<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Anurag Adarsh on Medium]]></title>
        <description><![CDATA[Stories by Anurag Adarsh on Medium]]></description>
        <link>https://medium.com/@anuad?source=rss-67bcc691e747------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*dmbNkD5D-u45r44go_cf0g.png</url>
            <title>Stories by Anurag Adarsh on Medium</title>
            <link>https://medium.com/@anuad?source=rss-67bcc691e747------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 06:24:13 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@anuad/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Reducing uncertainty in a data science defined product]]></title>
            <link>https://medium.com/@anuad/reducing-uncertainty-in-a-data-science-defined-product-db511b0fcde2?source=rss-67bcc691e747------2</link>
            <guid isPermaLink="false">https://medium.com/p/db511b0fcde2</guid>
            <category><![CDATA[data-science-development]]></category>
            <category><![CDATA[mvp]]></category>
            <category><![CDATA[ml-product]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[product-development]]></category>
            <dc:creator><![CDATA[Anurag Adarsh]]></dc:creator>
            <pubDate>Tue, 20 Feb 2024 03:47:04 GMT</pubDate>
            <atom:updated>2024-02-26T03:55:15.390Z</atom:updated>
            <content:encoded><![CDATA[<h4>How to plan and track progress on AI products</h4><p>In a <a href="https://medium.com/@anuad/minimizing-uncertainty-while-building-a-new-product-b4d64ed40f87">previous post</a> I talked about how to reduce uncertainty in product development and how that should be prioritised over anything else. One of the factors leading to uncertainty, that we did not discuss much, is the techno-economic feasibility of the product which I will discuss in this post now. We will first focus on quality defined products, and then we will briefly talk about other products as well.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*84HM8L687uU_mqaD" /><figcaption>Photo by <a href="https://unsplash.com/@m_malkovich?utm_source=medium&amp;utm_medium=referral">petr sidorov</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p><strong>What is a quality defined product or solution?</strong></p><p>A quality defined product is one whose primary service delivery can not be measured by a binary yes or no but has to be graded on a scale of quality. Take for example an app for a bank which lets you make a transaction. When someone builds such an app, it either works i.e. it lets you make a transaction, or it does not. There is no subjectivity in the outcome. Contrast this with an app which recommends you the next song to listen to. In this case there is an element of quality of the output which is core to the product. The bank transaction app also might have non-binary attributes like transaction-latency or time to settle a transaction, but they may not define the success of the core product. Quality defined products, such as a song recommender, are typically built using data science technologies.</p><p><strong>Challenges associated with quality defined products</strong></p><p>Planning to build a binary-outcome product is much easier because it is a question of putting some building blocks together. It might be non-trivial to estimate the timeline of building such a product but the question is not whether it can be built or not built. The uncertainty on timeline can be reduced by breaking the problem into various sub-components where estimating for a very small component becomes much more predictable.</p><p>Predicting qualitative outcome is not easy. It is a function of both the domain of the problem (the data) and the approach (tools or algorithm) used. The best that one can do to predict the quality is to look at the outcome of a similar approach on a similar problem. There is subjectivity in ‘similar problem’ and ‘similar approach’ itself. The exact nuances of a problem may become clear only when one starts to work on the problem. Therefore solution to such problems often involves research, and trial and error.</p><p>In such cases, it is important for one to understand if with trial and error one is moving in the right direction. Anyone who has done a (classical) ML model training has already faced this. You can play with hyper-parameters, input-features and/or the algo of training a model, but unless you have a test-set, you won’t know if you are necessarily making improvements.</p><h3><strong>Product level evaluation framework</strong></h3><p>Just like we use a test-set to evaluate an ML model for a well defined and isolated problem, we need an evaluation framework at the product level to monitor the progress achieved on a quality defined product. A product level eval measures the output of the product as seen by a user. It does not measure the efficacy of a specific algorithm or model — underlying algos and models are opaque to the user.</p><p>If there is a one-to-one mapping between the model being trained and the product then we may not need a separate evaluation framework at the product level for e.g. if the product is to find spam in email, then the training input-output is most likely going to be the same as input-output at the product level. However most often, the product or feature might be composed of multiple components, and the models being trained may be for a single component only. For e.g. imagine a song recommendation system which recommends songs by first classifying songs into different genres. Here the model that one would train would evaluate if the songs are being classified in the correct genre. The consumer may not care which genre a song falls into or if the song is classified correctly. The classification into genre may not even be exposed to the consumer. For a consumer, it might just be that after listening to song X1 and X2, song Y1 would be okay but not song Y2. The product level eval may then just evaluate whether the song that is recommended after X1 and X2 is acceptable or not.