“Scientists can predict the path of two billiard balls with precision, and the average behavior of two million gas particles. But what about the messy middle ground, where twenty or thirty components interact with one another in unexpected ways? … Complexity covers the untidy yet vibrant realm where so much of life unfolds.” – Simple Rules: How to Thrive in a Complex World [11].
Decision Policy Search: The Curse of Dimensionality
Our previous posts focused mostly on single decisions under uncertainty. We now examine decision cycles, i.e., finding a good decision policy to make a series of interconnected decisions over time in a complex world.
An intuitive observation from decision science and engineering practice is that the degree of complexity of a system is less dependent on the sheer number of components and more on the inseparability of their effects and non-ignorable level of interactions between them. This simple idea of complexity is hidden in everyday idioms like ‘two is company, three’s a crowd’ – move from 2 to 3 interactions and order turns to chaos. In classical mechanics, a 2-body problem is far easier to manage than a 3-body problem. Similarly, a fundamental mathematical result that governs optimal decision-search comes from the field of computational complexity: the2SAT (satisfiability) problem is easily solved, whereas 3SAT can be really hard.
‘Dynamic programming’(DP) is a general and widely studied decision-making approach that is often used to simplify problems involving several simultaneous interacting variables, into a sequence of simple univariate decision problems and without losing the original generality. This approach can identify a best decision policy if the decision-maker is able to extract such DP-friendly structure in a real-world problem. DP has been tested in diverse domains, from economics to identifying the best way to load a delivery truck, inventory planning for supply chains, and calculating the best spacecraft trajectory for space exploration missions. Starting from the 1950s work of Richard Bellman, DP has today evolved into methods like Reinforcement learning(RL) that propel recent AI successes in board games [6].
DP models require certain separability and monotonicity assumptions to be satisfied for them to work well in practice. Herbert Simon, who won the Nobel Memorial Prize and the Turing Award for his research in decision making, was also a DP expert. He concluded that “decision makers can satisfice either by finding optimum solutions for a simplifiedworld, or by finding satisfactory solutions for a more realistic world. [3].
Other researchers have also noted the practical limitations of DP in handling real-world complexity. John Rust remarks in his study of DP’s impact on decision making [4] that “ it is far more challenging to take things from pencil and paper and actually formulate and solve a host of real-world decision problems as mathematical DP problems”. He also cites other works that extend Herbert Simon’s point by comparing the performance of mechanistic ‘optimal’ models versus human ingenuity-based decision making: “.. humans evolved to survive in a world filled with a large and diverse set of ill-specified problems. Our ‘suboptimality’ may be a small price to pay for the flexibility and adaptiveness of our intuitive decision processes.”
Rust reminds readers that “the complexity bounds from computer science show that the curse of dimensionality cannot be broken by any algorithm, including the new stochastic algorithms from the reinforcement learning literature.” In light of this fact, it should not be surprising that media headlines play up the genuinely impressive AI results in the simplified lab/game world, only to follow it up with ‘delayed commercial implementation’ when AI has to face the full curse of dimensionality. AI users are left to face the consequences of the breakdown of modeling assumptions in the real world. The complex interaction between the various components of a real-world system is a key reason why ISRO legend Nambi Narayananrepeatedly stresses the practically irreplaceable value of human ‘know why’ over the more automatable ‘know how’ in making critical design decisions, as noted in an article in our daughter portal [20].
Our Vedic seers have studied the general idea of complexity at arguably the deepest level and use the powerful metaphor of ‘Indra’s Net’ in the Atharva Veda [14] to describe our world. “Indra’s Net symbolizes the universe as a web of connections and interdependencies among all its members, wherein every member is both a manifestation of the whole and inseparable from the whole”.
Assumptions to abstract out selected portions of Indra’s Net and build decision models are a distortion of reality. Western epistemology is comfortable with this loss in fidelity as it enables mechanization and theorem-proving within the idealized model. On the other hand, the Bharatiya approach directly interacts with reality based on the Vedic Ganita principle of the one manifesting as many.
From Dharma to Rta
The previous post studied some momentous decisions of Arjuna in the Mahabharata. There is a tendency to isolate and dissect such choices in the Ramayana and Mahabharata using a western lens and make sweeping claims. Context-driven decision making without consistent ethical grounding is basically an antifragile policy that optimizes a “pay off” based objective [21]. It leads to adharma. On the other hand, it has always been known to ordinary Hindu practitioners that there is a deeper unifying theme underlying the decision-making in our Itihasa-Mahakavyas.
