Cognification is the application of knowledge to boost the performance and impact of a process. According to Kevin Kelly, cognification is one of the 12 technological forces that will shape our future (I strongly recommend reading his book: The Inevitable or at the very least, watch this video).
We (myself, Robert Clarisó, Marco Brambilla and Sébastien Gerad) believe cognification will also be a revolution in the way software is built. In particular, we’d like to discuss the opportunities and challenges of cognifying Model-Driven Software Engineering (MDSE) tasks, and, conversely, how MDE could also speed up the development of cognified software. These ideas were presented at the GrandMDE (Grand Challenges in Modeling) workshop and now we have expanded them into a short paper that you can read below (to appear in the STAF 2017 workshop proceedings).
Check the slides or read the full description of the work below (also available for download as a free pdf).
- 1. Introduction: Cognification and its role in Software Engineering
- 2. Opportunities in the Cognification of MDSE
- 3. Model-driven Engineering applied to AI and Knowledge-aware Software
- Conclusions and challenges
1. Introduction: Cognification and its role in Software Engineering
It is hard to imagine anything that would “change everything” as much as cheap, powerful, ubiquitous intelligence and exploitation of knowledge. Even a very tiny amount of useful intelligence embedded into an existing process boosts its effectiveness to a whole other level . This process is known as cognification.
Cognification can be defined as the application of knowledge to boost the performance and impact of a process. It does not restrict itself to what we typically refer to as artificial intelligence, with Deep Learning as its latest and hottest technique. Cognification includes as well the combination of past and current human intelligence (i.e. all current humans online and the actions they do). Under this definition, collective intelligence and crowdsourcing  are valid cognification tools. Cognification does not imply either the existence of a super AI but the availability of highly specialized intelligences that can be plugged in depending on the needs of the problem to be solved.
Several initiatives aim to cognify specific tasks in Software Engineering, for in- stance, using machine learning (ML) for requirements prioritization , for estimat- ing the development effort of a software system  or the productivity of individual practitioners  or to predict defect-prone software components . This trend is also happening at the tool level, e.g., the recent Kite IDE claims to “augment your coding environment with all the web’s programming knowledge”. MDSE should join this ongoing trend.
Moreover, we know the limited adoption of MDSE is due to a variety of social and technical factors  but we can summarize them all in one: its benefits do not outweigh its costs. We believe cognification could drastically improve the benefits and reduce the costs of adopting MDSE.
In this paper we discuss the opportunities that cognification can bring to the MDSE field, in terms of possible tools, process steps, and usage scenarios and their empower- ment through cognification. We also discuss how MDSE itself can be applied to AI and knowledge–aware technology development, and we conclude with some challenges and risks that this roadmap may encompass.
2. Opportunities in the Cognification of MDSE
All MDSE tasks and tools are good candidates to be cognified, since they typically involve plenty of knowledge–intensive activities. Here, we comment on a few examples where the performance or quality of experience can be especially boosted thanks to cognification.
Cognification can enable modeling bots playing the role of virtual modeling assistant. Based on previous models on the same domain available online or on a comparison between your current model and general ontologies, the bot could suggest missing properties on the model, recommend best practices or warn that some model aspects differ from the way they are typically modeled.
Preliminary work in this direction considers recommendations during the construction of a model to reduce the effort made by designers.  proposes a recommender that suggests which model element should be the target of a newly created reference. Similarly,  studies the notion of an “auto-complete” that is able to infer the intended pattern that is being designed.
In some contexts, models are created by domain experts by using a Domain-Specific Language (DSL). This creation process includes manual activities like auto-completing a partial model or introducing small fixes to ensure it satisfies the integrity constraints of the domain. Proactive modeling  aims to automate these manual activities. Bots could even interact with domain experts in a more friendly environment (e.g. social networks, SOCIO would be a good example) to facilitate the knowledge acquisition process.
