Singapore’s educational reforms toward holistic outcomes

The paradoxes of student well-being in Singapore

Sentiment analysis and opinion mining on educational data - A survey

Changing assessments and the examination culture in Singapore


Using social media data from 2018 to 2025, we show a cyclical change in polarisation between academic vs holistic mindsets towards education amongst students in Singapore.

<PolarisationPolarityLiveline />

# Singapore's education reform

Human capital development through education is one of Singapore's top priorities. Though the nation has largely achieved success in developing an education system globally renowned for its academic rigour and performance [@ng-2020], many Singaporean students are highly stressed and anxious. In Organisation for Economic Co-operation and Development (OECD) comparative study, 86% of Singaporean students were worried about poor grades at school, compared with the OECD average of 66%. 76% of Singapore's students were anxious about a test even if they were well prepared, compared with the OECD average of 55% [@ng-2020]. At the same time, with shifting socio-economic needs of the relatively young nation, Singapore's education policies have shifted focus from educational development (1863 - 1997), economic development (1941 - 1965), social development (1941) to holistic development (1998) [@wong-kwek-tan-2020].

However, despite the official policy narrative of "holistic education", the narrow mindsets of success and culture of high-stakes examinations persist. This is amplified by a strong "shadow education system" of private tuition and intense parental expectations for top academic performance [@kwek-ho-wong-2023]. There is a disconnect between the education system's holistic goals and the reality of students' competitive academic goals.

# Measuring education reform

To solve this, policymakers need a better way to measure the impact of education policies. 

How do we measure the progress of holistic education? Is it possible to measure the shifts in mindsets of students across the entire population of Singapore? This is where sentiment analysis provides a powerful solution. By using computational methods to analyse large volumes of text - such as student feedback, online forums, and social media discourse - we can identify the underlying emotions and opinions being expressed by students [@shaik-et-al-2023].

Tracking these sentiments offers direct insights into student well-being and the true impact of educational policies. Such data informs more effective strategies for better development of reform policies and pedagogy, helping to create an educational environment in Singapore that is not only academically successful but also genuinely supportive of every student's holistic needs.



Quantifying mindset shifts in Singapore

Frames, biases, and rational decision-making in the human brain

The neurobiology of reference-dependent value computation

Medial orbitofrontal cortex codes relative rather than absolute value of financial rewards in humans

Prospect theory - An analysis of decision under risk

The neurobiology of context-dependent valuation and choice

Neural processes mediating contextual influences on human choice behaviour

The neural basis of loss aversion in decision-making under risk

Rational choice and the framing of decisions

Primate anterior insular cortex represents economic decision variables proposed by prospect theory


# Introduction
Humans are often not fully rational economic agents. We make choices that diverge from classical economic predictions and are strongly influenced by the context in which options are presented. Why do we choose to gamble when an option is framed as a potential loss, but play it safe when the same outcome is framed as a potential gain? Behavioral economics describes this with prospect theory [@kahneman-tversky-1979], which emphasizes three core principles: reference points, loss aversion, and framing effects [@tversky-kahneman-1989].

Prospect theory is powerful descriptively, but it does not by itself specify a neural mechanism. Context-dependent neural coding offers that missing mechanistic layer: the brain encodes value relative to surrounding context, such as available alternatives and internal state [@louie-demartino-2014]. This essay reviews work in neuroeconomics to show how context-dependent coding helps explain the same behavioral regularities captured by prospect theory.

# Reference point
Context-dependent neural coding explains the reference point principle. Prospect theory argues that outcomes are evaluated relative to a baseline rather than in absolute terms [@kahneman-tversky-1979]. Likewise, valuation in the brain appears relative rather than fixed. Neural systems can encode a reward differently depending on the local choice context [@demartino-2009;@elliott-2008].

This helps explain unstable preferences. A $50 reward can feel highly attractive when paired with $10, but less attractive when paired with $100. Brain regions involved in contextual processing and valuation appear to support this dynamic recalibration.

## Relative vs. absolute value coding
Elliott et al. tested this directly with fMRI by teaching participants to associate abstract cues with low, medium, and high monetary rewards [@elliott-2008]. Critically, the medium cue represented the same absolute payoff in both conditions, but its relative standing changed depending on the paired alternative.