One of the purposes of Itihasa is teach us how to ethically navigate through entiredecision cycles. Karma accrues with every decision and persists even across births and cannot be outsourced to some savior. Perhaps the exemplar of integrity in this context is Raja Satya Harishchandra, whose life story, at least until the last few decades, was known to almost all Indian families.
This broader unifying principle in Arjuna’s decision making under the guidance of Sri Krishna is the intent to uphold Rta: “Arjuna is not simply fighting for his side but for justice and the broader establishment of dharma, one that is in line with the ritam (inherent nature) of the cosmos.” – Rajiv Malhotra [13].
The goal of Dharmic Decision-Making is to uphold Rta, which leads to Ritamic (or Rtamic) Decision Policy. Rta represents the cosmic order and self-organized harmony of the world. Rta can be observed and experienced by us. For brevity, we refer the reader to this important essay [2] that summarizes among other things, the connection between Dharma and Rta, which is our focus here, as well as between Rta and Satya. The essay emphasizes that Rta does not stand by itself as Satyam is the truth of being/existing while Ritam is that truth in action, and Dharma is the upholding of Rta. Dharma is nested within Rta, which is in turn contained within Satya.
In the Kurukshetra war, even the most predictable of natural phenomena, the sunset, appears to be delayed by Sri Krishna’s intervention. This is not anritam as ritam is not an independent reality but serves the deeper transcendental truth embodied by the divine Sri Krishna. The Kaurava plans that implicitly assume a ‘separate and predictable Ritam’ fail and Arjuna becomes Bhagavan’s instrument to end Jayadratha’s lifetime of adharma that was leading to Anritam.
By Ramanarayanadatta Shastri (Public Domain, Link)
Another important discussion of Vedic Ritam is part of this WAVES talk by Sri Rajiv Malhotra.
This talk mentions the Vedic understanding of the nested nature of reality and how the “same reality of rtam is encoded/involuted at multiple levels“. Note the structural resemblance in Bellman’s ‘Principle of Optimality’ of DP. This principle encapsulates DP’s main idea of breaking down a complex decision problem into smaller sub-problems that are nested recursively within larger decision problems. This principle will become apparent to anyone solving the ‘Tower of Brahma’ puzzle.
Ancient India knew about individual gain from robust or antifragile decision making (see part 1) but accorded a higher priority to upholding Rta whose essence is harmony [2]. If we look around and pick any domain, we find that some decision policies work well for long whereas others fizzle out.
Vidura Neeti Guide to Good Decision Policy
Just as zero and the place value system arise from Rta, this cosmic order also enables us to differentiate between sustainable and unsustainable decision policies. Ancient Vedic texts [13, 17] mention how Prajapathi’s first version of the cosmos is a state that is too Jami and undifferentiated. This is too predictable and ordered; an “all mean and no variance” system. Decision making is unnecessary in a world with little uncertainty, and there is no role for Karma either. The second attempt is a cosmos that is Prthak, excessively differentiated and chaotic. Nothing is predictable, decision-making is useless, and Karma has no role here either. The third and final attempt uses the Vedic resemblance principle of Bandhuta to produce a cosmos that is self-organized by Rta and thereby avoids both extremes. Rta naturally reveals the role of Karma since the rewards of our actions are as per Rta [2] so this is a Karmic world where decisions come with consequences. Ritamic decision making incorporates karma and does not require or rely on the western modeling assumption of ‘randomness’.
Rta is directed from above and within by Satya that is supreme and invariant(Shruti) and keeping this ultimate truth as paramount, Rta permits contextual dharmic decision making (Smriti). This Vedic conception is echoed in the 5000+ year old Vidura Neeti to state a universal requirement for good decision policy:
“That is not strength which is opposed to softness. On the other hand, strength mixed with softness constitutes true policy which should ever be pursued. That prosperity which is dependent on crookedness alone is destined to be destroyed. That prosperity, however, which depends on both strength and softness, descends to [children and grandchildren] intact”. –[5].
Vidura and Dhritarashtra. Unknown artist. Public Domain, Link
To consistently translate this ‘bifocal’ vision into actual policy requires strength of character.For example, Chhatrapati Shivaji Maharajwas clinical in dispatching the most violent colonizers of Bhaarata and also treated their female family members with utmost respect and humility always, and without exception. Vidura Neeti also resolves the modern Indian non-debate on whether India should be a soft power or a hard power. This fundamental characterization of sustainable policy given in the Mahabharata manifests and hides in plain sight all around us.