A model inferencer is able to deduce a common schema behind a set of unstructured data (logs, tweets, forum messages,…). This model would be a useful “lense” to interpret the data. A good example of a model inferences is JSON Discoverer . Given a (set of) JSON documents, it analyzes the JSON data and generates for you a class diagram showing graphically the implicit JSON schema of your documents plus an object diagram representing their data
In a related context, process mining  aims to achieve a similar goal with respect to procedural information describing a process, action or activity.
Smart Code Generators
A cognified code generator would be able to learn the style and best practices of a particular organization and mimic them in the code it outputs. By learning from good code samples, the generator would be able to imitate the company’s best practices and style and maximize its chances to be accepted as a new “developer” for the company.
This line of thought is close to existing works on example-based generation of modeling artefacts. For instance, , where a search-based approach is used to derive model-to-model transformations from sample input and output models.
Real-time Model Reviewers
Machine learning has already been used in the context of verification and validation to tune verification algorithms by selecting the best choice for heuristics [15, 12]. However, here the aim would be constructing a real-time model reviewer able to give continuous feedback on the quality of the model from the point of view of consistency and correctness.
Such tool would resort to different verification and validation (V&V) techniques, using information about previous analysis to predict the complexity of the analysis  and select the most suitable method to verify a particular model. This problem is known in the A.I. field as the algorithm selection problem .
In addition to selecting the most adequate tool, previous experiences can be useful by highlighting the most frequent faults or the constructs that are most likely to produce problems. This type of information can be used to guide the search process, e.g. which kind of properties should be checked for a particular model. Some preliminary works addressed the use of semantic reasoning for validation and property checking for models: for instance , propose a simple tool able to address issues that cannot be identified by traditional type checking tools.
Another approach to model quality would be to evolve or adapt models rather than detect and correct faults. For instance, the notion of liquid models , which considers that design-time models should not be immutable artifacts, but that they should evolve to take into account potential deviations occurring at run-time. Cognification could be used as a means to study and explore candidate models during this evolution process.
Advanced Self-Morphing and Collaborative Modeling Tools
Automatic learning approaches can also be used within self-morphing modeling tool, able to adapt its interface, functions, behaviour (and even the expressiveness of the language offered) to the expertise of the tool user and to the context of the modeling problem addressed. This challenge is related to the problem of plastic user interfaces, i.e. interfaces able to keep its usability under changing uses and circumstances (responsive websites can be regarded as a limited example of a plastic UI). While there has been some work on using MDE to generate plastic UIs , building a plastic MDE IDE remains an open challenge.
The tool should not just adapt to different user profiles but also effectively support the collaboration of those users. Indeed, cognification also comprises exploitation of human and collective intelligence. As an example, we could leverage crowdsourcing techniques  to devise the best domain-specific modeling language to be used by the tool in order to optimize the communication process between modeling experts and domain experts. This has been so far applied to the problem of agreeing on the concrete syntax notation of domain-specific languages, as described in .
Semantic Reasoning Platforms, Explainability and Storification
At the purpose of making models as self-explanatory as possible, semantics-based techniques can be applied to the concepts in the model to enable automatic explanation of models. This would require a semantic reasoning platform able to map modeled concepts to existing ontologies and provide definitions (similar to Kindle WordWise3), references, relations to relevant concepts, services, and contextual information; and conversely also generate new knowledge through inference.
This would also enable automatic generation of thesaurus, data dictionaries, and even explanatory text out of models. Integrating MDE with resources such as ontologies, semantic processors, NLP tools, rule–based inference engines, and alike will be crucial in this setting.
Scalable Model and Data Fusion Engine
Big data architectures and data streaming platform enable the construction of data fusion engines that are able to perform semantic integration and impact analysis of design-time models with runtime data such as software usage logs, user profiles and so on. A typical use case is the integration of Web application logs, which are widespread in Web servers, with user interaction models of Web applications. An example is the work [2, 3] which integrates with a scalable approach the real-time big data stream coming from large–scale Web sites with IFML models. This close interaction between design and runtime models is also the focus of the MegaMart2 EU project.