Activity in medial orbitofrontal cortex (OFC) was higher when the medium cue was the better option than when it was the worse option. Because stimulus identity, response demands, and objective reward magnitude were held constant, this pattern supports relative value coding over absolute value coding.

## Neural coding of the reference point
Related work identifies specific neural substrates that may instantiate reference points. In macaques, anterior insular cortex (AIC) neurons encode current token wealth, effectively tracking an internal baseline for subsequent choices [@yang-2022]. In humans, hippocampal activity increases when contextual reward information is introduced, and this activation relates to contextual influence on behavior [@rigoli-2016].

Together, these findings move beyond the claim that value is relative. They suggest that particular circuits contribute to representing and updating the baseline against which outcomes are evaluated.

# Loss aversion
Context-dependent coding also helps explain loss aversion, the well-known asymmetry where losses loom larger than equivalent gains [@kahneman-tversky-1979]. This is behaviorally visible when people reject positive expected-value gambles because potential losses are weighted more heavily than potential gains.

Tom et al. examined this using fMRI while participants accepted or rejected mixed gambles [@tom-2007]. Activity in ventral striatum and ventromedial prefrontal cortex (vmPFC) increased with prospective gains and decreased with prospective losses. The decline for losses was steeper than the increase for gains, a pattern they characterized as neural loss aversion. Individual differences in this neural asymmetry predicted individual differences in behavioral loss aversion.

Conceptual visualization: this curve illustrates steeper subjective weighting for losses than gains.

<LossAversionDemo />

Comparable findings in valuation tasks, including ownership-based distortions between willingness to accept and willingness to pay, further support a neural basis for asymmetric valuation [@demartino-2009]. In non-human primates, AIC populations also show disproportionate sensitivity to loss-related signals [@yang-2022]. Taken together, these data align prospect theory's asymmetric value function with asymmetries in neural coding.

# Framing effect
The framing effect is another context-driven deviation from classical rational choice: equivalent outcomes can lead to different decisions depending on whether they are described as gains or losses [@tversky-kahneman-1989].

De Martino et al. used fMRI in a gain/loss framing paradigm with equivalent expected outcomes [@demartino-2006]. Choices consistent with framing bias were associated with greater amygdala activity. By contrast, choices resisting the framing bias recruited regulatory regions, including anterior cingulate cortex (ACC), and lower susceptibility was linked to stronger orbitofrontal and medial prefrontal involvement.

These findings suggest that framing does not simply alter verbal description. It changes neural valuation dynamics, creating predictable shifts in value-based behavior.

# Conclusion
Context-dependent neural coding provides a plausible neurobiological bridge between descriptive behavioral economics and brain-level implementation. Across reference dependence, loss aversion, and framing, the literature converges on the idea that valuation is not static or absolute; it is context-sensitive and neurally instantiated [@louie-demartino-2014].

Prospect theory explains what people tend to do under risk. Neuroeconomics helps explain how those patterns emerge from the interaction of contextual signals, valuation systems, and control processes. Linking these levels improves our understanding of why economic decisions systematically depart from normative rational models.


Neurobiology of prospect theory

A thousand empirical adaptive landscapes and their navigability

Career as a gradient descent problem

Algorithms to Live By

Climbing the wrong hill

Recombination and peak jumping

Developmental changes in exploration resemble stochastic optimisation

Hill climbing

Hill-climbing theories of learning

Rational life plans?

Hot streaks in artistic, cultural, and scientific careers

Understanding the onset of hot streaks across artistic, cultural, and scientific careers

The geometry of evolution

The pursuit of happiness in history

Critical core skills that employers want

Evolution on rugged adaptive landscapes

Simulated annealing: A life framework

Temporal discounting

Temporal discounting correlates with directed exploration but not with random exploration

Life is gradient descent

Bounded rationality

Simulated annealing beats metropolis in combinatorial optimisation

_(Note: This post will be continuously updated as I learn more about the world.)_

This project started as a final essay submission for my first semester in college, under the Module "Big Questions: Happiness & Suffering". I am still somewhat young and have much more to learn, so my opinions on this matter will likely change over time. Re-writing this in first-person, not to be taken as an objective paper.

