral damage'. The resultant formulation is: maximize profitability subject to safety regulations. However, this in itself is an insufficient specification if we want our algorithms to minimize harmful side-effects. An Ahimsa-based model would additionally consider (b) and eschew profit achieved at the cost of a reduced employee quality-of-work-life (QWL) or environmental degradation, as unsustainable and counterproductive in the long run. 

For large-scale systems such as a retail supply-chain or airline crew schedules, a reasonably skilled analytics professional should be able to incorporate requirements (a)-(b) within their decision support algorithm which, among alternative near-optimal solutions (and there are often many of these), selects one that also maximizes worker QWL, and/or minimizes harm (e.g. reduces carbon footprint). This requires the human-and-environment-variables in the system be treated positively as an active and equal partner based on mutual respect, by explicitly including their requirements as part of the primary goal (objective function), going beyond a legalistic/adversarial approach of treating these variables as a 'loss-making noise that has to be managed' by specifying a minimum tolerance constraint.

To summarize
It is possible to achieve sustainable profitable solutions via automated decision support systems that are also harmonious and sustainable, by paying due respect to all the stakeholders (including Ms. Nature), right from the design phase.

An old blog discussed Rajiv Malhotra's use of 'mutual respect' (as opposed to mere tolerance) as a simple but powerful basis for two heterogeneous groups of people, or people subscribing to conflicting thought systems, to achieve a fair and sustainable equilibrium in their interactions. It appears that such a mutual respect:

a) is implicitly present in the idea of Shubh Laabh, which in turn

b) can be employed as a key guiding principle of 'sustainable design' when building decision support algorithms for managing complex business problems, where multiple, and potentially conflicting, goals have to be delicately balanced.

The opinions expressed in this article are personal.

Gandhi and Operations Research

October 2nd is the birthday of Mahatma Gandhi, a major spiritual force behind the Indian freedom movement of the 20th century. Gandhi-ji also was a fundamental and direct inspiration for Martin Luther King Jr.'s civil rights movement of the 1960s, and Nelson Mandela's struggle against apartheid. In this post, we attempt to examine his idea of ahimsa from an optimization perspective.

Update (Oct 5): This Huffington Post article provides amazing insight into Gandhi's ideas.

What is ahimsa?
Indian textbooks mention that Gandhiji brought the colonial empire in India to its knees by using ahimsa and sathyagraha (both were 'spelling bee' words a couple of years ago). These words have no equivalent in English, and are often used to imply "passive resistance", "pacifism", or "non-violence". A mathematical optimization model provides a more useful translation.

The popular Sanskrit verses on himsa and ahimsa given below was popularized by Gandhiji:
ahimsa paramo dharmaha,
dharma himsa tathaiva cha
[Oct 2018 update: the second verse has been attributed to Swami Chinmayananda)

My translation:
Non-harming is the greatest virtue;
So too is righteous harm.

The second line suggests that allowing cruelty to go unchallenged is equivalent to willingly permitting harm, and therefore, must be resisted. The verses are a combination of the ideal (global optimality = zero harm), and a context-dependent violation of that ideal (soft rule = minimize harm). 'Local harming' is permissible in rare circumstances when it results in an overall reduction in global harm. The gangrenous foot has to be amputated to save the body, or a terrorist who attacks innocents in a mall or a school has to be taken down by security forces. In a recent talk, Narendra Modi tells us a story of how Gandhi would request his assistant at Sabarmati Ashram to pour back half a cup of water back into the river, because all he wanted was half a cup. Minimal harm!