3. Model-driven Engineering applied to AI and Knowledge-aware Software
Besides the advantages that cognification can bring to MDSE, the reverse is also true: MDSE techniques can be used to improve knowledge-aware technologies. As in any other domain, the use of models (and its abstraction power) can help simplify, interoperate and unify a fragmented technological space, as it is the case right now in AI, with plenty of competing and partially overlapping libraries and components for all kinds of knowledge processing tasks.
So far, this line of work remains largely unexplored. A few exceptions are , aim- ing to integrate machine learning results in domain modeling, , aligning MDE and ontologies, and works oriented to model-driven development of semantic Web based applications  and semantic web services .
Conclusions and challenges
As we described in this paper, a lot of possible scenarios and tools in MDSE can benefit from the application of cognification techniques in a broad sense. Even if only a few of them become a reality in the short-term, they have the potential to drastically change the way MDSE is used and perceived in the software community.
Nevertheless, some risks need to be addressed. The main challenge is that most of the above scenarios imply the availability of lots of training data in order to get high- quality learning results. In MDSE, this means a curated catalog of good and bad models, meta-models, transformations, code, and so on . While some model repositories are available (e.g. REMODD or MDEForge the number and diversity of the modeling artefats they contain is still limited. Improving this situation is a strong requirement towards the cognification of MDE.
- van der Aalst, , et al: Process Mining Manifesto, pp. 169–194. Springer (2012)
- Bernaschina, C., Brambilla, M., Koka, , Mauri, A., Umuhoza, E.: Integrating modeling languages and web logs for enhanced user behavior analytics. In: WWW’17. pp. 171–175 (2017)
- Bernaschina, , Brambilla, M., Mauri, A., Umuhoza, E.: A big data analysis framework for model-based web user behavior analytics. In: Web Engineering – 17th International Confer- ence, ICWE 2017, Rome. pp. 98–114 (2017)
- Brambilla, M., Cabot, J., Izquierdo, J.L.C., Mauri, A.: Better call the crowd: Using crowd- sourcing to shape the notation of domain-specific languages. In: Proc. of 2017 ACM SIG- PLAN International Conference on Software Language Engineering (SLE?17). ACM, New York, NY, USA (2017)
- Brambilla, , Ceri, S., Celino, I., Cerizza, D., Valle, E.D., Facca, F.M., Turati, A., Tziviskou, C.: Experiences in the design of semantic services using web engineering methods and tools. Journal of Data Semantics (JoDS) 11, 1–31 (2008)
- Brambilla, M., Ceri, S., Della Valle, , Volonterio, R., Acero Salazar, F.: Extracting emerg- ing knowledge from social media. In: WWW’17. pp. 795–804 (2017)
- Brambilla, M., Fraternali, : Interaction flow modeling language: Model-driven UI engi- neering of web and mobile apps with IFML. Morgan Kaufmann, USA (2014)
- Brambilla, M., Tziviskou, C.: Modeling ontology-driven personalization of web contents. In: 8th International Conference on Web Engineering, ICWE 2008, New York, USA. pp. 247–260 (2008)
- Brambilla, M., Tziviskou, C.: An online platform for semantic validation of UML models. In: 9th l Conf. on Web Engineering (ICWE) 2009 San Sebastia´n, Spain. pp. 477–480. Springer (2009)
- Gasevic, , Djuric, D., Devedzic, V.: Model Driven Engineering and Ontology Development (2. ed.). Springer (2009), https://doi.org/10.1007/978-3-642-00282-3
- Gogolla, M., Cabot, J.: Continuing a benchmark for UML and OCL design and analysis tools. In: Workshop on Software Technologies: Applications and Foundations. pp. 289–302 (2016)
