Selective lithium recovery from dilute brines remains challenging because Li⁺ must be separated from competing ions, particularly Mg²⁺. In this work, a commercial NF90 nanofiltration membrane was used as a non-dispersive interface between a simulated brine and the hydrophobic ionic liquid [P_6,6,6,14][PHOS]. Individual-salt experiments yielded apparent overall mass transfer coefficients of 638 ± 273, 100.5 ± 19.8, and 10.1 ± 0.7 L m⁻² h⁻¹ for Na⁺, Li⁺, and Mg²⁺, respectively. Replacing water with 10% (v/v) [P_6,6,6,14][PHOS] in hexane as the receiving phase increased Li extraction from 18.6 ± 3.0% to 40.9 ± 2.0%. Under direct Li⁺/Mg²⁺ competition, selectivities of 2.19 ± 1.10 and 3.83 ± 0.56 were obtained with the diluted and neat ionic liquid receiving phases, respectively. With neat [P_6,6,6,14][PHOS], Li extraction reached 40.3%, and 6.0% of the extracted Li was subsequently recovered in pure water under non-optimized conditions. CHNS analysis showed no detectable transfer of organic material derived from the ionic liquid to the aqueous feed. Overall, NF90 was the main source of Li⁺/Mg²⁺ transport discrimination, whereas [P_6,6,6,14][PHOS] mainly increased Li uptake.

Artificial intelligence (AI) is increasingly being investigated to support decision-making and process improvement in copper mineral processing and extractive metallurgy; however, the available evidence remains fragmented across operational units, methods, sustainability dimensions, and geographical contexts. This systematic mapping review, complemented by qualitative cross-study synthesis, analysed publications from 2014 to 2025 retrieved from IEEE Xplore, Scopus, and Google Scholar. Of the 410 records identified before screening, 71 studies were classified according to four predefined research questions addressing copper-processing operations, AI domains, sustainability contributions, and geographical distribution. The findings show an uneven distribution of research across the processing flowsheet. Comminution and froth flotation received the greatest attention, whereas lixiviation, solvent extraction, electrowinning, and electrorefining were less represented. Machine learning was the dominant AI domain, particularly supervised approaches based on historical operational data; however, no algorithm was universally superior, as reported performance depended on the dataset, target variable, operational context, and validation procedure. Most studies focused on prediction, monitoring, fault diagnosis, and operational optimisation, while reinforcement learning, hybrid mechanistic–data-driven models, adaptive control, and sustained industrial deployment remained comparatively limited. Sustainability assessments emphasised operational and economic outcomes more frequently than social, ethical, circular-economy, and broader environmental implications. Geographical results reflected the locations assigned to the reviewed studies and industrial cases rather than regional AI maturity. The review identifies data availability, heterogeneous validation practices, model transferability, and limited evidence of sustained industrial deployment as recurring challenges. It proposes a staged research agenda to develop more reliable, transferable, and human-supervised AI applications across copper-processing operations.