</p><p><strong>Product level eval vs standard ML eval metrics</strong></p><p>Standard ML eval metrics like precision, recall etc may work as a substitute for product level eval if it is a supervised learning problem and the ML problem maps one to one with the product problem. It might still be better to have product level eval as users perceive and subconsciously evaluate the product as a whole. Whether a certain error (in ML prediction) is acceptable or not may also depend on how it is surfaced on the UI.</p><p>Unsupervised learning metrics and sometimes even supervised learning metrics are a poor proxy for what the user wants. While they may help in guiding the model development process initially, product level eval is a better reflection of usability of predictions.</p><p><strong>Product level eval vs User metrics</strong></p><p>Product level eval may not be a user driven metric like engagement or retention. A product eval is evaluation of the product’s functionality itself while a user metric like engagement also captures the hypothesis that a better product would lead to a higher engagement. This may not always be true. For e.g. if a user is searching for some information, then the sooner the user finds the information, the sooner they may leave. You could argue that engagement is not the right user metric here but it is best to decouple product’s quality eval from user hypothesis validation.</p><p><strong>How to run product level evals</strong></p><p>Product level evals can, and should, be broken down into components if the components are exposed to the end user. For e.g. if the song recommendation app comes up with themes on which it can then create a playlists for the user, the problem, as seen by the user, is broken down into two steps — 1. Recommend playlist themes 2. Create playlist on the given themes. In this case, it would be better to have two product level evals as it can help the team identify which of the two components would give a higher ROI on future iterations.</p><p>These evals should ideally be run by an independent team who can get into the shoes of the actual consumer. The evals can be extended further to not just provide a rating on some scale but also provide a rationale for the rating. The reason for the rating often helps one identify the area of opportunity or ideas for improvement.</p><p>These evals should be done periodically and on fresh/latest results. This can prevent the team from over-fitting to a previous eval.</p><p>These evals also give visibility to other stake holders who may not understand the underlying technical solution or its challenges. Having a framework in place, helps them see if progress is being made and at what pace, and if it is worth investing further in the problem space.</p><h3><strong>What should be an MVP for a data science defined product?</strong></h3><p>Loosely speaking, an MVP may be defined as a collection of features which are required in a product for it to be usable and valuable to the consumer. As the question no longer is whether the feature exists or not, one needs to bring in a notion of quality of such features in the definition of MVP. An eval of the feature once again becomes important to gauge if the feature meets a quality threshold required for the product to be viable or not.</p><p>The threshold depends on consumer expectation which in turn depends on what the competition is offering, the segment of the consumer, and the ecosystem overall. Most likely competitive pressure and the need for growth will keep moving the threshold to higher and higher levels.</p><p>As consumer’s expectation can also be a influenced by how the product is positioned, the product architect may have a choice to re-envision the product if it seems like the current threshold can not be met in a suitable time-frame. The entire process should be iterative overall. So the feedback from evals need to be circulated back to product definition because the goal is not to build a product in a specific way, but to solve an actual problem for which there might be another way to go.</p><p><strong>Techno economic uncertainty in non-data science defined product</strong></p><p>In this post I have mostly focused on uncertainty around data-science defined products, and only mentioned timeline as the technical uncertainty on other products. It is possible to have other technical uncertainties on products, though it is not so common or concerning these days. The single biggest source of technical uncertainty is integration with another system or API. This also includes hardware or environment in which the application is supposed to run — in the early years of internet this was specially true for incompatible browsers. Before that it was the case for on-premise software installed on specific hardware. Then it became applicable for multiple mobile OSes. Other uncertainties could come from -</p><p>a. Latency: e.g cryptographic hacking or blockchain work hinges on how fast can you compute something.