How do you engage diverse stakeholders in machine learning?

Some more technical details to engage stakeholders: Domain Expert Involvement : Involving the domain experts to define the problem, and helping validate the results. Explainable AI: using interpretable explanation to understand the reason behind predictions. Data Sharing and Collaborations: facilitating data and code sharing for collaborative contribution, and comprehensive API documentation. Customisable Parameters : that stakeholders can adjust according to specific requirements to fine-tune to their business needs. Maintain a repository of technical documentation, including architectural diagrams, model explanations, to keep stakeholders informed about the technical aspects and for reproducibility.

As per my experience, while leading certain products in Machine Learning at Dyota AI, Stakeholders play a vital role as multiple models can't be implemented and it's very important to understand the end user and stakeholder point of view to leverage a stable solution. We can't implement 10 models , instead we focus to leverage the same with the specific use case and most important factor affecting any business to built a scalable solution at the same time.

In our area at Trullion, we work within the financial reporting and accounting industry, which is very conservative and highly regulated. There is an intense fear surrounding what many feel is the 'black box' nature of AI models and the lack of control, and transparency that this can entail. To this end, we work very closely with senior accounting and audit professionals, to both communicate and educate them about what we are doing and to get educated as to the standard of transparency and explainability we need to provide to ensure that we are improving the industry, and not undermining it.

Stakeholder prioritization is THE crucial step. It will have the highest impact on the outcome. Moreover, it won't show in almost any retro analysis and will fall into the bucket of 'you don't know what you don't know.'. For example, treating customer is the most important stakeholder, against time-to-market focused.

In these often resource-constrained environments, ML technologies are proving invaluable. For instance, ML-driven agricultural models can provide farmers with real-time insights into crop health, weather patterns, and optimal planting times, thereby increasing yields and food security. Telemedicine platforms powered by ML connect villagers with remote healthcare services, bringing vital medical expertise to underserved areas. Moreover, ML enables predictive maintenance of critical infrastructure like water supply and electricity, reducing downtimes and ensuring the reliability of essential services. These applications, among others, are transforming rural villages, enhancing their quality of life, and fostering sustainable development.

As per my recent experience with one of the big players, some of the time they just want to implement AI in some use cases, where there is no need at all(as the chance of failure is 1 in 5 Lakh). At Dyota Labs, we are very open to letting stakeholders know whether it's really needed to build such a solution or not before proceeding with the whole Product Development. If you say yes to everything, then in the end it will be a big failure. So, Commit to what can be done in that specific time which will help you to grow at the same time.

One thing that is important, especially when you are at the early stages of model development is setting reasonable expectations. In the current zeitgeist it seems as though many stakeholders can have inflated expectations of the outcomes of machine learning projects. However, because real world data is messy there are few guaranties that can be made at this stage. Hence, getting expectations out in the open and discussing them is tantamount. Remember, it's better to underpromise and over-deliver than overpromise and underdeliver.

I always make sure that what we are proposing is connected to a business objective and that we can measure it. One way I know this is because our non-technical customers understand most conversations around our models perfectly.

Vision: We aim to bridge the digital divide and improve the quality of life in remote areas. Through this project, we envision providing access to essential services, enhancing economic opportunities, and fostering sustainable development. Objectives: 1. Improved Agriculture 2. Telemedicine Access 3. Infrastructure Reliability 4. Education Accessibility Benefits - Enhanced Quality of Life - Increased Productivity - Resource Efficiency infrastructure services. - Education Access - Economic Growth Risks - Data Privacy - Algorithmic Bias - Technical Challenges - Acceptance and Adoption Ethical Principles - Fairness - Accountability - Transparency - Privacy Addressing Harms and Biases

Effective communication of goals and value proposition is crucial in managing ML projects successfully. It ensures alignment, informed decision making, and ability to adapt to changing conditions, ultimately increasing the likelihood of achieving project objectives and delivering value to the company.

Personally I find the involvement of stakeholders really important to the overall ML cycle. Involvement with trust and honesty has been paramount in delivering timely and quality ML models. A business or technical input from an internal or even an external stakeholder can sometimes provide insights to significantly improve or make sense of the results that might not seem logical at all. No matter how well we know businesses, there might be some instances that are critically unique to the specific business problem we are out to solve.

While mentioned methods are good, they are too generic. Most of the time, your stakeholders do not know what they want, and based on how you interview / survey them will dictate the outcome. A book in entrepreneurship but I think is applicable wherever user interviews feedback is gathered - The Mom Test - will definitely give you some insights.

While delivering one of the products for Traffic Enforcement , stakeholders engaged on a regular cycle even while testing the device on the Highway during the hot summer months back in Delhi. It's become the responsibility of the stakeholder to observe the ML Flow closely as they are the ones who have committed to some other clients for what we are building for them or they want to use at their end which again needs things to be perfect and should help any business at best.

