What are some successful data visualization applications in AI?

Dashboards are nice but you need to go further when executing AI at scale. At OCBC, our internally built framework links automatically to workflow and notifications so that key stakeholders are notified when there is an issue with model data quality, model drift or fairness metrics. We can review, investigate, comment & approve from the workflow. A dashboard is too standalone & has limited value without integrating to the end to end process.

From my experience, one key advantage of AI dashboards is their ability to democratise the understanding of complex models. When deploying models in real-world scenarios, it's often crucial to involve multiple stakeholders — from business leaders to operational staff. An effective AI dashboard serves as a bridge, turning abstract numbers into intuitive visuals. I've seen organisations swiftly pivot strategies or catch errors early because of insights gained from a well-crafted dashboard. Moreover, by integrating real-time feedback loops, AI dashboards can evolve iteratively, capturing changing business dynamics and incorporating user feedback.

One of the best frameworks i have used for this sort of tracking is streamlit which provides all the tools for graphing and creating a sharable application without much effort and knowledge of ui and frontends.

Being able to monitor AI models drift can be life or death of an AI startup. Being able to detect when models are giving wrong results particularly predictive results to customers is essential and dashboards are a great way for this

AI Dashboards can be a great way to explain the output of AI models graphically, along with easy ways to depict data anomalies, model explanations and model drifts. Hence, dashboards have both business and non-functional value for AI models.

We were building a customer churn prediction model. The initial models were simple, easy to understand, but delivered only 30% improvement over the base state. As we improved the model accuracy, we crossed over into black box algorithms - the realm of AI. These were 30% more effective than simpler white box models, but with zero explainability. When we rolled these 'superior' models, the users resisted and we had to kill the project. The importance of explainability cannot be overstated. It might sound illogical to engineering teams, but the real world operates a bit differently - prioritize explanation over accuracy!

AI explainability means understanding how and why an AI system makes certain decisions. It's essential for designers to build trust and transparency in AI, especially in important areas like healthcare and finance. We use data visualization to show why AI suggests certain things, making it easier for people to trust the technology.

Understanding why AI models make decisions/predictions is vital for trust and transparency. Data visualization techniques offer a path to achieve this. Here are some useful/known apps for ExplainableAI: InterpretML (suitable for various ML models, including neural networks), SHAP (Provides explanations for a wide range of ML models, including neural networks), LIME (Explains the predictions of black-box models), TensorFlow Interpret (for interpreting TensorFlow models, including CNNs), GNNExplainer (explainability for GNNs, IML-GNN (for graph data), DeepExplain (for deep learning models, suitable for CNNs). GANDissect (for GANs), ExBERT (for interpreting BERT-based models).

Understanding why AI models make decisions/predictions is vital for trust and transparency. Data visualization techniques offer a path to achieve this. Here are some useful/known apps for ExplainableAI: InterpretML (suitable for various ML models, including neural networks), SHAP (Provides explanations for a wide range of ML models, including neural networks), LIME (Explains the predictions of black-box models), TensorFlow Interpret (for interpreting TensorFlow models, including CNNs), GNNExplainer (explainability for GNNs), IML-GNN (for graph data), DeepExplain (for deep learning models, suitable for CNNs). GANDissect (for GANs), ExBERT (for interpreting BERT-based models).

Explainable AI (XAI) is essential for building trust, transparency, and fairness in AI applications. 1. Transparency and Trust: XAI enables users to understand the reasoning behind AI decisions which builds trust by demystifying complex AI systems and helping users feel more comfortable with the technology. 2. Accountability: In high-stakes scenarios, knowing how and why an AI system arrived at a particular decision is vital for accountability. 3. Bias Mitigation: XAI tools can uncover bias within AI models, allowing for bias mitigation strategies to be applied. In practice, different techniques can be employed to achieve AI explainability, such as feature importance analysis and decision path visualization.

"No one ever made a decision because of a number. They need a story," said Daniel Kahneman. Even the most powerful AI insights could be useless when not communicated in an understandable and actionable way. Data storytelling can help bring your story to life with insightful visuals, a relatable business context, and an actionable narrative.

Storytelling is a critical way to help demystify complex topics like cybersecurity. Imagine a plethora of log data from different sources, and that in turn is presented as data points. It’s not easy for people and management to see the real threat and vulnerability landscape. AI storytelling could evolve this by linking data points to create a real life scenario of how an attacker could infiltrate the organization, making it more impactful and faster at helping cybersecurity professionals act on issues rather than hunt.

GenAI has the power to revolutionise the field of data analytics and from my opinion it has already started. Besides of process, analysis and quickly interpret a vast amount of data, storytelling to business users was and is very important. GenAI assists effectively telling the right data story and interpreting the meaning of data.

A storytelling technique that is regaining steam recently is "PechaKucha". Give it a search in your favourite engine. The use of visuals, emotion-driving images, and colourful language can be highly effective in translating the technical aspects of AI to a more business-savvy audience. Effective storytelling will take your stakeholders through the journey of: 1. Listen 2. Understand 3. Believe 4. Buy-in Good luck!

The art of storytelling emerges as an essential asset when it comes to the visualisation of complex data. A prominent example for its need is big data obtained from clinical trials and real-world evidence studies, leading to a better understanding of the efficacy and safety of new treatments. With the rise of transformative technologies (such as AI) which have the potential to reshape the healthcare landscape, it becomes apparent that we need to discover new ways of how to communicate with patients in a comprehensible and relatable manner. Of particular importance is health literacy, which ensures that individuals outside of medical/scientific domains possess knowledge about disease development, progression, and treatment.