# Introduction
In recent studies on ‘hot streaks’, periods of peak performance, of over 20,000 artists, film directors, and scientists, researchers found that ‘hot streaks’ seem to occur randomly within a career [@liu-2017].

In my own life, in hindsight, most of the best things that happened to me were unplanned. Heck, I never even knew they existed.

We may find these random events (good or bad) changing the entire trajectory of our lives, leading us to states of greater happiness or success. We rationalise this in different ways. Some of us attribute it to God and higher beings, while some call it ‘luck’ or ‘serendipity’. However, that explanation might be not be good enough for some of us. What if there was a more quantitative way to understand how randomness contributes to human flourishing?

## From faith to formulas.
The field of computer science offers a promising model for explanation. Several authors proposed that the way humans learn and make life decisions, such as careers, is an iterative optimisation problem that is similar to the **hill-climbing** algorithm [@career-gradient-descent-2023;@climbing-wrong-hill-2009;@hill-climbing-2021;@langley-1987;@qu-2019;@tang-2021].

Imagine standing in a foggy mountain range, trying to reach the highest peak. Since it's foggy, we can only see our immediate surroundings. The rational strategy would be to keep walking uphill until we can't go any higher. This mirrors how we typically make decisions too. With limited cognitive resources and information, we rely on heuristics to make these decisions. We evaluate our next available options and make incremental improvements based on what we can observe over time, much like the **hill-climbing** algorithm.

## Limits of **simple hill-climbing**.
However, this approach to decision-making has a crucial limitation – the problem of local peaks. 

Consider how we approach career decisions. We may look at our current situation and nearby options - perhaps a promotion at our company or a similar role at another firm. We evaluate these options based on factors like salary, work-life balance, and personal interest. This is like climbing a hill by always taking steps upward. But just as a hiker might reach a small peak while missing a much higher mountain hidden in the fog, humans can become trapped on lower local peaks while missing higher global peaks. 

We might settle into a comfortable job while missing more fulfilling opportunities or maintain habits that provide immediate satisfaction while preventing greater long-term well-being. This is where randomness plays its crucial role, and the **stochastic hill-climbing** algorithm - an upgrade to **simple hill-climbing** - provides us with insights into how randomness can help us better flourish in life.

<LocalVsGlobalDemo />

## Escaping from local peaks.
To solve this problem, computer scientists suggest that we may need to temporarily worsen our solution if we want to continue searching for improvements, and randomness offers a few strategies for doing so: **jitter**, **random-restart hill-climbing**, and the **metropolis algorithm** [@christian-griffiths-2016].

**jitter** suggests for us to make random small changes to our lives when we are stuck, even if they temporarily worsen our level of well-being, and then continue **hill-climbing**. **random-restart hill-climbing**, as its name suggests, is to completely scramble our combinations of choices when we reach a local peak and restart again. Lastly, the **metropolis algorithm** suggests applying a small amount of randomness to every decision we make. This probability of making a random decision is controlled by a variable called temperature.

<RandomRestartDemo />

<MetropolisDemo />

# Empirical observations (maybe)
Interestingly, these **stochastic hill-climbing** strategies can be somewhat observed empirically and affect well-being on multiple levels: individually and evolutionary.

## Individual well-being.
**random-restart.** Research on over 20,000 artists, directors, and scientists reveals that "hot streaks" - periods of exceptional performance - follow patterns consistent with stochastic optimisation. Peak performance emerges from a specific sequence: extensive exploration across diverse fields followed by focused exploitation of promising areas [@liu-2021]. This mirrors **random-restart hill-climbing**, where individuals sample multiple "hills" before concentrating on the most promising peak.

**metropolis algorithm.** The **metropolis algorithm** parallels the entropy patterns surrounding these hot streaks too. Before peak periods, individuals exhibit high unpredictability in their work, which decreases significantly during the streak as they focus narrowly - resembling the temperature reduction in simulated annealing optimisation [@wegener-2005].

## Evolutionary well-being.
**jitters.** **stochastic hill-climbing** finds parallel support in biological evolution through adaptive landscapes. Organisms navigate fitness landscapes where height represents survival advantage, climbing via natural selection driven by genetic recombination and random mutations - analogous to algorithmic "jitters" [@mcghee-2006]. When trapped at local fitness peaks, organisms may also escape through peak jumping or environmental changes that reshape the landscape itself.