</p><p>b. Security: Theoretically any new product could have security vulnerabilities. Many of these come from the underlying system or libraries that are being used. As such, the vulnerability can crop up or be discovered after the product is built.</p><p>c. Scalability and Capacity: Can the system scale to N users, or N concurrent users or requests.</p><p>Reliance on cloud-service providers and our collective experience on these matters have reduced these uncertainty considerably, and unlike the data-science defined products, the outcome does not depend on the domain of the problem. These issues are therefor much more predictable.</p><h3>Conclusion</h3><ol><li>Products built using data science technologies have a higher level of uncertainty than other tech products.</li><li>ML evals are often theoretical and may not have a one-to-one correspondence with your product.</li><li>Product level evals define an evaluation with direct relevance to the product and consider the underlying algorithm a black-box.</li><li>Decouple product level eval from product hypothesis validation.</li><li>Run product level evals periodically and independently, ideally with extra comments from evaluators which captures the rationale behind their rating (evaluation).</li><li>Use the overall eval’s result to monitor your progress. Let the comments flow back to the dev team who may use it to plan further improvements.</li><li>Use the eval and its comment, along with your product hypothesis to priortize areas of improvements.</li><li>Define your MVP as a minimum threshold value on the product eval. If the metric does not seem attainable, be ready to re-envision the product.</li></ol><p>The rise of AI and projected ubiquity of AI features in every product means that the bulk of the uncertainty in near future is going to come from the quality of outcome of AI technologies, in what I call — data science defined products. Architects and product owners need to embrace the shift and be better prepared to navigate these uncertainties. Adopting a regular product quality evaluation framework in the development workflow can serve as a compass in uncharted waters.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=db511b0fcde2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Minimizing uncertainty while building a new product]]></title>
            <link>https://medium.com/@anuad/minimizing-uncertainty-while-building-a-new-product-b4d64ed40f87?source=rss-67bcc691e747------2</link>
            <guid isPermaLink="false">https://medium.com/p/b4d64ed40f87</guid>
            <category><![CDATA[new-product-development]]></category>
            <category><![CDATA[product-development]]></category>
            <category><![CDATA[mvp]]></category>
            <category><![CDATA[product]]></category>
            <category><![CDATA[agile]]></category>
            <dc:creator><![CDATA[Anurag Adarsh]]></dc:creator>
            <pubDate>Mon, 22 Jan 2024 12:20:54 GMT</pubDate>
            <atom:updated>2024-01-22T12:20:54.411Z</atom:updated>
            <content:encoded><![CDATA[<h4>Painting a cow or assembling a car: what are you building?</h4><p>It is pretty easy for someone these days to build a new product, or at least it seems so, and yet when it comes to actually building something, people falter. The core issue generally stems from what to prioritize and how to prioritize while building parts of the product. To understand this better, let us take two contrasting approaches to building something in two different scenarios — ‘painting a cow’ versus ‘assembling a car’. Yes, they are not comparable and that is the point!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*UuFCpC0ghRUzcd7X" /><figcaption>Are you sketching a cow or assembling a car?</figcaption></figure><p>As a kid, we all would have tried drawing and painting a cow, or whatever may have been our favorite animal. The way you would start is to sketch a drawing of the cow, maybe the head first, then the body, then its legs, maybe rub off parts of the drawings because it looked disproportionate, redo the parts, finish with the tail, etc. If it looked fine, proceed to the next stage, add shades, fill colors, etc. If it did not, rub off and redo stuff and repeat.</p><p>Contrast this with how we assemble a car in a factory. Each part of the car is built independently, to the spec, and then assembled together. We do not assemble the parts first and then redo one of the parts because it does not look good or fit well (<em>though at the initial stages of design, may be years ago, that may have happened as well</em>).</p><p>You may think that the key difference between the two is that of complexity and scale of investment, but that is not true. The difference is whether you have full clarity on what you want to build, which comes only if you have built exactly the same thing in past.</p><p>When we are building yet another copy or a version of a product that we have built in the past, we are focused on improvements and optimizations. We know exactly what we want. When we are building something new, we only have a vague idea of what we want. The key to success, or to an efficient process which leads to a successful outcome, is to make that idea concrete <strong>incrementally</strong> and <strong>iteratively</strong>, and not trying to perfect one part of the product before seeing the full picture.</p><p>The case of painting a cow is where we actually have a concrete example available to us i.e we know what a real cow looks like and yet we get it wrong all the time. When we are building software products, we rarely have a concrete example of <strong>exactly what we want</strong> in front of us. <em>The factory that builds replica of software products was built more than 50 years ago when the copy command was written. </em>When we are building something new there are many uncertainties.