Arghya

Infinite Shades of Mind: The Battle Between Human Feelings and AI

2026-02-08T00:00:00+00:00

Many of us live multiple lives at multiple places with the same identities. We want to pose ourselves in certain forms to benefit from the respective environments. However, every action leaves traces of changes in our DNA over time. Over a long period, we become what we went through, and we become what we wanted to be at different places and different stages of our lives.

Our mind is nothing but a reflection of our past actions and our life is the mirror of our mind.

Figure: The interplay between human consciousness and artificial intelligence

The Age of Information Overload

In the age of social media, the dynamic interactions among many individuals create a pool of diverse datasets in our memories. As the volume of data grows, our ability to utilize such extensive data drops gradually. As a result, a time comes when we start to get saturated, and our actions become numb, unilateral. We are on the verge of an addiction—a form of destruction of self-identity.

The Post-GPT Era

In the post-GPT era, people have started to realize the mammoth power of learning. Learning from a pool of data without any emotions is something that differentiates humans from AI models. And when machines learn more and more about human actions, they achieve human strengths and capabilities over time. However, the unsolved puzzle of ‘feelings’ is still the separator.

Our Competitive Advantage

One of the ways we can outperform such gigantic models of AI is when we can truly control our emotions/feelings and channelize our thought process to create a pool of actions that have a basis of rightful ‘feelings’.

In other words, our every action should be linked to a positive feeling so that models of AI fail to recognize any logical basis of such actions. It is a mapping between a pool of actions and a pool of positive feelings.

The Double-Edged Sword

On the contrary, our biggest strength is our biggest enemy. A significant portion of people on Earth do not realize they are gradually losing control of their thought processes over these models of AI in the form of Facebook, Instagram, Quora, LinkedIn, and many more.

The unfortunate thing is that no one came up with another powerful model that can help humans to recognize and reinstate the power of ‘feelings’ as an antidote to the AI models.

But we need such an antidote soon before the AI models completely take over.

— Arghya

Your thoughts? How do you preserve your authentic feelings in this AI-driven world? Share in the comments below.

Citation:

Mallick, A. (2026). Infinite Shades of Mind: The Battle Between Human Feelings and AI. 
Retrieved from https://amallick-tudelft.github.io/posts/2025/02/infinite-shades-of-mind/

Contributory Peer Review: A Fairer Academic Review System

2025-02-08T00:00:00+00:00

In this article, I propose a peer-review mechanism to provide a better alternative to the existing fragile peer review system in academia. In the following, I provide an equitable framework to assign reviewers of the submitted articles to a journal.

Let me first define a few notations that I will use later in this article (the reader can directly jump to the methodology section and come back here once required).

Suppose $N$ researchers write an article together and decide to submit it to a journal of their choice.

Notation

Author’s score: $a_i \in \mathbb{R}$ where $i \in [N]$ with $N$ being the total number of authors in a submitted article. This score signifies the expertise level of an academic researcher in a particular research area.
Expertise threshold: $a'$ - A threshold level of expertise of a researcher in a research area as decided by the editorial board of a journal.
Review criterion: $c_i \in \mathbb{R}$ where $i \in [N]$. It will be used as a criterion for evaluating the author’s participation level in the peer review process.
Author’s contribution level: $v_i \in (0,1]$ where $i \in [N]$. This is the level of the author’s contribution to creating a research article, and it will be provided by the authors while submitting it to any journal.
Historical contribution average: $v'_i$ - the historical average of contribution level for author $i$ to the journal.
Total submissions: $p_i$ - total number of articles submitted by $i$ to a journal.
Total publications: $x_i$ - total number of articles published by $i$ in the same journal.
Historical review score: $r'_i$ - the historical average review score in $[0,1]$ received for $p_i$ articles of author $i$. The editors grade the quality of each review submitted by the $i$-th author and update $r'_i$.

We define the author’s score as:

\[a_i = \left(\frac{x_i}{p_i}\right) \times r'_i\]

Methodology

The main idea of the Contributory Peer Review (CPR) mechanism is that every author is obliged to contribute to the peer review system of the journal where he/she wants to submit a research article. In other words, the submitted article will enter the peer review process of a journal only if the authors agree to contribute to the peer review system of that journal. I will now particularly focus on developing an algorithm to automate the reviewer allocation process among the authors.

Algorithm

The following algorithm assigns research articles among the authors who have submitted their article to a journal.

Input: $\{v_i, v'_i, a_i, c_i, r'_i\}, \forall i \in [N]$

 for i in [N]:
     if a_i > a':
         c_i = c_i + v_i + v'_i × (x_i / p_i)
         if c_i ≥ 1:
             Send one article from similar research area to author i for reviewing
             Set c_i = 0
         end if
     end if
 end for

Update: Update and store $\{v'_i, p_i, a_i, c_i, r'_i\}$.