Ensuring transparency in the ML lifecycle for stakeholders is essential. Without clarity, stakeholders might lack trust due to uncertainties about the ongoing processes. Often, there's a need to simplify complex concepts for them, which is no trivial task. As an ML engineer, it's crucial for me to collaborate with other business partners to find solutions.

Any machine learning model is only as good as the data that goes into it. Too often, ML is viewed as a solution for an organizational issue, or a data issue. Make sure ML is an appropriate tool. Many problems can be solved with less advanced statistical models, with higher interpretability. Make sure you have access to quality data, and enough data, to train a model robust and reproducible enough for your use case(s).

It’s really important to test your algorithms against a trained bias classification tree. What does that mean? A model trained by men or just by women may include bias over certain topics. To tackle that, you need to have a blind model that will be like a second brain testing your models. By testing that, your models will have a less tendency towards some features. Also, you can do that by running a supervector machine. Thus, reducing the amount of features. Or you could also drop the features that are less important than the mean of score of your choice. What you choose will depend on your model and what you are trying to find.

I've found that what data individual stakeholders want can change based on their position in an organisation. For example given a summarization model, the type of summary needed will greatly vary based on the person reading the summary

Acknowledge and respect the diversity of your stakeholders, recognizing their varying backgrounds, cultures, languages, experiences, and opinions. Understand that these differences can shape how they perceive and engage with your ML project. Avoid making assumptions or generalizations that could lead to misunderstandings or conflicts. Strive to balance and integrate diverse perspectives, addressing any trade-offs or disagreements thoughtfully. Foster an inclusive environment that values all stakeholder inputs and promotes collaboration towards a shared goal.

In my experience, stakeholders must understand ML recommendations and ensure traceability for ethical accountability. I address this by thinking: - How did ML arrive at its recommendation? (understand the workflow) - Can decision-makers understand inputs, analysis, outputs, and processes? (simple terminology) - Engage broad stakeholders for equity. (over-communicate) This feedback shows an understanding of results, and it's crucial to keep parties involved in the process throughout each step. While it can be challenging, it's important to communicate the practicality of ML and not present it as a solution to every problem. This approach fosters trust with stakeholders and promotes a more transparent and collaborative environment.

Establish trust and rapport with stakeholders by demonstrating your credibility, competence, and integrity as an ML practitioner. Show commitment to meeting their needs and expectations by being responsive and dependable. Maintain open, regular communication to keep stakeholders updated on the progress and outcomes of the ML project. Acknowledge and appreciate their contributions and feedback, fostering a sense of collaboration. Celebrate milestones and successes together, reinforcing positive relationships and shared achievements throughout the project lifecycle.

Clear communication holds a major key to stakeholders engagement. Keep stakeholders informed about project progress, challenges, and milestones. Regular updates build trust and demonstrate transparency. Be responsive to their concerns and questions. Analyze feedbacks to identify areas of improvement. Adjust communication strategies and plans based on what you learn from stakeholders.

Evaluate and improve stakeholder engagement by measuring the effectiveness and impact of your activities, focusing on participation, satisfaction, understanding, and influence. Collect and analyze feedback from stakeholders to identify strengths and weaknesses in your engagement approach. Use this insight to make necessary changes and improvements, refining your methods for future ML projects. Continuously learn from your experiences to enhance stakeholder relationships and ensure more successful and collaborative project outcomes. Adapt your approach to address evolving needs and challenges.

Don't overstate what AI or ML is used for! I have been an Anti Money Laundering Consultant for 5 years and the most abused word and field I have seen is AI and ML. The reason is that people think of AI and ML as a silver bullet able to solve all woes. But it is not, and the fact that subject matter experts understand this and do not communicate this, basically adds to the distrust in the system. Having a background in ML and Deep learning myself, I can clearly see most applications in Anti money laundering and even Retail sector can be accomplished using basic stats and/or process automation without ML. This is a bit harsh, but I have heard this and seen it with my own eye enough, to not be able to deny it. I wish good luck to those

Being transparent about the true expectations and abilities of ML is crucial! This will guarantee that trust is built and maintained between you and the stakeholders. Making sure that there are regular updates and check ins, which are delivered in a language that is understood by everyone will also aid in carrying projects to the desired outcome.

In a very traditional business, folks are often wary of change. It might be because they're used to a certain way of doing things or just nervous about new technologies that seem too good to be true. If you think this new technology can help people, make sure you show off what it can do. The big bosses might not get it right away, so listen to their feedback, understand where they're coming from, and keep doing your thing. After all, there's a saying, "People won't use something new until they really like it." It might sound weird, but I've found it to be spot on.

From my perspective, we should put our focus on the limits that our ML model have, so the stakeholders may have a bigger picture of the output, especially with decision takers (managers) which will grab our ML insights to take decisions that will change the business course of action.