</p><h4><strong>Sources of uncertainty</strong></h4><p>In the initial stages of building a product, the focus should be on reducing the uncertainty. There are three different sources of this uncertainty and depending on the actual product, the extent to which each one of these is applicable could be different -</p><ol><li>Is the product really desired by the consumer — is there a real need for the product?</li><li>Has the architect of the product factored in all the details which could influence the outcome?</li><li>Is the product technically (and economically) feasible?</li></ol><p>A lot has been already written about #1, typically on the topic of product-market fit. I will mostly be talking about #2 next, and will take up #3 in a different post.</p><h4><strong>Can we picture the ideal product?</strong></h4><p>The builder / designer / architect of the product is not the typical consumer of the product. Even when the ideal product, as envisioned by the architect, is built to the spec, when it is taken to the consumer, one often realizes that it does not match consumer expectation. This is true even when the product is trying to solve a real problem. It happens because of tacit assumptions made on behalf of the consumer or because of a cognitive oversight. There is no point in perfecting something when the outcome is not what is desired.</p><p>There are a few cases where we might be able to picture the ideal product —</p><p>a. When we are translating an offline process to an online or a digital one. In such cases, with diligent research/survey, one can formulate what product is to be built with reasonable accuracy.</p><p>b. When we are building next version of the software, primarily to optimise the performance.</p><p>c. A copy cat product or feature.</p><h4><strong>Implication on team size and organization</strong></h4><p>If the goal is to evolve clarity fast, and work in a fail fast manner, then working in smaller teams should be prioritised. If multiple people are sketching different parts of a cow, it is highly likely that it will not look good together (<em>this is a metaphor for building things in general and should not be confused or conflated with UI design only</em>).</p><p>Communication is definitely important, but it is also important to realize that communication itself is not productive. In fact communication eats into available bandwidth. It is a necessary condition but not a sufficient one. The bigger the team, the more the need for communication. Therefore one should always minimize the team size initially and grow team as and when clarity emerges. In my view, being present physically in the same location at the same time, mitigates a lot of risk associated with lack of communication.</p><h4><strong>Isn’t Agile the answer then?</strong></h4><p>Yes and no. It depends on what you mean by agile. Conceptually, the notion of agile exists and is so popular in software development primarily because of the problem that we described earlier — what needs to be built only becomes clear as and when we start building. Yet, if we use Agile methodologies to structure the process from the very beginning itself, it might defeat the purpose.</p><p>In a typical Agile process, people often take a feature or a component at a time and try to build it; while what I am advocating is to not build any feature to a polished state. There are also predefined roles and processes to track delivery of features and timeline. Too rigid a structure might make it difficult or slow to change course. If we are building a novel product, we need a process that <strong>prioritizes reducing uncertainty over delivery of features</strong>.</p><p>Most Agile frameworks exist to support a team to work together. As mentioned earlier, it is better to start with a smaller team initially than to adopt a framework.</p><p>At this stage, I would also like to clarify that the two approaches that we discussed do not exist as binaries. In reality, there is a continuum. One expects clarity to emerge and features/requirements to converge as we start building and iterating. Some parts of the solution may start feeling like they are well defined. They can be modularized and stabilized, while others might still be fluid. So, adopt Agile in spirit and bring in the structure and process as and when it starts to feel like it would make things more efficient.</p><h3><strong>Conclusion</strong></h3><p>Having clarity on exactly what needs to be built cannot be overemphasized. Market pressure means we must ship a product out fast, but building the right product is more important. While we imagine, or may intuitively feel, that we know exactly what the right product looks like, we can only be sure once we see it concretely. Our product might be built of N components, each of which may have a different level of novelty. For us to minimize wasted effort, it is often more efficient to build an unpolished version of the product (<em>not to be confused with an MVP</em>) which gives us enough clarity to have the full picture in front of us, and identify the components which have higher uncertainty. In the cycles of iterations that then follow, one can prioritize the elements which have the highest risk of affecting the success of the product.</p><h4><strong>Related theories</strong></h4><p>There are many related theories which say the same thing i.e. the same guiding philosophy applied to a slightly different context -</p><ol><li>Working in small batches</li><li>Doing things which do not scale</li></ol><h4><strong>Relevance in the AI world</strong></h4><p>When it comes to building AI powered systems, this philosophy becomes even more relevant. You can not, in advance, predict the quality of output of the system unless you have tried it. It depends on the domain of the problem and the nature of the data that you have. I will cover some aspect of this in my next post.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b4d64ed40f87" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What should India do now — 4th day of the coronavirus lockdown?]]></title>
