How do you share your AI model with non-technical people?

Your audience doesn't even need to know that it's AI! One of my most vivid memories as a Digital Transformation Advisor was introducing an AI solution to a board of directors, most of whom were not technical. Earlier in my career, I made the error of diving deep into algorithms and losing the attention of such an audience quickly. I learned that successful communication starts with empathizing with your audience. Before diving into the "how," it's vital to understand "what" matters to them. Grasping this can steer the conversation from technical jargon to tangible benefits, in a language accessible to them.

KISS. Keep it simple. Use simple langauge, as if explaining it to a 5-year-old child, including a simple diagram if possible. Do not use tech lingo. Then everyone will be on the same wavelength.

I always like to explain reinforcement learning as the akin to the method parents employ when teaching a child not to touch fire or rewarding them for good behavior. Just as parents shape their child's actions through incentives and consequences, this approach guides AI models to make better decisions by rewarding positive outcomes, ultimately enabling them to learn and improve over time.

- Tailor your message to suit your audience's needs and interests, and avoid using jargon or technical details that may confuse or bore them. - Use analogies and metaphors: Explain AI concepts using analogies and metaphors that your audience can relate to. - Provide real-world examples and demos: Show examples and demos that demonstrate how AI works and what it can do, using real-world scenarios that your audience can relate to. Focus on the value and impact: Explain how your AI model can solve problems, bring value, and make an impact on your audience's goals and concerns. Tell a story: Connect the data, the model, and the results to the real-world context and implications.

Storytelling: Incorporating storytelling elements into the presentation can make the content more relatable and engaging. Narrating the development and application of the AI model through a story can help connect with the audience on an emotional level, making the information more memorable. For example: A farmer uses a computer vision-enabled drone to monitor crop health. The drone identifies a small area of early pest infestation, allowing the farmer to address the issue promptly, thereby saving a large portion of the crop and ensuring a successful harvest. This story illustrates the practical, beneficial integration of computer vision in modern agriculture.

Storytelling is an important aspect of AI adoption. While explaining an AI intervention, pls keep these points in mind - 1. AI is a scientific intervention in a business process. One needs to understand it thoroughly before explaining it to others. 2. Understand the as-is problem before advocating for AI 3. Use simple terms or analogies to explain it to non-tech audience. AI fundamentals can be easily correlated with routine tasks. 4. Don’t make them sound fool or outdated. Complex explanation won’t make you smarter. Don’t forget to set the context. 5. Tools that you can use - data visuals, charts, and inferences

Again just use the ChatGPT Microphone Extension and explain away all this context Then ask IT to come up with the solution. The more and more context you give it, the better the answers will be.

It’s vital to tailor the explanation of an AI model to your audience’s specific needs and concerns. Leveraging frameworks like “Jobs to Be Done” and “Value Proposition Design” can be invaluable. These frameworks force you to focus on the outcomes and value your AI model provides to users, shifting the conversation from capabilities to addressing real problems or challenges. Understanding the “so what” factor is crucial in making AI models relevant and valuable to your audience.

Again, I repeat what I said above. Keep it simple. Use simple langauge, as if explaining it to a 5-year-old child, including a simple diagram if possible. Do not use tech lingo. Then everyone will be on the same wavelength.

Finding a balance between Technicality & Communication is key. I often use Conversational AI tool like ChatGPT or Claude to simplify technical terms and break them down into easy-to-understand words or terms. Example prompt: "Explain Deep learning as you would to a 6 year old" This prompt will help you learn to communication complex ideas. Infact over time, you would begin to learn to communicate complex ideas just by thinking about it.

With respect to showcasing Computer Vision models, I feel like it's important to expose the non-technical person to all the different cases that can occur with respect to images or videos and the relevant results. A non-technical person may not have a clear idea of what is possible with respect to the AI model and what is not possible. I've experienced this during meetings with clients who want to understand why an object detection model has given a false positive prediction, and they are frustrated that the model performance isn't comparable to human eyesight. This is avoidable if, before deployment, you make it very clear what an AI model can actually do and what it can understand.

To present your AI model to stakeholders, remember they want assurance that it's technically sound and beneficial. Clarify basics, connect loss functions to business goals, and emphasize overall performance vs. occasional outliers. Use interactive demos to build confidence. Aim for balance between technical depth and practical application. This approach will showcase the model's strengths and build trust with your audience.