(updated October 2)
Optimization Model of Ahimsa
From an optimization modeling perspective, these ahimsa verses represents an objective function of minimizing harm. In normal circumstances, the optimal value should be zero, but in all circumstances, it should be minimal. When some non-zero harm is inevitable, the goal is to limit the total harm to a minimum, i.e., the employed level of harm is optimal if and only if it is necessary and sufficient to restore dharma. The 'necessary' condition implies minimalism of the counteracting harm, while the 'sufficient' condition implies the safe neutralization of the source of the harm. It's a tough balancing act for humans even though nature itself effortlessly adheres to Newton's third law. A pure hard-constraint version of ahimsa would discourage self-defense and even celebrate cowardice, while a pure soft constraint version could open the doors to unnecessary use of force, and justifying cost versus benefit approaches. (The legal system dictum of "let a hundred guilty go unpunished, but a single innocent should not be wrongly convicted" is an interesting case study in this regard.) Hence, applying any one of these two lines is an incomplete specification and can lead to unpredictable results.

We argued a while ago that these verses are an improved 'fail-safe' choice for the 'zeroth law' of robotics. In the real world, when we build decision models to aid decision making, we can optimize decision variables to pick a pareto-optimal solution that also results in the least disruption to the system ("don't fix what isn't broken"). For example, if we are scheduling workers to maximize efficiency or minimizing cost, then an optimal solution that also minimally disrupts (and preferably, enhances) their quality-of-work-life is more likely to be sustainable over the long run.

Gandhiji's Swaraj
Many feel that Gandhiji was partial to the first line, and quotes attributed to him support this claim. On the other hand, Gandhi's 'Hind Swaraj' and his lesser known quotes on preferring violent self-defense to cowardly capitulation suggests that he was aware of both verses. His book 'Hind Swaraj' (Indian self-rule) implies that his primary objective was not merely an overthrow of colonizers, but to achieve the strategic and deeper goal of ending the cultural genocide of India (restoring its Sanskriti and dharma). Applying the ahimsa verses would yield a path to Swaraj that results in minimal incremental harm to India's Sanskriti and dharma. Such a path may not necessarily also be optimal in terms of being the shortest-time path, or the least painful, or one that maximizes regained territory.

Optimizing Schedules: QWL Considerations

A first blog post from India.
Impact of Decision Variables on Humans
This practice related comment was triggered by yet another useful 'Punk Rock OR' post  - on 'optimization and unhappy truckers', which briefly reviews a Tom Vanderbilt article that noted that mathematical optimization may having contributed to the incremental unhappiness of employees who were affected by the decisions prescribed by the model. TV's article also talks about the optimization of airline crew schedules, which is a useful example to analyze some of the side-effects of optimization.

Scheduling Objectives

The rules that govern the safe scheduling of airline crews are incredibly complex, and used to appear in a bound book form (every single one must be programmed into the optimization system, scarring an Operations Researcher for life). Additionally, there are hundreds of different 'cost' components that typically go into an airline crew scheduling system that is so neatly abstracted to "Min cx. Ax=1, x binary" in OR textbooks. Some of the objectives are listed below:

1. cost, utilization, efficiency

2. quality of work life (QWL)

3. schedule regularity

4. operational resilience

5. Downstream system compatibility

.. and many more..

Among these, a component that is most relevant to the discussion here is QWL, a non-negotiable component of "soft rules" that go beyond what the FAA prescribes and diligently adheres to the letter and spirit of a collective bargaining agreement (CBA) between the management and representatives of the crews. QWL metrics are audited and checked before schedules are published, and tracked over time. A drop in QWL metrics can result in followup phone calls from crew representatives, and keeping the call volume (and decibel) to a minimum is a clear and track-able goal.

Anonymous Schedules, Personal Impacts

While traveling on company flights, I initially used to strike up a conversation with flight-attendants (FAs) to get their opinions on their schedules, and any particular issues they had. There were some harsh complaints, but also the occasional compliment based on their feedback that compared their QWL to FAs in other carriers.  Nevertheless, schedules are initially anonymous, and thus indifferent to personal needs, while also being free of privacy concerns. It is safe to say that unless schedules are personalized, there's bound to be unhappy crews. Personalization is at odds with automation, and the task of optimally synchronizing and scheduling 30, 000 FAs and pilots, and hundreds of expensive aircraft that operate thousands of flights per day, while trying to keep costs down, reliability high, and crews happy is non-trivial. Luckily, the space of feasible schedules contains many trillions of possibilities, and is diverse enough to accommodate many, many management and crew objectives to produce tons of alternative near-optimal solutions. In fact, this feature plays a vital part in designing new and improved crew safety rules during CBA negotiations. To summarize, modern, large-scale industrial optimization systems are sophisticated, robust, and flexible enough to accommodate a myriad of human-impact objectives without breaking a sweat. Who knows, truly personalized schedules that sync with personal calendars, while also keeping utilization high, may well be technically feasible now. Preferential bidding systems (PBS) have already been in place for more than a decade now.

