Despite many advances in automating system and software engineering, the knowledge, skills and experience of human engineers and domain experts are, and will continue to be, critical factors in the success of complex system development projects. In a new research direction we term towards wise computing we aim to make some of these abilities available to all project-team members, at any time, by adding several radically-new capabilities to the development environment. For example, the development environment will be able to, on its own, notice and suggest a required action for, an unexpected, unspeci ed, emergent system property. A report on the ultimate wise-computing vision and an initial demonstration of the desired functions have been published separately. Clearly, advanced tools, such as formal veri cation and synthesis, machine learning and automatic reasoning, which were not previously available, are now maturing and will be essential to wise computing. Yet, a comprehensive feasibility plan was not shown until now, and despite the above arsenal of tools, a challenge is often presented: is it indeed possible to automate some critically relevant abilities of humans, like free association, applying general knowledge to a wide variety of speci c cases, or \thinking outside the box" to nd a solution to a problem. In this report we outline research directions for several software capabilities and design principles that we have begun to work on, and which we believe can remove key remaining obstacles to feasibility of wise computing.
Rich knowledge and sophisticated skills, acquired by experienced engineers over years of training and hard-earned practical lessons, are key ingredients, and even critical success factors, in complex software and system engineering projects. In a new research vision that we term towards wise computing, we aim to endow the software development environment with those engineering abilities that are particularly considered unique to humans and make them available to the project team at all times. Some of the relevant capabilities we are most interested in are: { Free association: i.e., when an event can trigger a distantly related action { Applying engineering and domain expertise: e.g., to new and varied situations { Noticing irregularities: e.g., answering \what's wrong with this picture?" ? With apologies to Lewis Carroll { Creative problem solving: i.e., \thinking-outside-the-box" { Seeing a bigger picture: e.g., considering problems, causes, work methods, etc. { Adapting to new demands: e.g., new work methods due to cyber risks
As a speci c example, in developing an autonomous car, a wise development environment will be able to do the following spontaneously: (1) identify the modules participating in a communication protocol and apply model-checking to that part of the system only, in order to verify its correctness; (2) notice, in testing or in nal use, without it having been speci ed for the system, that the car occasionally slows down, slightly, but still unnecessarily, and then search for causes and solutions; and (3) use domain-speci c information, such as classical mechanics or applicable tra c laws, in code generation, in test generation and in formal veri cation.
A description of the full vision of wise computing, as well as an initial
proofof-concept demonstration of the capabilities that such a wise development
environment will have, has been published separately [
Wise computing will rely on advanced computationally-intensive capabilities, such as formal veri cation using model-checking and SAT/SMT solving, program synthesis, machine learning, pattern recognition, AI planning and reasoning, program static analysis, and automatic test generation. These sophisticated tools, which were not previously available, are now maturing and will play a central role in wise computing. Yet, a comprehensive feasibility plan for wise computing was not shown until now, and despite the availability of the above arsenal of tools, a challenge is often presented: can the human skills described above, and others that are required for accomplishing the wise computing vision, really be automated? Do we know enough about the human mind to even start pursuing such an ambitious goal?
The purpose of this short paper is to describe elements of our present research, which show that in the area of software and system engineering (\SE" hereafter) this is indeed feasible, and that such human activity can be mimicked, independently of the need to understand the biology and psychology behind it.
In a nutshell, we believe that to harness and integrate into a development environment (\DE" hereafter) the latest analytical tools and the not-yet-formalized knowledge and skills of humans, will require the development of new structural and behavioral principles. These will involve issues in system modeling, in knowledge representation, and in the languages used by humans and machines to converse about systems. As part of our broader research plan for wise computing, this paper is dedicated to these enabling building blocks. In the following sections we brie y present several of these as particular research directions, discuss their merits and feasibility, and describe preliminary results supporting their continued pursuit.
Related Work. Due to space constraints we refer the reader to [
The main contributions of the wise computing vision and of the feasibility plan presented here are the breadth of target system applicability, range of human skills addressed, and the search for fundamental game-changing engineering concepts that would streamline this undertaking. 2
Common development practices and methodologies often rely on di erent models and languages for requirements, speci cations, actual system code, test scenarios, simulation results, bug reports, and system documentation. We propose to research a method for modeling all the above artifacts in a consistent way that connects all of them throughout the project life-cycle, in a single model or few closely integrated ones. Combined with extended aggregation and re ection capabilities, discussed in Sec. 3, this will help in application of knowledge in multiple contexts, and in \seeing the bigger picture" of the system, the domain, and the methods and tools used in development. Externally-speci ed connections between artifacts, heuristic searches and systematic explorations of such models will be able to exhibit behavior that may mimic the human ability for free association.
To that end, we propose to research integration of statecharts [
Speci cally, we plan to endow statecharts with the semantics unique to SBP,
including must vs. may transitions, associating statechart states with events that
should be considered for triggering and/or are forbidden in the state, and,
systemwide enforcement of event order subject to all concurrent scenarios. This work has
started and a draft speci cation of the integration is available in [
As an interim step we integrated a statecharts development-and-simulation
tool with the PlayGo SBP tool, allowing not only exchange of events between
the two systems, but also a common, integrated user-interface, and a well-de ned
semantics for joint execution [
Moreover, such a model immediately suggests a very useful approach to
knowledge representation: describe it in scenarios that capture the application of the
knowledge, similar to how the program itself is speci ed. This way, a particular
unit of knowledge constantly tries to apply itself wherever it is relevant, perhaps
in more than one way. An example of this approach is demonstrated in our
preliminary result of a scenario-based biological model for the intra-cellular citric-acid
cycle (a.k.a. Krebs cycle) [
This kind of knowledge representation can also help in detection of emergent properties or missing elements, complementing machine learning methods, with explicit, much simpler scenarios that look for behavior patterns or checklists of elements that must be present. 3
Two of the most attractive features of statecharts are their intuitive and powerful aggregation of states, and the ability to reference states of any object from anywhere. We propose to extend these two capabilities to other system entities, such as scenarios, and to also allow a single entity to be contained in multiple containers, which themselves are rst-class citizens in the model.
