Comment optimiser un algorithme pour le bien commun ?

I believe that goals should be aligned to the way social good is perceived in a specific domain, for example in stock management domain the social good would be to make sure algorithm always has demand a little lower than stock amounts. Social good should be measured in all domains objectively taking fairness, rationale, implications, contribution to macro-economy/household economy/healthcare status/individual's welfare and ethicality.

The goals should not only include the accuracy and fairness but also include a reason to be responsible. A parameter as this will improve the reasoning towards social good.

Greenwashing for "social good" is a risk. Think like a cybersecurity expert and look for exploits in the algorithm and test against them. This requires a dedicated community and ongoing vigilance. Adoption is a second key. Like a tree in the forest if no one is there, who can say it fell? Dare I say it needs to be marketed?

For optimizing algorithms for social good, we should follow human centric design approach by invovling relevant users in every stage of design process. This should help us in maximizing positive impact and minimize the negate consequences from social perspective. The algorithm should be designed for delivering positive social impact in accordance with the priorities as outlined in the United Nations "17 Sustainable Development Goals (SDGs)". Plus algorithm design and implementation should focus on fairness, bias, ethics and responsibility aspects for achieving social good.

Developers can optimize algorithms to address societal challenges such as healthcare accessibility, education, and environmental sustainability. By focusing on developing algorithms that directly contribute to solving these challenges, software development can be optimized for social good. This is for addressing societal challenges. The second point is for ethical considerations: When optimizing algorithms for social good, it's crucial to consider ethical implications. This involves ensuring that algorithms are designed and implemented in a way that prioritizes fairness, transparency, and accountability, while also addressing concerns related to biased results and digital divides.

Ensure it aligns with ethical standards and respects privacy, fairness, transparency, and accountability. Identify and mitigate biases in the data used. Implement techniques like re-sampling, re-weighting, or using de-biasing data/view to address potential biases.

Always learn your business domain! If you know how the business flows, you can understand the meaning of you own data. That will give you huge insights.

For any algorithm or software development requirements are also plays a major role to define the data. And it is very important to re-evaluate the requirements from Stakeholders, Software Development, Integration and Testing Point of view else it will contradict at each stage because of many limitations at each level for examples in case of end users, platforms, testing tools etc.

A thorough data analysis involves examining every aspect of the data in detail. This includes verifying the source, assessing quality, identifying trends, addressing missing and inconsistent data, and considering ethical implications. Using statistical and visual tools can provide deeper insights, ensuring that the data is reliable and valuable for decision-making.

Gather and examine data relevant to your goal. Continuing with the same example: This could involve looking at patterns in food distribution and consumption, and identifying where the most wastage occurs.

Understand any restrictions you have on how your data can be handled and processed. Some data - e.g. health data - can have legal and ethical restrictions on how it can be used.

Once you have the data the easiest way to identify the methodology/algorithm is to assess what the solution needs to address: 1. Is this A or B ? (classification algorithm) 2. Is this normal or not? (anomaly detection) 3. How much or how many? (regression algorithm) 4. How this is organized? (clustering) 5. What should it do next? (reinforcement learning)

"If you have only hammer in your hand, everything seems like a nail" Learn alternatives, how the others do the same job, even how they failed...

Depending on your goal, you'll need to choose the most appropriate algorithm or technique. This could be a machine learning model that predicts where food will be wasted, so you can intervene and reroute the food where it's needed.

Let software and algorithms help everyone: poor, ritch, whoever. If AI is involved, don't try to get information which *could* give biased answers, try to circumvent here and solve real problems.

An algorithm for testing results for social readiness is much needed and evolving in pockets, a consolidated effort would make big difference

Be Agile!! In my experience its very important to evaluate the results from very beginning. It gives a fair response and review of Stakeholders and provide the expectation. Which also improves the process and provides much better outcomes than expection at the end as we could able to analyse the results at each stage of development.