In my experience, in the cybersecurity space simulations of table-top exercises are common in the risk management process. But often the case, humans have to decide on the specific exercise to perform the table-top. AI simulations could take this process to scale, and potentially come up with simulations which we may not even have considered. These new insights could be automated and provide risk scores and impact scores for security personnel to work on. Moreover, chaining these simulations together could possible unearth complete new risks to the organization. It would be like putting Honeypotting on steroids.

When we consider AI simulations we can see that there are great applicational benefits, especially in professional environments. For instance AI stimulation can offer a realistic training environment for professionals while also allowing opportunity for rapid prototyping and risk free experimentation. This in turn provides effective data driven insights while enhancing learning, decision making and innovation across a wide array of industries and even academic fields.

AI-generated simulations have greatly evolved many fields, especially robotics. For instance, being able to generate kinematic simulations for edge cases, as well as simulate those scenarios are both very complimentary and game-changing functionalities when it comes to validation and verification. This is particularly the case for mission critical workloads, which are generally regarded as realms in which the use of any machine learning may be regarded as “too black of a box”, and thereby disregarded.

In the life insurance domain, AI simulation is revolutionizing risk assessment. Imagine a scenario where an insurer employs sophisticated AI models to simulate various life scenarios for policyholders, incorporating factors like health data, lifestyle choices, and genetic information. By running thousands of virtual life scenarios, the AI can predict the likelihood of health issues and life expectancy with remarkable precision. This approach allows for more accurate and personalized insurance premiums, benefiting both the insurer and the insured.

AI Simulation is often overlooked but serves as a potent data visualization tool, especially for decision-making. It provides multi-dimensional, real-time visualizations that allow stakeholders to grasp the impact of AI-driven strategies in complex systems like supply chains. For example, a change in one network node can be visually traced to see its ripple effect across the entire chain. Furthermore, AI simulations can quantify and visualize risks, offering a more nuanced understanding of possible outcomes. By incorporating AI Simulation into data visualization strategies, we move beyond static charts, enabling dynamic, context-rich decision-making.

AI Creativity in data visualization represents a fascinating frontier in AI. It involves using artificial intelligence to generate innovative and captivating data visualizations that go beyond conventional charts and graphs. AI creativity can help uncover hidden insights in your data by presenting it in fresh, unexpected ways. For instance, AI algorithms can generate art-like visualizations that appeal to human emotions and senses, translating data into a form that is more engaging and memorable. This is particularly valuable for conveying complex data to a broader audience, making data-driven decisions more accessible and persuasive.

AI-powered data visualisation tools can help your business gain insights from your data far more efficiently and effectively. AI can also help identify patterns and trends that might not be immediately apparent. If you’re interested, head over to selfiecity and check out the average smile score for a few capital cities and then see if angry people tilt their heads more in a selfie….

With the aid of AI, we can generate the most effective visualizations for specific types of data, taking into account the dataset's characteristics and past user interactions. This allows us to display crucial data more prominently.

AI can be used to generate data visualizations that use natural language to explain the data in a clear and concise way. For example, AI could be used to create a data visualization that explains the relationship between different variables in a dataset, or a data visualization that summarizes the key findings of a study.

Before the "ChatGPT" moment, AI mostly focused on predicting patterns based on past data. One of the key applications were Recommender Systems (those that enable you to find a nice TV show in Netflix, or to get precise product recommendations from Amazon). With Generative AI, we are unleashing a new realm of opportunities. Imagine this: what if the AI piece becomes and "Inspiration System", where recommendations are just inputs for the human agent to leverage them for their work? Bing Create is a great example. You can create different versions of any image you have in mind, and take it as base for your artistic activities. Other formats include text, audio, video, etc. Try it. Explore it. Welcome to the AI creativity era.

here some examples: Exploratory Data Analysis (EDA): Data visualization plays a crucial role in EDA, where AI systems create interactive graphs, charts, and plots to help data scientists and analysts gain insights and discover patterns within data. Tools like Tableau and Power BI are commonly used for this purpose. Natural Language Processing (NLP): AI-driven NLP applications use data visualization to present textual data in a more understandable format. Word clouds, sentiment analysis charts, and topic modeling visualizations help users grasp the meaning and sentiment of large text datasets.

AI for ethical hacking research is a proven way where AI can help co-pilot and technically test theoretical hypotheses. Assume you are cybersecurity researcher and have some concepts where you think it could be possible to infiltrate or attack a system but not quite sure how to start technically, or you know how to start but have a roadblock to jump between software components. With the right context and prompts, AI could help with the experimentation process and co-pilot the proof of concept. The benefits of AI to cybersecurity are potentially enormous.

I would add a section on Data Privacy and Security in AI Data Visualization. As AI models process and visualize data, it's essential to safeguard sensitive information and comply with data protection regulations. This involves data anonymization to protect individual identities, encryption for secure data transmission and storage, and strict access controls. Compliance with data protection regulations is mandatory, and the integration of AI allows for the automation of security measures, such as anomaly detection. By incorporating security measures, organizations can harness the full potential of AI in data visualization while upholding the trust and confidence of their stakeholders and users.

Beside the application of AI in data visualization, the "streamlit" framework is recent, open source, and easy to use python framework for visualizing data in web pages. Streamlit is awesome for data applications.

What is obvious for a data scientist, for example, is not that trivial for business users. Therefore, it is critical success factor to engage end user as earlier as possible to make sure the data visualization meet the expectation. Otherwise, all the hard work will value nothing.