Live Demonstrations: Organizing live demonstrations where the audience can witness the model in action in real-time, addressing real-world problems, and providing solutions. This approach can create a dynamic and engaging experience, allowing the audience to see the immediate impact of the model. In my experience, I have worked on a Smart Parking project. If any customers are interested, the best approach is to take them on a tour of a smart parking garage where our solutions have been implemented, and to present our applications and solutions in person.

Yes, show me and I will start to understand, but it is better to involve me, then I will learn much faster and more completely.

Interactive demos are king when it comes to sharing models with non-technical people. Use a tool like Streamlit and Gradio to create the demo quickly. Then setup an organization on Hugging Face Spaces, add the people to that organization, and deploy the application privately for them to see.

Anticipating the audience's questions and needs is a must! You need to be very clear on the AI model limitations. Stakeholders will be aware and worried about bias and ethical implications so you need to explain how how you have integrated ethical principles in the design of the model and how you have addressed potential issues with data bias.

Alweays have a full summary explanation of why this model, how it works, when to use it, where to use it, and who is imvolved -- and how they benefit.

Yes understanding audience is very important but it does take hell lot of time. Also sometimes we need to hit arrow in the dark. Knowing persona is simple but knowing an audience in case of AI demo - whose purpose is to get acceptance and adaption as well. Then different route the key! In design thinking , It’s fairly simple to tackle at the define stage wherein audience set , itself list down the questions and use cases . I have modeled our design thinking practices against 2 goals : 1. Acceptance 2. Adaption Beat technology, A.I or any change , DT has worked for us in more then 80+ acceptance model.

Sharing a document with the model specifications and your own performance assessment might help presenting it completely and responsibly.

Agreed, you need to be able to predict the type of questions the audience might have on the limitation of the system and address them within the presentation.

I find this is of little value. When people read articles or reports, they invariably ignore all external references. Much better to give actual use-case examples in the report itself.

Make yourself available for follow-up and direct contact. Invite stakeholders to dig deeper into aspects of interest. Offer to provide references to technical papers, tutorials, or experts. Suggest meetings to address outstanding questions. Enabling further exploration shows commitment to education and transparency. It empowers audiences to become ongoing partners in optimizing AI impact.

In my perspective, this is the most vital aspect of sharing AI models with a non-technical audience. While AI models may seem intricate at first glance, breaking down the high-level concepts into more accessible resources like beginner-friendly articles, blogs, and videos can make a world of difference. By offering these materials, we empower non-technical folks to grasp the fundamental workings of AI models. It's like providing a roadmap that guides them through the complexity and helps them gain insights into the incredible capabilities of AI. It also opens doors for meaningful discussions and interactions, ultimately strengthening the connection between AI enthusiasts and experts.

Regular communication and dialog are critical — not only for building the relationship, but also for maintaining it. Staying in touch will make it easier to work on any improvements or projects next time around. Also, nothing in business is static; neither are your model or your stakeholders tasks. So, keeping your stakeholder relationships “warm” let’s you pick up where you’ve left off.

When demonstrating your model, aim for a balance between technical depth and practical application. Make it relatable, connect it to business outcomes, and always be prepared to address the exceptions. This approach will showcase the model's strengths and build trust with your audience.

You can use ChatGPT or Claude2 to get suggestions of how to build a narrative of the model ( the tool) and use cases ( how to use the tool and why this tool is better) in an way that your audience could easily understand. For example, using such prompts like “ explain like I am five .. or my audience is [personas], explain …”

A model's performance depends on its training data. Biased data can cause problems, so diversity is essential. While metrics like MAE matter, the real-world impact determines success. Take the chess-playing AI that surprised its creators. Beneficial outcomes make for the most compelling stories. Always consider AI's ethical implications and uphold fairness, transparency, and privacy.

Also storytelling! In one of my recent presentations, I employed a storytelling technique to elucidate the concept of secure and private AI. I drew an analogy with the practice of using sniffing dogs in airports. These specially trained dogs can detect illegal items like guns or drugs in luggage without ever examining the contents. I find this analogy particularly effective in conveying the essence of differential privacy. It helps illustrate how AI can glean valuable insights from data without compromising individual privacy, much like how these airport dogs uncover contraband without revealing the contents of the bags.

People don't even need to know it's AI. Focus on what they care about. What will the model do for them? A perfect example is Apple's recent announcement of the new Apple Watch. It was full of new AI features, but they never mentioned AI once. They said "Siri is more accurate with speech and can work without a wifi connection." Technically, they beefed up the AI processing capability of the watch and new the models are running on the watch instead of remotely. People didn't need to know that, they needed to know what that meant for them.