Actions Reflect Priorities

Some of my purely personal observations based on the data I have seen: the QWL metrics for a schedule is correlated to the negotiating clout of the organization for whom the scheduling is done, and the importance given by management to maintaining harmonious relations with them. Higher up the food chain, the better the QWL. Not surprisingly, some employee organizations may have their own optimization systems that enable them to evaluate their schedules (and also 'game' the system). 

In between the scheduler and the 'schedulee' is the OR layer, the secret sauce. I'd like to believe that OR'ers can make and have made a positive difference by paying attention to the net human impact of a binary variable changing from a 0 to 1 to find win-win stakeholder-friendly alternative optima. I've seen analysts devote many days trying to figure out how to make excruciatingly complex experimental QWL constraints work cost-effectively in the optimization system to break an ongoing CBA negotiation deadlock: for example, how to limit the flying done by west-coast based pilots when it is dawn, Eastern Standard Time (EST). Putting the plane on autopilot and going to sleep is not an option. I have even seen prototypes that used "crew happiness variables" :)

It is interesting to look at the optimized crew-aircraft schedules for fractional jets that ferry well-heeled folks and time-starved execs on Gulfstream-Vs to various parts of the world between tiny airports. Needless to say, non-bottomline 'costs' and degree of personalization play a prominent role in the objective function. The customer is both king and queen. In the end, a well-designed optimization system's objectives can accommodate the considerations of all the stakeholders to consistently (and merely) reflect their relative importance from a human decision-maker's perspective. Nothing more, nothing less. As Gandhi ji said, actions reflect priorities.

Analytics and Cricket - XI: Using DRS Optimally

The Ashes
The first Ashes cricket test that concluded earlier today ago in England triggered this post.

(pic source link: guardian.co.uk)

This post is again related to the Decision Review System (DRS) that combines machine and human intelligence to support evidence-driven out/not-out decisions in cricket. The previous cricket-related post can be found here where a reader critiqued an earlier post on the false positives issue of DRS (that saw a lot of visits last week during this cricket match).  It's now apparent that England won the match thanks in part to their superior use of DRS compared to Australia.

DRS
The DRS consists of 3 components:
1. The human (a team of umpires, camera manners, and hardware operators)

2. The hardware (hot-spot, slow-mo cameras, snick-o-meters, etc). These are the data gathering devices.

3. The Analytics and Software (ball-tracking and optimal trajectory prediction, aka 'Hawkeye' based system)

This gives us three separate (but possibly correlated) sources of error:
a. Operator error

b. Hardware error: Technology limitations - resolution, video frame rates, hardware sensitivity, etc. may be inadequate at times for sporting action that occurs as fast as 100 mph or spin at 2000 rpm ...

c. Prediction algorithm error: Given such variations in sporting action, a forecast of the future trajectory of the ball is also subject to uncertainty.

(pic source link: dailymail.co.uk)

A smart user, after sufficient experience with the system, will be able to grasp the strengths and limitations of the system. In test cricket, a team is allowed no more than two unsuccessful DRS reviews per inning. Thus it is a scarce resource that must be cleverly used to maximize benefit.  In fact, the DRS is an example of a situation where the use of a decision support system (DSS) itself involves decision-making under uncertainty using a meta optimization model.