Examples of Ritamic Policy
We now review some Ritamic decision-making scenarios to confirm that this is not some abstract concept but real, useful, and non-ignorable in this era of powerful AI (sic) decision making systems.
Example 1: Engineering Design
A concrete column stands strong when subject to heavy compressive loads but crumbles when subject to bending forces. It needs to be suitably reinforced with flexible steel bars (rebars) to increase its tensile strength and become RCCthat is used in everything from buildings to bridges today. Similarly, a very rigid vehicle (e.g., Nehruvian-era Sarkari Ambassador car) performs poorly in accidents as it transfers all the shocking force to its occupants. A car with the right balancecombines rigidity and flexibility that allows it to crumple appropriately and absorb the collision impact and save the passengers.
Example 2: Flight Routing in sync with Ritam
In October 2016, Air India (AI) changed the route of its nonstop flights between New Delhi (DEL) and San Francisco (SFO). Previously, AI followed the conventional to-and-fro polar route policy which minimizes travel distance. However, it was calculated that by switching to a non-polar route to SFO in tandem with the original polar route on the return trip to DEL, AI would be a couple of hours faster despite traveling more than a thousand kilometers extra. This routing decision rides the tailwinds due to the Jetstream across the Pacific Ocean and results in a roundtrip around the earth. This route also conserves fuel, a win-win for passengers and the airline and reduced environmental himsa. The Vimana circumambulates Bhumi Devi in harmony with Rta.
Example 3: Ritamic Forecasting for Decision Support
Prediction of natural phenomena and weather has influenced the outcomes of decisive events throughout history, from the Mahabharata to theD-Day landings in Normandy in 1944. There’s been tremendous progress in weather forecasting accuracy in recent times. While some would expect these gains to come from a machine learning (ML) capability that captures complex nonlinear correlations, a bigger reason is the ability to effectively integrate (causal) Ritam into weather calculations: “Meteorologists succeed because they understand the laws of fluid dynamics governing weather patterns and can incorporate these physical laws into their models.”. [11].
Similarly, Modern ML was unable to beat traditional methods at time-series predictions until the fourth (M4) forecasting competition in 2018 [18]. The winning approach combined deep learning with traditional time series forecasting to train an ensemble model. Traditional time-series methods can extract the latent ritamic patterns in human activity (e.g., seasonality of sales or travel) within the training data, at fixed hourly, daily, monthly, or annual time intervals. The origin of time-series forecasting can be traced back to the Vedanga Jyotisha.
Ritamic Forecasting in the Vedanga Jyotisha (VJ)
The Ganita (science of computation) in the VJ is primarily to ensure the timely performance of Yajna that allow us to constructively join and contribute to the cosmic order. VJ’s Ganita is naturally useful for planning decisions as it can track the various natural cycles from intra-day to the VJ yuga level.
A natural byproduct of timekeeping for Yajna is that it aids agricultural policy by helping Indian farmers schedule their activities in sync with the Varsha Rtu, the rainy season and other seasons of the traditional Indian calendar. These primary astronomical calculations of the VJ also enable us to keep track of other ritamic quantities such as tides, monsoon, etc. Note that these calculations are backed by causality and not mere correlationthat drives much of today’s AI/ML. Ritamatic calculations become reliable inputs to traders to refine their planning decisions for domestic and overseas trade missions over land and sea. Of course, such estimates are also useful in planning military campaigns to most effectively thwart invaders. Thus, the performance of Yajna to be in harmony with Rta benefits all sections of the society and the environment and enabled India to remain a global powerhouse for millennia.
Vidura Neeti and Prisoner’s Dilemma
“The Prisoner’s Dilemma (PD) is simply an abstract formulation of some very common and very interesting situations in which what is best for each person individually leads to mutual defection, whereas everyone would have been better off with mutual cooperation”. – [7].
PD is a non-zero-sum game (both opponents can ‘win’ by mutually cooperating) and shows up in a variety of real-life settings, but it was found that there’s no guaranteed a-priori ‘best’ winning strategy. The challenge was to find a practically good decision policy for PD. In 1980, Prof. Robert Axelrod started a computer-based tournament to empirically analyze various strategies for PD. The tournament consisted of a series of PD games played in a round-robin format by participants, each of whom encoded their own favorite strategy into a computer program. Points were gained or lost in each game based on the outcome.