Get "golden standard" data (buying or generating, collecting, whatever). Test the algorithms on expected or exceptional results. Don't be agile - SW development gets too much interrupted being AGILE. Agile forces not thinking to the end, bringing up half ready solutions.

When ethical and social implications are a part of the evaluation, it is important to do the evaluation not only by yourself, but involve other experts in the evaluation, such that you prevent that the scope is just your opinion, but a set of opinions where the results are viewed from different perspectives.

One of the main key aspects of a well written application is to be able to log and inspect every action that js taken in the system. That alone will also be immensity helpful for auditing purposes and cost savings to enhance performance.

In my experience, I have come to understand that no solution is too perfect as problems are dynamic and ever-evolving, hence it's also great to ensure that your algorithms are not just good, but scalable too

I would like to offer valuable advice to fellow developers: "Do not rush to refactor your solution too early." While it is essential to strive for efficiency and optimization, prematurely attempting to optimize your code can often lead to unnecessary complexity and wasted effort.

Magic values are limitations to long term robust systems. Reusable blocks of code with optimal tests should be written to enhance scalable systems.

In my opinion, to insure continuous improvement in optimizing algorithms for social good: -Regular assessmnets -Feedback loops -Stay informed about ever evolving societal needs. -Collaborate with diverse stakeholders -adapt strategies based on real world impact

As I always mention: There is always place for perfection. But as business needs solutions in limited periods of time and assuming we know "the theory of the 90%" it's always not perfect when deliveder, because chasing the ideal solution is time consuming. So the iteration is the key to perfect results (is that code or anything else). As famous saying (unfortunately don't know the real source) tells us do it in the following steps: 1. Make it work (even ugly and not optimal) 2. Do it as it should be, in the proper way (as we someones call "according to Feng shui") 3. Improve it (make it work better, prettier and faster) 4. Repeat step 3 untill it's perfect (btw it never is) And remember the meaning is not the goal, it's the way.

Here is something totally different to consider: I lead some path breaking algorithms developments in the MedTech area. Hence , I find all the inputs in this entire discussion thread very valuable. But then, I had a heart attack on December 2022. In short, it happened because there was no ECG machine in the general practitioner's clinic in an advanced country. When we are struggling with such basic things, what is the social good about new algorithm development that we are talking about? Along with focusing on new developments, it's the responsibility of all of us to ensure that 'social good' of earlier innovations have duly reached the majority of the society.

One thing I've found helpful,r egarding the subject under discussion, I must say that a new approach is based on a repeating pattern and based on user operations in the network, and specifically, on Internet social networks, social banking, online stores, etc., in which users' operations They choose their interest, it is executed. First, based on users' interests or activities in the network, we discover a number of small communities consisting of similar users, and then expand each community using social connections. The results of F-measure evaluation on two real-world datasets show that the proposed method improves the quality of community discovery. repetitive pattern exploration algorithms such as CAN and CAT...

I believe algorithm optimization is a topic of computer science than ethics. For example, fb optimized its algorithms to spy on people's activities and analysing behaviours. It might not bear any social good but against when applied to observing criminal activities, it can carry boons - the whole topic of good and bad is subjective and based on context, whereas optimisations are made based on parameters that are devoid of any subjectivity but numbers.

If you can measure the outcome then utilize all your viable solutions in different tests (like A/B testing) in parallel then have an automated/manual system to adjust your target percentages until the winner solution takes the majority of the traffic.

Ethical bias is something that is largely unprioritized, and that is much more crucial than we think. Large dataset, tools benchmark are usually tested on dataset or on Caucasian European, thus providing bad result or putting aside dark skin tone or specific face shape or other behavior. It costs more money and time to thoroughly test on a decent subset of population, but it’s up to our responsibility to not add ethical bias by not being « lazy » in order to live in a more fair society. And also simply to have a good product working for everybody and not putting people aside unknowingly.