Optimal Usage of DRS
There are several factors that dictate when the trigger must be pulled by a cricket captain to invoke DRS to try and overturn an on-field umpiring decision.
i. The probability of success (p)
ii. The incremental reward, given a successful review (R)
iii. The cost of an unsuccessful review = cost of status quo (set to 0, normalized)
iv. The expected future value of having no more than k reviews still available in the inventory (concave, decreasing f(k), with f(0) = 0).

Reviewing only based on (ii) is a like a "Hail Mary" and banks on hope. On the other hand, paying exclusive attention to (i) may not be the best approach either, since it can result in a captain using up the reviews quickly, reducing the chances of taking advantage of the DRS later when "you need it the most". A person who doesn't use DRS at all (or too late to have an impact) leaves unclaimed reward on the table.

Probability Model
We'll start with a simple model. It's not perfect, or the best, but merely a good starting point for further negotiations.

The value of do-nothing  = f(k).

The value of a DRS review = p[R + f(k)] + (1-p)[f(k-1)] = pR + pf(k) + (1-p)f(k-1).

It is beneficial to go for a review when:
pR > (1-p)[f(k) - f(k-1)] or

p/(1-p) > [f(k) - f(k-1)]/R

i.e., odds must be greater than marginal value of a review / marginal reward

In other words:
it is good to review when the odds of overturning the on-field decision exceeds the ratio of the expected cost of losing a DRS review, to the expected incremental reward.

Use Case: for fifty-fifty calls (p = 0.5) with a single DRS review in the inventory, you would want to review only if you are convinced that the present reward is likely to exceed the value of not having DRS for the remainder of the innings. For a fixed reward, the RHS increases steeply after the first unsuccessful review due to the concave f. To be really safe, you want to risk a second and final unsuccessful review only when you can trigger a truly game-changing decision that greatly increases the chances of winning the match.  In general, R may neither be a strictly increasing nor a decreasing function of time. This is especially true in limited-overs cricket where a game-changing event can occur very early in the game. However in soccer, baseball, or basketball, R can be reasonably approximated as an increasing function over time. In general, it makes sense to save the review for the end-game. In any sporting event, including cricket, which is heading for a close finish, it may be beneficial to delay the use of a review.

In the Ashes test, Michael Clarke, the Australian captain appeared to pay more attention to 'R' and less attention to 'f', and was left without recourse at a crucial stage, and this hurt his team. On the other hand, the England skipper Alastair Cook delayed the use of DRS: The last wicket of a closely contested match fell when the game had reached a climax (R = R_max), and was DRS-induced. Thus, optimally delaying DRS involves the constant assessing and updating of risk versus reward and pulling the trigger when the odds are in your favor.

Analytical Decision Support Systems
A smart organization will aware of the strengths, weakness, and value of DSS based decisions. In some industries that are characterized by shrinking margins, even small incremental gains in market-share or profitability using DSS can alter the competitive landscape of the market. This motivates an interesting question: If two firms employ the same decision analytics suite provided by the same vendor, does it necessarily cancel out? or like in cricket, can one firm do a better job of maximizing value from the DSS to gain a competitive advantage?

Updated July 17, 2013:
It turns out, that wicketkeeper Matt Prior was instrumental in ensuring England's good DRS strategy. As we all know, a good prior saves your posterior during crunch time!

Are Operations Researchers Eulerian Decision Makers?

It is interesting to read Venkatesh Rao's book 'Tempo' (and the blog) from an Operations Research (OR) perspective and explore its ideas on human decision making. To better comprehend the themes in the book, I've been trying to map the examples in the book to familiar OR-type problems. For example, timing the purchase of consumer products at Amazon.com over a period of time while also managing to satisfy the free-shipping threshold to the extent possible feels like playing a game of tetris where we try to optimally delay the inevitable ("entropic optimization").