It does seem that this pattern - in biological evolution to human careers - suggests a fundamental need for stochastic optimisation in complex, uncertain environments.

# We can't jump alone.
If exploration through randomness is beneficial, why don't we simply choose to be more random in our decisions? The answer might lie in behavioral barriers that prevent self-directed exploration.

Temporal discounting - our tendency to overvalue immediate rewards compared to future benefits - creates a systematic bias against voluntary exploration [@reed-luiselli-2011]. Research shows this phenomenon correlates significantly with reduced directed exploration [@sadeghiyeh-2020]. Even when we intellectually understand exploration's value, we struggle to voluntarily leave comfortable situations. Chris Dixon's example illustrates this: his acquaintance knows joining a startup aligns with his dreams yet remains trapped in an unfulfilling finance job due to immediate security concerns [@climbing-wrong-hill-2009].

This explains why externally imposed randomness often proves crucial for escaping local peaks. Environmental disruptions - whether evolutionary pressures, global pandemics, or personal upheavals like job losses - force exploration that behavioral biases would otherwise prevent. These external forces compel us out of lower-peak situations, enabling the discovery of higher peaks we might never have voluntarily pursued.


# Limitations
## Defining the goal.
This whole time, I used ‘well-being’ as a general term to describe the outcome humans are optimising through our life choices. However, this very idea of ‘well-being’ itself is subject to individual goals and values. While we have explored the optimisation of ‘well-being’ through the lenses of individuals and evolutionary biology using the concepts of ‘hot streaks’ and organisms’ fitness to survive ecologically, for one to effectively apply this model to their individual lives, we need to consider how success or well-being is defined for ourselves.

## Assumption of motivation.
This model of **stochastic hill-climbing** assumes the desire to climb the hill. It assumes the individual making decisions has the motivation to optimise their well-being. ‘Hot streaks’ would not be possible if the artist, film director, or scientist had no motivation to even produce works. On the other hand, adaptive landscapes assume that the organism is influenced by natural selection and an innate or biological desire to survive. 

Notably, the idea of randomness contributing to flourishing also depends on our ability to adapt and climb the hill from new locations. If random events were to throw us off our current hills, are we capable of climbing back up? Similarly, as the environment changes, which it rapidly does as we often observe in the world, are we capable of identifying these changes, the positions we are in, and subsequently learning to adapt to the new environment? Behavioural traits like learning agility [@myskillsfuture-2024], trait curiosity, and resilience will be necessary for individuals to maintain or improve their well-being in these dynamic environments. Perhaps it hints at how education systems around the world, like Singapore [@ng-2020], should adjust curriculums to better facilitate the development of these traits amongst students.

## Too many dimensions.
In adaptive landscapes, a criticism is raised by R.A. Fisher on the complexity of dimensionality. The probability of local peaks decreases with increasing dimensions in a high-dimensional genotype space [@obolski-2017]. As individuals, it might be difficult to mentally evaluate our well-being simply because there are so many dimensions that influence one another. Similarly, our choices in life are highly interdependent, be it due to opportunity costs or other relationships. A simple choice of going to the gym may increase our physical well-being but lead to opportunity costs of time and result in lower capacities in other life commitments.

A possible step would be to operationalise specific dimensions of well-being clearly. For instance, narrowing the dimensions down only to the possible career options and salary. While such models can be narrower in scope, they may be more practical for evaluation.

# Conclusion
**stochastic hill-climbing** offers a unique computational lens for understanding how randomness contributes to human flourishing. While it may not capture every aspect of well-being yet, it provides a promising framework for understanding how unexpected events and enforced changes, despite their initial challenges, can lead to improved life circumstances.

In our quest for well-being or 'success', perhaps we should view life's randomness not as an obstacle to overcome but as a feature that helps us escape the limitations of our bounded rationality and discover new peaks of flourishing.


Role of randomness in human flourishing

Hi. I'm Ray.

About Me

Experience

United Nations, UNDP Bangkok Regional Hub

CORDY

National University of Singapore

Singapore Armed Forces

Education

Some encouragement from others

Some projects that I happen to have the time to write about: