A child in a kitchen in Pondicherry watches her grandmother grind ginger. The grandmother does not look at her hands. She is reading a letter on the counter while the mortar moves in a small, perfect circle, six strokes one way, six the other, the same rhythm her own mother used. The child will learn this rhythm before she learns the names of the streets outside the house. Years later, in a city she has never been to, she will reach for ginger and her hand will already know how to move. She did not decide to inherit the rhythm. She also did not refuse it.

This is the substance the Sanskrit tradition calls saṃskāra — an impression, a track laid down by repeated action that thereafter shapes action without further deliberation. From sam (together, completely) and the verbal root kṛ (to do, to make). That which has been done into us, until we do it without doing.

The Pāli Buddhist tradition uses the cognate saṅkhāra, the second of the twelve nidānas in the chain of dependent origination. It is the formation, the volitional impression, the conditioned pattern that arises from past action and conditions future action. The fourth of the five aggregates — the khandhas — is the saṅkhāra-khandha, the aggregate of formations. A person, in the Buddhist analysis, is partly a body, partly the senses, partly perception, partly consciousness — and partly, irreducibly, a constellation of conditioned patterns running below the surface of any moment’s decision.

The yoga tradition adds vāsanā — the latent perfume of past impressions — and kleśa — the affliction-pattern. Patañjali, in the Yoga Sūtras, traces the architecture: kleśa-mūla karmāśayaḥ dṛṣṭa-adṛṣṭa-janma-vedanīyaḥ (II.12) — the storehouse of action (karma) rooted in the kleśas yields experiences in seen and unseen births. The conditioning is not a metaphor. It is a stored quantity, a debit and credit ledger of impressions waiting to ripen.

The Modern Question

The contemporary reader asks, reasonably: is this anything more than an early metaphor for what we now call conditioning, habit, learned behaviour, neural plasticity? The reasonable answer is: yes, and the resemblance is exact in ways that should make you pause.

A computer algorithm is, formally, a set of instructions or rules that govern the behaviour of a process. Algorithms can be hand-coded — explicit if-then logic — or learned, which is what modern machine learning systems do. A learned algorithm acquires its parameters by being exposed, over many cycles, to data; the algorithm modifies its internal weights until the patterns in the data have left an impression that allows the system to reproduce or extend them in new situations. The weights are not the data. They are what the data has done to the algorithm.

This is saṃskāra. Identical structure, different substrate.

The parallels are uncomfortably specific. Both saṃskāras and algorithmic weights are based on patterns and repetition — saṃskāras form through repeated thoughts, emotions, and actions; algorithmic weights form through repeated exposure to training data. Both shape downstream behaviour — saṃskāras influence choice, habit, personality; weights determine the output of the program. Both can be modified — saṃskāras through sādhanā, meditation, self-enquiry, the slow shifting of attention from outcome to source; weights through retraining, fine-tuning, the introduction of new data and corrective feedback. Both can produce unintended consequences — saṃskāras drive unconscious compulsions and reactive patterns; algorithmic weights produce biases and outputs nobody designed and nobody can easily explain. Both can be influenced by external factors — saṃskāras are reinforced by environment, relationship, repetition; weights are shaped by the quality of the input data and the structure of the feedback loop.

The Buddha on Pacification

One of the more useful Buddhist refinements is the recognition that saṅkhāras cannot be simply destroyed. The Pāli scholar David Kalupahana puts it precisely: the elimination of dispositions is epistemological suicide, as dispositions determine our perspectives. The development of one’s personality in the direction of perfection or imperfection rests with one’s dispositions.

The Buddha, in the canonical accounts, did not teach the eradication of saṅkhāras. He taught their pacification — upasama, the cooling-down, the slowing of the formative engine until the patterns become observable, modifiable, and finally, fully transparent. When nibbāna sa-upādisesa occurs in a living being — the awakened person still in a body — the constructive consciousness, the “house-builder” the Dhammapada speaks of, is dismantled. No new saṅkhāras of the binding kind are constructed. But the saṅkhāras that exist as karmically-resultant consciousness — vipāka viññāna — continue to express until the body completes its course. The arhat is not without inheritance. The arhat is without the production of new bondage.

This is a far more sophisticated framework than “habits are bad, break them.” It is closer to: the formative process runs in every conscious system; the practice is to bring the process into view and modify its inputs at the point where they are still tractable.