A recent tempo blog post compares Lagrangian versus Eulerian decision makers (let's call them LDMs and EDMs). These two approaches appear to have their origin as choices of frames of reference for studying fluid flow. An LDM tracks a single 'parcel' of flow though its entire journey while the EDM examines many parcels that flow through a particular location. Let's apply this to an airline OR (crew scheduling) scenario:

an LDM follows a particular crew operating a sequence of flights through a network before returning to her/his base, while the EDM looks at many crews arriving and departing through a particular airport. LDMs track flow-paths, while EDMS zoom in on activities within a specific node in the network. Another example from airline network revenue optimization: an LDM may focus on a passenger flowing and the accumulated revenue through an itinerary consisting of multiple flights, whereas the EDM zooms in on the variable price and fixed cost associated with a seat on a specific flight that is a part of many passenger itineraries.

Intuitively, it feels as if the LDMs focus on the clearly visible primal decision variables, whereas the EDM focuses on the latent, underlying dual problem. Column generation as an iterative optimization approach works well when the EDMs compute and transmit useful 'dual' signals from specific locations to LDMs (e.g., the rate of pilots arrive and depart at their node) who in turn, collect and apply this updated information to stitch together new and improved flow-paths across the network (i.e., columns), which in turn ensures more balanced and smoother flows across the nodes (i.e., rows). For some problem classes, if we solve the dual problem, our primal decisions are also at hand, and if the dual is unsolvable, the primal is likely to be problematic as well. Conjecture: Lagrangian and Eulerian decision makers build mental models that represent (informal) primal and dual formulations of the same problem. Depending on the context, one or the other approach may be more convenient or effective. 

The ability to extract information from noisy dual signals is useful in solving complex or large-scale industrial optimization problems. For example, in the airline revenue optimization example, it is known to be more convenient and reliable for airline revenue/scheduling planners to take decisions based on the shadow prices of flights, rather than rely on the optimized prices of the millions of possible itineraries. Dual methods are insightful and can help solve a decision problem where a frontal assault fails. To the best of my knowledge, the default out-of-box method invoked by the leading mathematical optimization software packages today for solving linear programs is their dual algorithm.

In general, Eulerian decision making or as this post sees it, 'dual-driven decision making' may be the more attractive first-choice approach for human decision making. The tempo blog notes:
"... When flow gets turbulent, the fluid mixes a lot. To properly follow a “parcel”, you have to let it expand as flow lines diverge and churn. This means there is more fluid in your parcel than you started with, more “noise.” Eventually you are trying to analyze world hunger — the entire body of fluid.
But the Eulerian static parcel stays the same size. It just bleeds causal structure and gets more entropic. The action gets a lot more random and choppy, but still tractable in size. It is also easier to shrink what you’re paying attention to when things get complex — it’s called focusing – than it is to reduce the ambition of a model you’re tracking (generally called pruning)..."

Are Operations Researchers Eulerian Decision Makers?

Jugaad Innovation: Stuck in a Local Optimum

Updated July 5, 2013:
CNN link on Jugaad Innovation.

The book cover blurb sounded exciting: Do more with less. An alternative to risky expenditure-driven, resource-hungry growth using a "bottom up approach to frugal and flexible innovation". Endorsement from a salesforce.com CEO. No doubt, Jugaad is a useful concept - it's a Hindi word that implies a improvised and clever work-around, but there is such a thing as stretching an idea too far. Jugaad arises from the Indian way of doing "more with less" - possibly a public response to artificial scarcity induced by genocidal British colonization since the late 18th century when a resource-rich India's share of world GDP plummeted from 25% to a negligible quantity in a short time to turn it into from a knowledge and manufacturing economy into an agrarian, impoverished nation. A scarcity-driven economy that has been nurtured by Nehruvian socialist politics over the last 60+ years. However, the word itself ties in nicely with ideas in Operations Research that deal with the optimal allocation and utilization of scarce resources, and hence this post.

The book has some heart-warming and splendid examples of Indian innovation, which are nice to read. The remarkably decentralized and entrepreneurial nature of the workforce, and their seeming comfort in operating effectively within what appears as "chaos" to the external observer is one of the salient features of the native Indian economy. A discussion of the Indian practice of the "missed call" (the closest that mankind has come to 'half a bit' of information) adds humor. The authors could have lent depth to the book by exploring the deep-rooted cultural origins of the decentralization, and entrepreneurial spirit that drives the Indian way of doing things. They do make some useful comparisons between the profligacy and rigidity of some Western CEOs with the adaptive nature of the "jugaad entrepreneur". However, celebrating every success from Bharti airtel to PepsiCo's direct seeding to Mitticool (now that is really cool) as a win for "Jugaad" dilutes the message. When the book hyped the crappy 'Aakash' tablet, the populist, Indian government's tax-payer subsidized tablet for the unwashed masses (a cheaper and cynically sub-optimal alternative to constructing and maintaining primary schools), it was time to cry halt. Celebrating artificial scarce resource allocation, however efficiently done, feels like a cop-out.

Operations Research practice is not just about optimizing within constraints and declaring victory. Value can be unlocked by using OR to expose the more expensive bottlenecks in the system. Constraints that turn out to be artifacts must be eliminated whenever possible. I'm all for (also) 'thinking small', but not at the expense of losing the context.

Synchronized Scheduling Case Study - Part 2

The case study was introduced in the previous post. Read the introduction here.

.... Despite their suboptimal approach, the schedulers of the LEG carrier did some nice work with their synchronization approach.

2SAT versus 3SAT
There is a jump in decision complexity when moving from a two-body conflict resolution problem to three-body conflict-resolution. Intuitively, we can observe this in linear programming (LP) problems. LPs limited to two variables per constraint are easier to solve. The dual (two constraints per variable) is a generalized network flow problem that can be solved using purely combinatorial methods. Intuitively, one can adopt an alternating heuristic that iteratively fixes one variable and optimizes the resulting univariate problem. In computational complexity theory, SAT-2 is easier to manage compared to SAT-3.

Rather than trying to simultaneously juggle flight attendant (FA), pilot, and aircraft maintenance schedules, the LEG schedulers adopted an simplification. They combined the two human elements into a single 'crew' scheduling entity. This can be achieved via "co-pairing", i.e., by building common pairings (partial schedules) for FAs and pilots employing X = {X_PILOT ∪ X_FA}, i.e., every pairing would be feasible to both pilot and FA contracts and FAA regulations. Column generation allows you incredible flexibility in accommodating such a change. This flexibility is invaluable when hit by a perfect storm. Now, they could work with a two-variable per constraint structure so that their telephonic "alternating heuristic" negotiations could converge in a timely manner. Unfortunately, such a beneficial reduction in complexity was accompanied by inefficiency. The approach erred inordinately on the conservative side. The resources were now too tightly integrated, resulting in a cost spike due to under-utilization of expensive resources. LEG's R&D team was invited to resolve this problem.

Integrated Crew and Aircraft Routing (ICAR)
The OR team quite naturally proposed an integrated scheduling approach that would reduce costs, improve reliability of operations, and also enable tactical synchronization of flight connections. However, the number of possible connections to be evaluated is O(n²) for a n-flight schedule. Methods in the literature are typically data-agnostic; sophisticated alternatives to the straightforward approach designed to perform an acceptable job even in the worst case. Luckily, LEG's OR team had to solve only LEG's problem and not that of every future airline and space shuttle.

The Beauty and the Brute Force
Given the problem size spread over a five-day period, there are a few million candidates to synchronize in the theoretical worst case. However, our focus is only on same and overnight connections, which reduces the candidate list to a few hundred thousand. When we recognize that we only have to focus on the set T of connections of the same sub-fleet type at the same airport (duh), and exploiting a few other contextual restrictions, this final number is data-driven. It turned out that there were no more than tens of thousands of synchronization possibilities to deal with, 99% less than the theoretical bound. This finding allowed the possibility of cheerfully discarding (as we found out a couple of years later when some of the first journal papers on ICAR came out) all the neat out-flanking theorems and approximations presented in the literature.  The method chosen may score low on aesthetics but was effective and flexible in managing the full blown ICAR instances in context. There is a childlike joy in discovering that enumeration solves your practical business problem - akin to finding out that this emperor has very little clothes. This basic counting step gave the team a clear goal.

