*1.2. Engineered Artificial Intelligence*

Data bursts within engineering disciplines. For years, data was used in other areas where models were less developed or remained quite inaccurate. Data collected massively was successfully classified, cured, distilled, ... using artificial intelligence (AI) techniques. Thus, correlations between data ca be removed, proving that a certain simplicity remains hidden behind a rather apparent complexity. Data-driven modeling developed exponentially and advanced artificial intelligence techniques were developed, covering six major domains: (i) Multidimensional data visualization [9]; (ii) Data classification and clustering [10,11]; (iii) Learning models from input/output pairs of data, with adequate techniques enabling real-time learning and able to operate in the low-data limit (e.g., sPGD [12], *Code2Vect* [13], iDMD [14–16], NN [17], ThemodynML [5,18], ...); (iv) Knowledge extraction in order to identifying combined parameters and model richness/complexity, discovering hidden parameters, discarding useless parameters or even to extract governing equations; (v) Explaining for certifying; and (vi) Hybridizing physics and data for defining advanced and powerful Dynamic Data-Driven Application Systems, DDDAS [19].

However, these data-driven models, when used in engineering and industry, were quickly confronted with three major and recurrent difficulties: (i) the need for a huge amount of data to make predictions accurate and reliable, knowing that data is synonymous with cost (acquisition and processing costs); (ii) the difficulty of explaining and interpreting predictions obtained by artificial intelligence; and (iii) related to the the latter, the difficulty of certifying engineering products.
