PANDA: Facilitating Usable AI Development

Abstract: Recent advances in AI and machine learning have created a general perception that AI could be used to solve complex problems, and in some situations over-hyped as a tool that can be so easily used. Unfortunately, the barrier to realization of mass adoption of AI on various business domains is too high because most domain experts have no background in AI. Developing AI applications involves multiple phases, namely data preparation, application modeling, and product deployment. The effort of AI research has been spent mostly on new AI models (in the model training stage) to improve the performance of benchmark tasks such as image recognition. Many other factors such as usability, efficiency and security of AI have not been well addressed, and therefore form a barrier to democratizing AI. Further, for many real world applications such as healthcare and autonomous driving, learning via huge amounts of possibility exploration is not feasible since humans are involved. In many complex applications such as healthcare, subject matter experts (e.g. Clinicians) are the ones who appreciate the importance of features that affect health, and their knowledge together with existing knowledge bases are critical to the end results. In this paper, we take a new perspective on developing AI solutions, and present a solution for making AI usable. We hope that this resolution will enable all subject matter experts (eg. Clinicians) to exploit AI like data scientists.

• The paper introduces PANDA, an end-to-end AI platform that overcomes usability barriers by simplifying data preparation, modeling, and deployment.
• It emphasizes interactive, user-friendly tools for data visualization and annotation, ensuring cost-sensitive and secure handling of sensitive data.
• The work demonstrates practical interfaces like drag-and-drop model construction and distributed hyper-parameter search, balancing performance with interpretability.

PANDA: Facilitating Usable AI Development

The paper "PANDA: Facilitating Usable AI Development" (1804.09997) addresses the pressing challenge of democratizing AI for practical applications across various domains, particularly for non-AI experts such as clinicians in healthcare. The authors propose PANDA, an end-to-end AI development platform aimed at overcoming the barriers that hinder the widespread adoption of AI in domains like healthcare and finance. This essay explores the essential components of PANDA and the significant implications of its framework for facilitating usable AI development.

Barriers to AI Adoption

The paper emphasizes the multifaceted nature of AI application development, which includes stages such as data preparation, application modeling, and product deployment. A gap exists between the rapid advancements in AI and the practical usability requirements in real-world applications, particularly in areas requiring specialized domain knowledge.

Figure 1: AI Development pipeline for Healthcare.

Data Preparation Challenges

The initial stage of AI application development is heavily reliant on data preparation. Despite the ubiquity of data, its manual creation and refinement pose significant challenges. The quality and structure of raw data, especially from sources such as electronic medical records (EMR), play a critical role in determining the efficacy of AI models.

Visualization and Interaction

The paper advocates for user-friendly data visualization tools that facilitate domain experts in effectively engaging with and annotating data. An emphasis on interactive tools with collaborative capabilities aids experts in contributing domain-specific insights to the AI workflow.

Cost-sensitive Acquisition and Data Privacy

Efficient data preparation demands a focus on cost-sensitivity, ensuring that only informative samples requiring expert input are utilized. Furthermore, data privacy is paramount, especially in sectors like healthcare, where integrated data from multiple institutions necessitates secure storage and analysis.

Figure 2: Collaborative analysis by integrating data from multiple hospitals.

Application Modeling and Deployment

PANDA outlines the importance of model selection and cost-sensitive modeling for practical application development. The challenge lies in making these processes accessible for AI researchers, beginners, and domain experts with varying levels of expertise.

Model Selection and Cost-sensitive Modeling

Tailored solutions such as distributed hyper-parameter search and drag-and-drop interfaces for model construction (Figure 3) are proposed to cater to novice AI developers, while experienced researchers benefit from efficient model management systems. The paper also underscores the trade-offs between model complexity and resource efficiency, emphasizing the need for optimization without compromising performance.

Figure 3: Drag-and-drop interface for model construction.

Ensuring Reliability and Security

Deployment necessitates rigorous checks and real-time monitoring to ensure the reliability and interpretability of AI applications, especially in high-stakes domains. The paradigm shifts towards developing AI models that operate robustly within reliably defined data regions, thereby enhancing trust and reducing human oversight.

Conclusion

PANDA represents an effort to surmount inherent challenges within AI development for domain-specific applications. By focusing on usability, efficiency, and security across critical stages such as data preparation, modeling, and deployment, PANDA aims to democratize AI. The implications of this work suggest further research opportunities that include developing interfaces that bridge the gap between domain expertise and AI proficiency, as well as models that balance performance with interpretability and cost-effectiveness.