            <link>https://medium.com/@anuad/what-should-india-do-now-4th-day-of-the-coronavirus-lockdown-98fdb6dd0c4e?source=rss-67bcc691e747------2</link>
            <guid isPermaLink="false">https://medium.com/p/98fdb6dd0c4e</guid>
            <category><![CDATA[coronavirus]]></category>
            <category><![CDATA[lockdown]]></category>
            <category><![CDATA[covid19]]></category>
            <category><![CDATA[india]]></category>
            <dc:creator><![CDATA[Anurag Adarsh]]></dc:creator>
            <pubDate>Sat, 28 Mar 2020 10:40:52 GMT</pubDate>
            <atom:updated>2020-03-28T10:40:52.184Z</atom:updated>
            <content:encoded><![CDATA[<h3>What should India do now — 4th day of the coronavirus lockdown?</h3><p>We are in the 4th day of the lockdown and are going to cross 1000 known infections by the end of the day. A lot of people are suffering and in all likelihood it is going to get worse in the coming days. The question is if the 21 day lockdown is sufficient to fight the epidemic.</p><p>Most people who have reasonable access to information, and the discretion to sort out the reliable from the rumour, know that the 21 day lockdown is not good enough. WHO has said that the lockdown only buys you time. As per some sources, WHO had recommended a 2 months lockdown on the country. Wuhan, the original epicentre of the virus, had to go through a 2 months lockdown. We all know that in India we can not afford that — definitely not pan-India, definitely not how we are doing it currently. Government is supposed to build up the healthcare infrastructure during the lockdown to handle the influx of patients later. Despite our best efforts, India won’t have the resources to handle the surge, if the lockdown is lifted entirely after 3 weeks. <strong>Most people in India, do not know or appreciate this.</strong></p><p>It is also worth considering may be the eventual, may be 1 year from now, likely state. As per experts, it is one of the following (depending on our chosen course of action) -</p><ol><li>No action: disease spreads to 70–80% of the population, killing tens of millions of people and then subsides. People develop herd immunity (more on this later).</li><li>Strict lockdown for 2 months, international borders closed: Disease does not infect many more people. All those infected, are identified and resolved. This is what Wuhan did. This is local eradication. International borders will have to be kept shut or under strict vigil for a much much longer time (or the entire world follows the same path and somehow eradicates the virus globally).</li><li>Intermittent lockdowns till the world develops an affordable cure or a vaccine: This is what most of the world is preparing for. A single short term lockdown prevents the spread of the disease. As the curbs are lifted after 2–3 weeks, the virus is likely to resurface somewhere after another couple of weeks. (This is because a lot of people could remain asymptomatic and keep infecting others silently even under a lockdown). A shutdown is ordered again — if not at national level, then at local level. The game of cat and mouse goes on till a vaccine or cure is discovered.</li></ol><p>It is assumed that no one is thinking about #1, and #2 is not possible for a country like India — both for economic and socio-cultural reasons. We are left with some form of #3 — but are we prepared for that? It is not clear if we are even prepared to handle the current 21 day lockdown. The only thing worse than tens of millions of people dying due to the virus would be tens of million dying due to virus along with another millions due to starvation and ailing economy. Government has the dual task of 1. Building the healthcare infrastructure —build hospitals, quarantine units, procure masks, sanitizers, ventilators, train people, nurses, paramedics etc. and 2. Support the marginalised and the poor, and keep the economy afloat. Both these tasks are humongous. We need to come up with ways to make this work better. So what can we do?</p><p>It is important to acknowledge that the I am not an expert on the topic (perhaps nobody is as there isn’t just one topic that we are talking about now). So all comments and critique are welcome.</p><p><strong>Inform, prepare and empower people</strong></p><p>Government should declare and admit to people that this is not a one time 21 day exercise and that life is going to be fine after this. Yes, people are going to panic if plainly told this. They have to be empowered and made responsible. The lockdown has to be treated at district, town and village level. All the borders have to be sealed. It is the people who have to guard their borders to protect their co-habitants. Government can not do this on a long term basis. Probably not even for 21 days. Prime minister’s Junta curfew was an excellent idea. It needs to go a level further. People have to take charge and guard their borders. For thousands of towns and villages this is more important than whether there is any lockdown inside the borders.