This paper outlines a strategic framework for developing general-purpose scientific AI by addressing the challenges of interpreting multimodal scientific data. To facilitate this, the authors introduced the ALD/E-ImageMiner benchmark and the ICDAR 2026 Competition on Information Extraction from Atomic Layer Deposition/Etching Scientific Figures. This dataset comprises 1,951 expert-annotated figures sourced from 205 distinct publications. The benchmark evaluates multimodal capabilities through four primary tasks: classification, data table extraction, summarization, and visual question answering (VQA). Rather than focusing solely on basic OCR or pattern recognition, the researchers aim to probe higher-order cognitive abilities, including quantitative reading, domain-grounded reasoning, and evidential justification. The methodology emphasizes Bloom-informed question design to move beyond surface-level descriptions toward deeper scientific comprehension. Looking forward, the authors propose "scientific conceptual understanding from images" as a long-term benchmark objective. Future research directions include expanding to diverse figure types and domains, enabling cross-document synthesis, evaluating hypotheses, and managing uncertainty and provenance. Ultimately, the work seeks to transition from simple data extraction to creating machine-actionable, verifiable scientific visual knowledge capable of supporting open-ended multimodal research.