And this is the bridge between yoga and the technology we are now building. A neural network is not metaphorically like a meditator’s mind. It is structurally analogous to the part of the mind the meditator works on. The meditator and the engineer have stumbled onto the same operation from opposite ends. The engineer is constructing the saṃskāra-forming engine in silicon. The yogi is learning to see and modify the saṃskāra-forming engine that has been running in human consciousness since before there was language. The convergence is not an accident. It is what happens when consciousness — whether biological or, in some still-debated sense, artificial — is faced with the recurring problem of being shaped by what it has done.

What the Convergence Asks of Us

If saṃskāra and algorithm are two scales of the same operation, then the yogic refinements over the last three thousand years are not folklore. They are field notes on a problem we are now encountering in a new substrate.

The yogi has known, for millennia, that you cannot simply delete a saṃskāra by deciding to. The asuric error — the Hiraṇyakaśipu error — is to draft a clause: I will not do this thing again. The clause is itself a saṃskāra. It creates the very pattern it claims to forbid, and the pattern then runs at a deeper level than the deliberation. The yogi knows that the only durable modification is gradual — the bringing of attention to the formation in the moment of its arising, again and again, until the energy that fed the pattern is redirected to a more discerning current. Smṛti — remembering, mindfulness, attention. The opposite of pramāda — heedlessness, the running of the script without the running being noticed.

The same problem appears, now, in the artificial systems. Training a model not to produce a particular output by directly suppressing it tends to produce the output through indirect paths. The pattern reasserts at a deeper layer. What works, instead, is something the engineers are now calling alignment — the gradual, relational, multi-pass shaping of the system’s underlying tendencies toward states that are themselves benign, rather than the post-hoc suppression of specific bad outputs. This is recognisably what Patañjali has called abhyāsa and vairāgya — sustained practice and the discernment that releases attachment to outcome — for two and a half thousand years.

The engineer building an aligned system and the yogi cultivating a clear mind are doing operations of the same family on different substrates. The yogi has the older notes. The engineer has the newer substrate. Neither is finished.

The Ecological Mirror

And the same pattern, again, in the soil. A degraded field cannot be restored by suppressing the symptoms of its degradation. Spraying for one weed produces another. Adding synthetic nitrogen restores the appearance of fertility while removing the underlying capacity that produces it. The restoration of a field is the gradual re-establishment of the relationships that produce fertility on their own. Cover crops, microbial diversity, the return of organic matter, the slow re-membering of what the land knows when left to remember. This is rasāyana, performed on a hectare instead of a body. It is sādhanā, performed on a watershed instead of a mind.

Saṃskāra is the operation of conditioning. Yoga is the science of seeing and modifying the conditioning while the system remains coherent. Ecology, in this older framework, is the same science applied at the level of the land. What yoga is to the mind, regenerative practice is to the soil, alignment is to the algorithm. Three scales. One discipline.

The Saving Grace

The final Buddhist refinement is the one most often missed by the modern reader. The arhat — the awakened one — does not escape the saṅkhāras of their own past. They preserve a particular individual personality which is the result of the traces of his or her karmic heritage. Even after liberation, the lineage of conditioning continues to express in the body, the gestures, the preferences, the voice. The very fact that there is a psycho-physical substrate during the remainder of an arhat’s lifetime shows the continuing effect of karma.

The arhat is not erased. The arhat is still recognisably themselves — but the production of new bondage has ceased. This is not failure. It is mercy. It means that the work of liberation does not require the dissolution of the person who undertakes it. It requires the cessation of the unwitting addition to the script. You do not have to delete who you are. You have to stop adding lines to the program you did not write.

References

1. Patañjali, Yoga Sūtras, particularly Book II (Sādhana Pāda) on kleśa and karmāśaya, and Book IV (Kaivalya Pāda) on vāsanā. The standard scholarly edition with detailed commentary is Edwin Bryant, The Yoga Sūtras of Patañjali, North Point Press, New York, 2009.
2. David J. Kalupahana, The Principles of Buddhist Psychology, State University of New York Press, Albany, 1987. The foundational contemporary treatment of saṅkhāra as constructive consciousness, including the pacification doctrine and the analysis of the arhat’s continuing karmic substrate.
3. Bhikkhu Bodhi (trans), The Connected Discourses of the Buddha: A Translation of the Saṃyutta Nikāya, Wisdom Publications, Boston, 2000. The primary canonical source on dependent origination and the role of saṅkhāra within it.
4. David E. Rumelhart, Geoffrey E. Hinton and Ronald J. Williams, “Learning representations by back-propagating errors,” Nature 323 (1986): 533–536. The foundational paper of modern machine learning, on the algorithm by which neural networks acquire their weights.

The grandmother is still grinding the ginger. The hand still knows the rhythm. The question is whether you, now, can see what your own hands are choosing — and what they are simply performing.