- Haim, S., Walsh, : Restart strategy selection using machine learning techniques. In: Pro- ceedings of the 12th International Conference on Theory and Applications of Satisfiability Testing. pp. 312–325. SAT ’09, Springer-Verlag, Berlin, Heidelberg (2009)
- Hartmann, , Moawad, A., Fouquet, F., Le Traon, Y.: The next evolution of mde: a seamless integration of machine learning into domain modeling. Software & Systems Modeling (May 2017)
- Hutchinson, E., Whittle, J., Rouncefield, M.: Model-driven engineering practices in indus- try: Social, organizational and managerial factors that lead to success or failure. Sci. Comput. Program. 89, 144–161 (2014), http://dx.doi.org/10.1016/j.scico.2013.03.017
- Hutter, , Babic, D., Hoos, H.H., Hu, A.J.: Boosting verification by automatic tuning of decision procedures. In: Proceedings of the Formal Methods in Computer Aided Design. pp. 27–34. FMCAD ’07, IEEE Computer Society, Washington, DC, USA (2007)
- Hutter, , Xu, L., Hoos, H.H., Leyton-Brown, K.: Algorithm runtime prediction: Methods & evaluation. Artif. Intell. 206, 79–111 (Jan 2014)
- Izquierdo, L.C., Cabot, J.: Jsondiscoverer: Visualizing the schema lurk- ing behind JSON documents. Knowl.-Based Syst. 103, 52–55 (2016), https://doi.org/10.1016/j.knosys.2016.03.020
- Kelly, : The inevitable: understanding the 12 technological forces that will shape our fu- ture. Viking (2016)
- Kessentini, M., Sahraoui, H.A., Boukadoum, M., Benomar, O.: Search-based model transformation by example. Software and System Modeling 11(2), 209–226 (2012), https://doi.org/10.1007/s10270-010-0175-7
- Kotthoff, , Gent, I.P., Miguel, I.: An evaluation of machine learning in algorithm selection for search problems. AI Commun. 25(3), 257–270 (Aug 2012)
- Kuschke, T., Ma¨der, P., Rempel, P.: Recommending auto-completions for software modeling In: Proceedings of the 16th International Conference on Model-Driven Engineer- ing Languages and Systems – Volume 8107. pp. 170–186. Springer-Verlag New York, Inc., New York, NY, USA (2013)
- Lopez-Martin, C., Chavoya, A., Meda-Campaa, M.E.: A machine learning technique for predicting the productivity of practitioners from individually developed software projects. In: SPND’2014. pp. 1–6 (2014)
- Mazak, A., Wimmer, : Towards liquid models: An evolutionary modeling approach. In: Business Informatics (CBI), 2016 IEEE 18th Conference on. vol. 1, pp. 104–112. IEEE (2016)
- Pati, , Feiock, D.C., Hill, J.H.: Proactive modeling: Auto-generating models from their semantics and constraints. In: Proceedings of the 2012 Workshop on Domain-Specific Mod- eling. pp. 7–12. DSM ’12, ACM, New York, NY, USA (2012)
- Perini, A., Susi, A., Avesani, : A machine learning approach to software requirements pri- oritization. IEEE Transactions on Software Engineering 39(4), 445–461 (April 2013)
- Shepperd, , Bowes, D., Hall, T.: Researcher bias: The use of machine learning in software defect prediction. IEEE Transactions on Software Engineering 40(6), 603–616 (June 2014)
- Sottet, J., Ganneau, , Calvary, G., Coutaz, J., Demeure, A., Favre, J., Demumieux, R.: Model-driven adaptation for plastic user interfaces. In: Human-Computer Interaction – IN- TERACT 2007, 11th IFIP TC 13 International Conference, 2007, Proceedings, Part I. pp. 397–410 (2007)
- Steimann, , Ulke, B.: Generic model assist. In: Proceedings of the 16th International Con- ference on Model-Driven Engineering Languages and Systems – Volume 8107. pp. 18–34. Springer-Verlag New York, Inc., New York, NY, USA (2013)
- Stol, K.J., Fitzgerald, B.: Two’s company, three’s a crowd: A case study of crowdsourcing software development. In: ICSE 2014. pp. 187–198. ACM (2014)
- Wen, , Li, S., Lin, Z., Hu, Y., Huang, C.: Systematic literature review of machine learning based software development effort estimation models. Information and Software Technology 54(1), 41 – 59 (2012)