</p><p>India has more than 600k villages. Even if the real number of infected people today were to be 10k, most of these villages would be untouched by the virus so far. With the borders sealed off, the villages could function inside their boundaries normally. In a couple of week’s time, those neighbouring villages which report healthy could be merged together and their inter-village border be opened. This information management will have to be done by the state. The town borders will have to managed by police. In a month’s time, we may have many of the districts in the country declared covid-healthy. However their borders will have to kept sealed while the local economy inside keeps running. In a way this may become #3 at a district level for some time.</p><p>In order to do this (i.e. seal the village/district borders for a much longer duration), the government will have to allow a small window of time for people to go where they want to be for this extended period of time. This is a social requirement — we already have thousands of migrant labourers stuck at state borders and highways. These people may either be stamped with strict rules — may be even with gps trackers on their wrists, or quarantined in a separate place for sometime. All such people’s name and location may have to be made public. This is not a time to advocate privacy. Once everyone is well informed and made aware, the risk of these people infecting others goes down drastically.</p><p><strong>Recruit the cured</strong>:</p><p>There is some lack of clarity currently whether someone can get the infection again. Though some such cases have been reported, experts suspect that this is likely a case of false-negative on the first test or false-positive later. We assume that there will be immunity to the virus for at least 2–3 months after someone is infected. (Otherwise the idea of herd immunity itself may suffer a setback along with that of finding vaccination soon).</p><p>People who are infected and cured, and hence now immune could form the backbone of the country. These people would be able to do what others are not supposed to. They can form the link between isolated communities. While the govt is stamping people who are quarantined, government should also start sharing green-passes to people who have recovered. This makes it important to test even those people who do not have serious condition — currently at some places the authorities are not testing those who may not require medical intervention as there is not much that they can do with them except for quarantining. The need for making testing cheaper and affordable is much higher. Once the recovered people become part of active economy, they will compensate the cost of testing.</p><p><strong>A counter intuitive idea</strong>:</p><p>The eventual state of the virus is where most of the population is immune to the virus — this can happen either through vaccination, or through catching the infection itself.</p><p>Achieving herd immunity requires 60–70% of the population to get infected. Current data shows that about 80% of those infected develop the disease in a mild form such that they do not need any medical intervention. Could that be our ray of hope? Can we identify the 80% population who will not have any adverse effect of the disease and let them volunteer to get infected and develop immunity?</p><p>The idea may sound crazy, but here are a few points to consider -</p><ol><li>A vaccine is at least a year away, and we have to survive for that long with some form of lockdowns.</li><li>If we get a set of 100 people go through this process and everyone comes out safe and sound, and is then declared to not have any restrictions (because they are immune), a lot more people will gradually opt for this. Getting the first set of 100 people may be the most difficult.</li><li>At the same time we know that a lot many people wanted to travel to Mars on a one way ticket! Soldiers who go to a conventional war, face much more dire consequences and risk. It may not be that difficult to get 100 people out of the 1B population.</li><li>Last, and most importantly, today we have the medical capacity to handle and observe such people. Tomorrow, when there is a surge of critical patients, we won’t be able to do this. It would be much safer to be under medical care and be infected. If most people are likely to get infected anyway (as some experts claim), then it may be in our self-interest to get infected first when we may have all the medical attention, than to do it later.</li></ol><p>It does still require isolation of those voluntarily getting infected from those who are in the potential risk category. However this may be easier to administer and manage compared to the case where one is caught unaware.</p><p>None of these solutions or ideas may be what we need, but we are definitely in the times where we can not think conventionally. A fortnight ago we could not have imagines a pan-India lockdown for 3 weeks. It is not time to get complacent with what we have. No other country has the size or the diversity (and limited resources) as ours. We may not have the luxury of picking up solutions from elsewhere. Despite the best intentions, what we are doing may not be enough. We can not risk another demonetization.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=98fdb6dd0c4e" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>