Flexible hierarchy and re ection will help in creating contexts for scenarios, and then applying rules to entire contexts|e.g., day-time vs. night-time behaviors of a home-assistant robot, or separating the robot's self-maintenance activities from house duties. They will also simplify the application of external knowledge: e.g., the knowledge that a baby needs a certain number of hours of sleep, can, among others things, be re ected in a scenario that forbids the robot from doing any of the actions previously tagged or identi ed as noisy, when the baby is asleep. External reference to scenario states will also be valuable in re nement and repair of existing behavior.
Aggregation may also enable the DE to exhibit what may be perceived as creative behavior, or \thinking outside the box". As is often advised in innovation workshops, one should try to expand his or her \box", namely, by having a list of generic solutions whose applicability should be considered. Flexible aggregation enables the categorizing of certain system elements, as, e.g., \concurrency related", or \safety-critical", thus enabling automated examination of generic solutions such as a variety of locking and serialization mechanisms for the former, or added sensors-and-alerts or human supervision for the latter.
We have already prepared an initial speci cation for adding hierarchy and
aggregation to SBP in the context of live sequence charts (LSC) [
In addition to any visual modeling and classical programming tools for the above, we propose to provide an extensible textual language with precise, unambiguous semantics for creating and discussing the models. A key feature of the language is that its statements are aligned with how human experts would prefer to converse, with people and with the DE, about the above systems|focusing on one fact at a time, in arbitrary order. Each of the statements can thus be viewed as a standalone entity, and their order or organization will not carry additional semantics.
A highly relevant preliminary result is our group's work on programming scenarios in controlled natural language with precise semantic rules|which may thus be considered as a form of a domain-speci c language (DSL). We have also created interfaces for displaying LSCs as English sentences. Our experience has shown that while a picture (say, of an LSC) is \worth a thousand words", a concise statement such as \when the driver presses the brake pedal, the brake light turns on" is both useful and convenient. Such an approach, based on stand-alone statements, is common also in declarative programming (e.g., Prolog), in text encoding of UML models (e.g., UML/P), or in assertions to SAT/SMT solvers (e.g., CVC4 and Z3). As for a DSL for statement-based description of statecharts, extending the above is straightforward, especially in the presence of existing text-based encoding of statecharts such as SCXML.
Thus, since all system entities, environment knowledge and meta information can be described in statecharts and scenarios, we are con dent that we now have a good basis for a natural interface for the two-way text-based interaction of a wise DE and its users. 5
We introduce a design principle, termed self-integrating behavior (SIB), which we believe can dramatically accelerate the endowing of DEs with new knowledge and analytical abilities. We propose that it be adopted for speci cations, actual system modeling and code, DEs, domain knowledge and other SE artifacts.
SIB-designed components \take care of themselves". They seek out their inputs, and make their outputs available to all other components. But, most importantly, they are also prepared to comply with constraints speci ed by other components. They can thus join or leave complex processes in an incremental, selfstanding way with little or no change to existing components. The latter would be more di cult in common OO designs, where existing modules have to be modi ed with calls of relevant methods or for discovery of con guration changes1.
One implementation we propose for SIB-design is that of scenario-based
programming (SBP) [
As an example of SIB, consider a game-playing program where rules and
playing strategies are each speci ed as a separate scenario, much as they would be
when described verbally, or appear, say, in paragraphs of the printed instructions
without having any program explicitly invoking all these modules in some order
1 Aspect-oriented programming (AOP) provides limited, and somewhat fragile, SIB
capabilities in standard programming languages, and indeed, some SBP implementations
use AspectJ internally. For additional discussion of SBP and AOP see [
Marron et al. (see, e.g., our group's paper in EMSOFT'13 on compositionality in SBP). Another SIB example, focused on knowledge representation, is our preliminary-result modeling of the Krebs cycle, described in Sec. 2.
We propose to further investigate SIB. E.g., see whether any additional semantics are required for accomplishing it, beyond the native SBP principles. One such candidate is adding a context scope to the event-blocking idiom, limiting its otherwise global and unilateral e ect. Another direction is to identify any special properties SIB may have, such as its communication or module-complexity costs, its possibly-improved succinctness under certain conditions, or its allowance of accelerated formal veri cation using, say, assume/guarantee techniques. A most intriguing research direction would be to measure the contribution of SIB and aggregation to tackling state explosion.
As mentioned above, wise-computing will incorporate, and appropriately extend, a broad range of SE ideas and tools. We intend to nd ways to allow existing tools and automated techniques to be embedded in the framework of SIB, in order to allow DEs to harness their power smoothly and e ciently.
Finally, as in parts of the PoC in [
Since the early days of computing, the software engineering community has been aspiring for a development environment that can automate some of the more sophisticated tasks carried out by human experts. We presented four research directions: a single cross-function model for all aspects of development, exible and extensible aggregation and re ection, a text-based interface based on stand-alone statements, and building models, systems, and development environments from units of self-integrating behavior. We believe that these can form important aspects of future system and development-environment design, and help integrate the most advanced tools and human knowledge and skills in SE and related disciplines, thus beginning to suggest the feasibility of the wise-computing dream.