The winner of that tournament was the basic ‘tit for tat’ strategy. This was also the simplest strategy among all entries and outperformed the various enhanced “tit-for-tat 2.0” versions and highly complex policies that had entered the tournament. The reason why Tit-for-Tat won is not simply because of its instant retaliation as many believe, nor purely because of its softness in forgiving (both ‘rigid’ and ‘soft’ as strategies performed poorly), but because the winning version combined strength with softness and was rooted in unchanging clarity.
Simply put, Tit-for-Tat wins consistently when it follows Vidura Neeti:
“What accounts for TIT FOR TAT’s robust success is its combination of being nice, retaliatory, forgiving, and clear. Its niceness prevents it from getting into unnecessary trouble. Its retaliation discourages the other side from persisting whenever defection is tried. Its forgiveness helps restore mutual cooperation. And its clarity makes it intelligible to the other player, thereby eliciting long-term cooperation.
… there is a lesson in the fact that TIT FOR TAT succeeds without doing better than anyone with whom it interacts. It succeeds by eliciting cooperation from others, not by defeating them. “
Note the convergence of the discussion to harmony, which is the essence of Rta.
A soft policy of hoping to be adopted (Ganga-Jamuni tezheeb, Hindi-Chini Bhai-Bhai, etc.,), or being over-smart and opaque (e.g., misuse Arthashastra ideas to out-maneuver the opposition) are recipes for failure in the long term. Neither of them satisfies the Vidura Neeti requirement that is rooted in Vedanta. Non-ritamic strategy that seek a crooked or exploitative ‘edge’ over the ‘other’ may have curb appeal but are myopic and unsustainable. As Axelrod notes: “If you are using a strategy which appears random, then you also appear unresponsive to the other player. If you are unresponsive, then the other player has no incentive to cooperate with you. So being so complex as to be incomprehensible is very dangerous“.
It is worth stating the obvious that in a lifetime we go through entire decision cycles, and this really requires us to adopt a policy that is sustainable. Axelrod gives the example of deals closed amicably in diamond markets only based on verbal agreement: “The key factor is that the participants know they will be dealing with each other again and again.” In fact, Bhaaratiya commerce is the best example of such transparency, and this is how the traditional ‘informal’ sector operates [8], as noted by Prof. P. Kanagasabapathi: ” In India, we see crores of rupees changing hands everyday without any agreements, and a number of businesses running for years with only oral agreements with regard to investments and sharing of profits and losses.”
To summarize, the best decision policy in practice invariably turns out to be that which maximally upholds Rta. What practically works for a diverse and complex group of people is to follow the principle of Sahakarana Dharma, which of course includes the wisdom of Vidura Neeti. This is exactly what the world requires today and this essay on Sahakarana Dharma [19] is important to read and internalize.
While a computer tournament is closer to a simplified world than reality, the next example is a modern business strategy that was applied and found to be practically successful, and we again show how (when executed in accordance with dharma) lasting success emerges from Vidura Neeti.
Simple Rules for a Complex World
“.. simple rules work because they provide a threshold level of structure while leaving ample scope to exercise discretion. Complex rules, in contrast, attempt to anticipate every contingency and dictate what to do in each scenario, thereby reducing people to automatons who do what they are told. But human discretion is not a defect to be eliminated, it is our greatest hope in the battle against complexity” – [11].
In the 1990s, engineering graduate student Shona Brown (later an influential SVP at Google) and her advisor Prof. Kathleen Eisenhardt at Stanford University analyzed the key features of sustainable business strategy in a highly disruption-prone and competitive environment. They found the traditional strategy of long and elaborate planning followed by execution to be rigid and inflexible. At the other end, a purely game theoretical approach produced good competitive responses but lacked a clear long-term goal and structure. They concluded that what worked well is an agility to navigate at the edge of chaos [9], that small region between order (well defined rules and structure) and chaos (flexibility, reaction, improvisation, innovation), i.e., a bi-modal capability. Prof. Eisenhardt later collaborated with Prof. Donald Sull from MIT to assess the performance of a few key decision rules to manage complexity across different domains [11].
“In this kind of agile organization, there are a small number of very tight rules—that is the rigidity—but flexibility otherwise—that is the chaos. The counter-intuitive insight is that simple organizations drive complicated strategies .. companies should avoid the extremes of too much or too little structure ..” – [11].
Note the remarkable similarity to Prajapathi’s self-organized construct.
In a 2001 HBR article [10], the authors give the example of a movie-making company that became successful in an ultra-competitive industry by ensuring that “there are a few rules about movie content. The movie must revolve around some very basic human condition like love …or honesty … The movie needs a coherent story with a clear beginning, middle and end. There is also a firm lid on costs. But beyond these rules, there is wide flexibility in moviemaking..”.