Problem Statement (at last)
Given a flight schedule, jointly construct feasible crew pairings and aircraft strings such that their scheduling intersection (elements within the connection set T) are optimally synchronized.

And, yes, there were a few TSP (Traveling Salesman Problem) instances to be managed within ICAR. To be continued....

Conflict Resolution - 3: Contextual Optimization

Asimov's Zeroth Law
We continue this discussion from where we left off a few weeks ago: robot-ethics and how Asimov's robots resolved conflicts. A key update to Asimov's original three laws is the inclusion of the zeroth law (from Wikipedia):

"A robot may not harm humanity, or, by inaction, allow humanity to come to harm."

A fundamental difference between this law and the others is its abstract specification, with little clue on how it will be implemented. Furthermore, by giving this law the highest priority, Asenion robots are designed to first and foremost safeguard 'humanity', while also minimizing injury to individual humans and themselves as secondary and tertiary objectives. A robot is given the difficult computational task of proving that the cost of hurting a human is less than humanitarian benefit derived from an alternative action, and must do so within a finite amount of time.

The Universal Conflict-Resolution Model
Immanuel Kant's 'categorical imperative' is an example of an universal conflict resolution model that furiously strives to be context-free, and one that has greatly influenced western thought. Similarly, the Ten-Commandments' "Thou shall not kill" is absolute. Any machine that includes this hard constraint would be unable to kill, even defensively in order to protect a large number of humans under threat. Kantian rules are easy to 'encode-and-forget' within machines and systems since one does not have to ever worry about the context of its application. As we saw in the previous post, the robotic laws are context-free and Kantian in design. The original three laws operated as hard, must-satisfy constraints and necessary conditions. Per Wikipedia, Kant's

"perfect duties are those that are blameworthy if not met, as they are a basic required duty for a human being."

The problem of course is that (see prior post), the rigidity of all-hard rules is not practical and later versions of Asenion robots appear to additionally operate based on the concept of Kant's imperfect duty (again from Wikipedia):

"unlike perfect duties, you do not attract blame should you not complete an imperfect duty but you shall receive praise for it should you complete it, as you have gone beyond the basic duties and taken duty upon yourself"

Thus, imperfect duties are soft, rather than hard constraints, and the aim is to maximally satisfy (minimally violate) requirements. However, the optimization weights that trade off the degree of importance assigned to each of the Asenion robot's 'imperfect duties' are hard-coded, and additional context-specific inputs are required from humans to achieve a satisfactory result in resolving a dilemma. The robots have no authority to perform context-specific conflict resolutions on their own.

How then does this zeroth law practically work in Asimov stories? It's Kantian abstraction is incomprehensible to all except a couple of enlightened robots (with telepathic ability, no less). In general, the zeroth law remains useful only on paper for most of the robots.

Contextual Ethics: The Indian Way
Rajiv Malhotra's path-breaking book "Being Different: Indian Challenge to Western Universalism" provides a fascinating contrast between the traditional Indian approach of 'contextual ethics' (CE) that arose from its Dharmic thought system, and the 'context free ethics' that largely guides the western approach (for an earlier post based on his work, see here. Some of the methods in this post are applications of ideas in this book). From an optimization perspective, we can think of a CE-embedded robot as one that maximally satisfies a combination of hard, soft, and firm constraints, where a 'firm' constraint refer to a hard constraint that is minimally and temporarily relaxed, depending on the specific context of the dilemma, and for the benefit of the 'greatest good', at the expense of incurring a context-specific penalty. This flexibility must not be confused with moral relativism - where a set of soft constraints are tactically and optimally manipulated according to context to maximize some convenient self-serving objective. 'Optimally timing an apology' can be thought of as an example of a non self-serving, contextual optimization model.  It is worth doing a deep dive into this concept, by reviewing some passages in the aforementioned book:

"The frequently leveled charge of moral relativism against this contextual morality is inaccurate, because the conduct and motive are considered consequential in judging the ultimate value of statements.

.... Dharmic ethics are formulated in response to the situation and context of the problem in a way that makes Western ethics seem unduly codified, rigid, monolithic and even simplistic. A.K. Ramanujan, in his influential essay 'Is There an Indian Way of Thinking?', uses the terms 'context-free' and 'context-sensitive' to contrast the West and India in their respective approaches to ethics: "Cultures may be said to have overall tendencies to idealize, and think in terms of, either the context-free or the context-sensitive kind of rules. Actual behavior may be more complex, though the rules they think with are a crucial factor in guiding the behavior. In cultures like India's, the context-sensitive kind of rule is the preferred formulation" ....

.... Dharmic traditions, on the other hand, have long sought to arrive at truth by balancing universal truths and acts with those that can be determined only in the context in which they occur. Dharmic cultures have thus evolved to become comfortable with complexity and nuance, rejecting notions of the absolute and rigid ideals of morality and conduct....

....dharmic thought offers both universal and contextual poles – not just the latter, as that would be tantamount to moral relativism."


The dharmic approach lies in between an "all-soft constraint" and the Kantian "hard and soft constraint" approach to decision optimization.


Applying contextual optimization
Asimov's telepathic robot Giskard formulates and solves a probabilistic optimization problem where it trades off the opportunity cost (in terms of human lives) against the expected benefit to humanity. However the degree of uncertainty in this conflict-resolution model is too high and the robot eventually crashes. This episode comes across as an example of applying the CE approach to resolve a dilemma. The great Indian epics - the Ramayana and the Mahabharata, contain several brilliantly narrated instances of contextual conflict-resolution. Indian sci-fi movie buffs would not be surprised to know that George Lucas' Star Wars was inspired by the Ramayana.

Contextual optimization in the specific area of 'mathematical decision support software' would mean: allowing the rules of engagement to be configurable depending on the context. For regular users, advanced settings are greyed out, with only universal (default) rules enabled. Only super users, who are well-trained and comprehend the nature and consequences of the beast, get to work with 'firm' constraints, and on rare occasions. For example, an airline crew schedule optimization system should be configured to satisfy contractual and FAA rules, except during emergencies (e.g. post 9/11 recovery) where 'crew welfare' is only achievable by overriding one or more of these rules. Practical decision support systems should be carefully designed to allow such controlled contextual optimization.

Amending the Zeroth Law: The Dharmic Robot
The four laws do not quite protect the rest of the cosmos (e.g., from humanity) given their anthropocentric nature. From an Indian point of view, this gap can be closed by modifying the zeroth law based on the contextual ethics of dharma. Rajiv Malhotra, in his book, provides the etymology and a working definition of dharma:

"Dharma has the Sanskrit root dhri, which means 'that which upholds' or 'that without which nothing can stand' or 'that which maintains the stability and harmony of the universe'. Dharma encompasses the natural, innate behaviour of things, duty, law, ethics, virtue, etc. For example, the laws of physics describe current human understanding of the dharma of physical systems. Every entity in the cosmos has its particular dharma – from the electron, which has the dharma to move in a certain manner, to the clouds, galaxies, plants, insects, and of course, man. Dharma has no equivalent in the Western lexicon."

In such a framework, Asimov's laws would delineate a robot's various dharmas. At the highest level, we can require that a robot abide by the following fundamental law, that is based on an ancient Indian text:

"Non-harming is a robot's highest priority, except in the defense of dharma"

Conflict-resolution is always performed by first applying this highest dharmic principle and customizing it to the specific context. Note that by operating on the fundamental dharmic principle of least harm, a robot would usually satisfy Asimov's zeroth law, albeit in a context-specific manner, while also being in harmony with the original laws, as well as any new laws that get written in the future. Interestingly, the Hippocratic oath of medical doctors is based on a similar idea that represents a non-negative bound: "do good, or at least no harm".  If complex systems, new drugs, etc., are designed by always keeping this fundamental principle in context, it may well minimize the risk of catastrophic failure.