Artificial intelligence has emerged as a translational tool at the interface of biomarker discovery, tissue diagnostics and clinical decision-making. This lecture follows the trajectory from detection to decision, again contrasting lung cancer and sarcoma. In common, data-rich malignancies such as lung cancer, artificial intelligence assists in interpreting abundant information, extracting additional signal from imaging, pathology and blood-based assays already generated in routine care. In rare malignancies such as sarcoma, it addresses the opposing problem of data scarcity, extending specialist-level diagnostic capability to centres that lack it; in this setting the algorithm may itself function as a biomarker. The lecture reviews how artificial intelligence is shortening the biomarker development cycle, and why the barriers to clinical translation differ between common and rare disease. In relation to treatment decisions, it critically appraises artificial intelligence as an adjunct to the multidisciplinary tumour board, including evidence that fluent, confident model outputs may conceal unreliable reasoning. Attention is directed to the challenges of greatest current relevance: prospective validation, generalisability across populations, bias, regulatory approval, and the distinction between an accurate model and a clinically usable one. The near-term role of artificial intelligence is framed as augmenting expert judgement and distributing scarce expertise equitably, rather than replacing it.