RAISE 2013 - Day Two

Twitter discussion from the second day of the NSF workshop on Realizing AI Synergies in Software Engineering - http://promisedata.org/raise/2013/index.html


  1. The first day of the 2013 RAISE workshop took a dive "over the horizon," discussing future synergies between artificial intelligence and software engineering. Seven short vision statements were presented on potentially fruitful research directions, along with a perspective session and a keynote from John Clark.

    Like the first day, side-discussion on Twitter was encouraged under the hashtag #icse13raise. Some of the tweets are collected here:
  2. Following the vision statements (and lunch), the workshop participants broke into two groups to brainstorm and debate the following topics: 
    1: SWOT (strengths, weaknesses, opportunities, and threats) of the synergy of AI and SE
    2: How people should be considered in both topics
    3: A call for research proposals on RAISE
  3. Group 1 returned the following thoughts:

    1. SWOT analysis of AI and SE


    Deep knowledge of SE from experts
    AI quite good at analysing natural language; SE often deals with human
    written documents - potential synergy
    AI can offer guidance on demand, very quickly
    Communities are talking to each other (AI and SE)
    More data coming online


    Weak AI knowledge from software engineers
    Data not being shared
    Obsession with manual software engineering methods (which is slow)


    Many systems to evaluate, a large volume of data to be collected
    Cognitive burdens to be taken off human software engineers
    Will be great if people can share their data
    Deeper analysis still possible (in the similar fashion to text mining)
    Social networks like stack-overflow may be a source of (SE) domain knowledge
    More data coming online


    Not enough accuracy to be useful: issue of trust and/or worry of
    creating even more manual work
    Not understanding each other community, not speaking the same language
    Better learners are not doing better than dumb learners: unexplained
    variations in results, lack of data analysis patterns

    2. People

    We should be learning more from people
    Venues: NLP, AI/SE, text mining, traceability analysis - very diverse
    research interests. Do we need more specialised workshops?
    Present SE problems to AI researchers as a productive, impactful, and
    meaningful intellectual challenge
    Educational opportunities for AI people to transfer skills to software engineers

    Search-based methods
    Machine Learning
    Expert systems, including agents

    They do Turing test for generic conversation: why not SE?
    Present things in a cooler light (Pex4Fun)

    3. Ideas

    Success of simple/stupid text mining techniques: why?
    What is the best way for an academic to transfer technology? Students
    are not enough.
    What is the longest term goal that we can set?
    Obsession about correctness in software engineering

    4. CfP

    We need more software to improve the general quality of life.
    More data: learn from the history.
    More precision in the information we see.
    Provide software in a way people can use, understand, change/personalise
    We need to understand and learn from history.
    Software engineering for under-developed countries: what unique
    challenges and requirements?
    Computation as self-organising "thing" (not unlike biology)

    Group 2 added:


    AI produces prototypes and tools (much more than SE)

    many public open source AI tools (e.g. data miners)

    Availability of SE data

    Recent development of big data and cloud technologies

    Strongs advances in automated SE (e.g. source code feature extraction)


    SE produces fewer prototypes and tools

    Not much involvement by traditional AI researchers

    Lack of precise definitions of goals

    AI are technologies for solving a range of problems

    SE is methods for build specific products

    We often speak in generalities while only offering very specific results.

    Absence of a culture of challenge problems and challenge systems

    A lack of system models and methods to build the


    Extraordinary computational power and memory can now be applied to problems

    On-the-fly creation

    Creative dialogs between AI and humans, rather than merely between humans

    Theory of testing of complex systems

    Better understanding of the types of determinism needed for classes of systems

    Human centric synergy for making systems and tools more usable by people

    Involvement of cognitive science

    If we restrict our domains, many specific tools and techniques often succeed

    New HCU interaction technologies (e.g. NLP) may create opportunties for ASE

    Strong awareness of the centrality of human’s role

    Green computing


    AI and SE use different ways to build systems

    Distance between communities, no common language

    SE research not dealing with real systems

    AI may help in traditional lifecycles but not useful for agile (not enough data)

    Complexity of upcoming systems

    Overconfidence in our knowledge and ability about assumptions

    The lack of a big picture

    Obstruction from individual communities