Rationale

The rationale of line 2 in the algorithm is obvious: a novice researcher should not review a paper. The idea behind the update of $c_i$ in line 3 of the algorithm can be understood in two steps:

First part: The level of contribution for author $i$ of the submitted article should be proportionally reflected in his/her level of contribution in the peer review system.
Second part: The more an author’s articles get published in a journal, the more he/she should review other articles.

Summary

The above idea, if implemented with fairness, is expected to greatly regulate the scarcity of reviewers. Any further ideas for improvement are welcome in the comment section.

Note: Please cite this article if you use it for further research.

Citation:

Mallick, A. (2025). Contributory Peer Review: A Fairer Academic Review System. 
Retrieved from https://amallick-tudelft.github.io/posts/2025/02/contributory-peer-review/

Global Productivity Loss from Watching Reels and Its Cure

2025-02-08T00:00:00+00:00

With the rapid development of learning-based technologies, the reel industry (YouTube, Facebook, Instagram, etc.) has taken the world into the booze of cocaine-like drugs. Even the best minds of a country are regular customers for hours every day. Let us first understand mathematically (in a very non-rigorous way) why reels are so addictive.

The Mathematical Framework

Consider the average person watching reels as the Audience, and there is an Agent (the drug lord) who has designed an algorithm $F(\cdot|H): \mathcal{Z} \to \mathcal{Y}$, where $H$ is history, $\mathcal{Z}$ is the set of design parameters, and $\mathcal{Y}$ is the set of strategies. Step by step, let us define the sequence of events that take place while watching reels:

Event Sequence

1. Content Generation

The Agent runs the secret algorithm and generates a particular reel at time $t$ (say $x_t$) from a pool of finitely many reels (say, set $\mathcal{X}$).

2. Audience Interaction

After the audience watches it, they give likes/comments/swipes to the next reel without watching it in full. All of these actions of the audience are formulated in terms of rewards $r_t(x_t)$ at time $t$. For example, for a given $x_t$:

‘Like’ $: r_t(x_t) = 1$
‘Swiping without watching’ $: r_t(x_t) = 0$

Note that the Agent always wants to maximize the reward.

3. Regret Formulation

We define the notion of regret ($R_T$) after $T$ rounds as:

\[R_T = \max_{x \in \mathcal{X}} \left(\sum_{t=1}^{T} r_t(x)\right) - \mathbb{E}\left[\sum_{t=1}^{T} r_t(x_t)\right]\]

where:

The first term is the best possible achievable cumulative reward (if the Agent could have known the reward function $r_t(\cdot)$ beforehand every time; in other words, if the agent had known what the audience likes most before showing a reel)
The second term is the actual realized cumulative reward after $T$ rounds

The Addiction Mechanism

The goal of the Agent is to develop an algorithm where $R_T$ grows sub-linearly with $T$, i.e., in the order of $\mathcal{O}(\log T)$ (for example).

Note that if $R_T$ grows sub-linearly with $T$, it implies asymptotically that $R_T$ is almost constant, which means every new $x_t$, as given by the algorithm, is approximately the optimal one. When this happens, it means you like almost every reel you are watching and are under the arrest of an addiction.

The drug lords know the secret formula to design such algorithms that want to steal away your precious time by engaging you in reels.

Proposed Solution

With every technology, a grave amount of responsibility comes into the hands of the innovators. Therefore, each user must be given a hyper-parameter (say $p$), which will signify the time control for watching reels on any platform.

Implementation

Once a user sets $p = 1$ hour (for example), the Agent behind the algorithm changes its strategy after 1 hour to maximize the Regret ($R_T$), which will eventually show such reels that the user starts to dislike and stay away from watching it further.

This simple modification transforms the optimization objective from:

\[\text{Current: } \min R_T \quad \text{(minimize regret = maximize engagement)}\]

to:

\[\text{Proposed: } \max R_T \quad \text{(maximize regret = minimize engagement after time } p\text{)}\]

Summary

By introducing user-controlled time limits that trigger a strategic shift in the recommendation algorithm, we can transform addictive engagement mechanisms into self-limiting systems. This approach respects user autonomy while addressing the societal problem of attention exploitation.

Citation:

Mallick, A. (2025). Global Productivity Loss from Watching Reels and Its Cure. 
Retrieved from https://amallick-tudelft.github.io/posts/2025/02/global-productivity-loss-reels/

What are your thoughts on this proposal? Share your ideas in the comments below.