To pick a more recent example, the content produced by big-budget Bollywood cartels is either rigidly formulaic or so bereft of meaningful context and coherence as to appear random to the audience. Contrast this with the simple, rooted yet flexible movie-making emerging from the relatively small-budget and decentralized Indian (i.e., non-Bollywood) film industry, and it is apparent why the latter enjoyed pan-India success in the post-pandemic period.
Unlike Vidura Neeti, the Ivy league authors focus on rules for gain. Ethical restrictions have to be added on subsequently to constrain gain or as some “tolerance” on himsa and crookedness. This is a built-in ethical deficiency in the western lens that invariably leads to what is today labeled the‘alignment problem’ in the AI domain. Similarly, Prof. Taleb adds the ‘Golden Rule’ of Christianity [21] to contain habitual antifragile positioning at the expense of others.
The only way out of this ethical mess as far as I can see is to switch the western order of priority and follow the Bhaaratiya template shown by Vidura Neeti: a mixture of firmness (few simple and transparent ground rules rooted in integrity) and flexibility to take context-specific decisions that promote harmony, with gain as a byproduct, i.e. shubh laabh.
Swami Jitatmananda has explained this Indic view of ethics [1] that radically differs from the western lens (there is no ‘other’ here): “The final Rta, the socio-moral order, is to Vedic vision, always based on the essential divinity and the underlying unity of all life. What is ethics but acceptance of the unity of existence?”.
It must also be pointed out that the metaphor of ‘Improv’ used by the authors more accurately describes Sangeetam, Indian music. They either do not reveal this fact or are entirely unaware of theIndian origins of music and improv.
“Improv is a music form that has two central characteristics: specification of a few crucial rules, often around who plays first and permitted chords, and a high amount of real time communication. The result (especially for skilled musicians) is innovative, high quality music.”.
If they had realized this fact and explored Sangeetam, they would have also been able to address the weak ‘alignment’ of strategy with ethics. The motivation of Sangeetamand western music are not the same. Troy Organ’s explanation of these differences [15] is given below and he remarked on this stark contrast in all art forms and fields of human endeavor:
“Raga and tala constitute the invariable; the musicians supply the variable. The spectators are privileged to watch a cooperative and some times competitive effort to bring forth from the given raga and tala every conceivable variation. Indian music is far more than disciplined sound; it is a revealing of the pluralities within oneness. .. In architecture, in poetry, in music, in sculpture, and in every conceivable dimension of Hindu life the theme is repeated.”
On a lighter note, Sangeetam performances can also include a musical tit-for-tat!
Strengthening Ritamic Decision Making Capability: Vratam
“In order to be in harmony with ritam, an individual, as well as a society, must strive for the four pursuits known as purusharthas: dharma, artha, kama, and moksha. – [12].
Dharma also teaches us how to develop the strength of mind and body to pursue decision policies that maximally uphold Rta. When the intent behind choices is aligned with our Purusharthas, we can expect good things to happen. The ‘good’ here means that which is good, right, and just for all, and not just us or our group or organization [2].
Vrata or right conduct strengthens our ability to consistently take the right decision and accept the outcome. Conduct that continually adheres to Rta is vrata. Varuna is revered as dhrtavrata for holding steadfast and steering the moral, cosmic laws [1]. Note that like Yajna, the various vrata followed by dharmikas synchronize with natural cycles and are not opportunistically performed for gain or antifragility (see part 3).
This Yajurvedic verse [16] shows the path one can follow. It begins with the observance of Rta (vrata) that prepares us for deeksha, which builds shraddha within that gradually aligns us to the one supreme reality, Sathya.
Acknowledgment: thanks to Nripathi garu for reviewing this article and his valuable feedback.
References and Further Reading:
Rta: The Cosmic Order; Edited by Madhu Khanna. DK Print World Pvt. Ltd. 2004.
Troy W. Organ. The Hindu quest for the perfection of man. Wipf and Stock Publishers, 1998.
Vedic Hindu Portal. vedicheritage.gov.in.
Brian K. Smith. Reflections on resemblance, ritual, and religion. Oxford University Press. 1989.
Spyros Makridakis, Evangelos Spiliotis, and Vassilios Assimakopoulos. The M4 Competition: Results, findings, conclusion and way forward. International Journal of Forecasting (Vol 34). 2018.
By Ramanarayanadatta Shastri (Public Domain, Link)
