I went to a workshop earlier this week on “the Future of Software Engineering Research” in Santa Fe. My main excuse to attend was to see how much interest I could raise in getting more software engineering researchers to engage in the problem of climate change – I presented my paper “Climate Change: A Software Grand Challenge“. But I came away from the workshop with very mixed feelings. I met some fascinating people, and had very interesting discussions about research challenges, but overall, the tone of the workshop (especially the closing plenary discussion) seemed to be far more about navel-gazing and doing “more of the same”, rather than rising to new challenges.
The break-out group I participated in focussed on the role of software in addressing societal grand challenges. We came up with a brief list of such challenges: Climate Change; Energy; Safety & Security; Transportation; Health and Healthcare; Livable Mega-Cities. In all cases, we’re dealing with complex systems-of-systems, with all the properties laid out in the SEI report on Ultra-Large Scale Systems - decentralized systems with no clear ownership; systems that undergo continuous evolution while they are being used (you can’t take the system down for maintenance and upgrades); systems built from heterogeneous elements that are constructed at different times by different communities for different purposes; systems where traditional distinctions between developers and users disappear, as the human activity and technical functionality intertwine. And systems where the “requirements” are fundamentally unknowable – these systems simultaneously serve multiple purposes for multiple communities.
I’ve argued in the past that really all software is like this, but that we pretend otherwise by drawing boundaries around small pieces of functionality so that we can ignore the uncertainties in the broader social system in which it will be used. Traditional approaches to software engineering work when we can get away with this game – on those occasions when it’s possible to get local agreement about a specific set of software functions that will help solve a local problem. The fact that software engineers tend to insist on writing a specification is a symptom that they are playing this game. But such agreements/specifications are always local and temporary, which means that software built in this way is frequently disappointing or frustrating to use.
So, for societal grand challenge problems, what is the role of software engineering research, and what kinds of software engineering might be effective? In our break-out group, we talked a lot about examples of emergent successful systems such as Facebook and Wikipedia (and even the web itself), which were built not by any recognizable software development process, but by small groups of people incrementally adding to an evolving infrastructure, each nudging it a little further down an interesting road. And by frequently getting it wrong, and seeking continual improvement when things do go wrong. Software innovation is then an emergent feature in these endeavours, but it is the people and the way they collaborate that matters, rather than any particular approach to software development.
Obviously, software alone cannot solve these societal grand challenges, but software does have a vital role to play: good software infrastructure can catalyze the engagement of multiple communities, who together can tackle the challenges. In our break-out group, we talked specifically about healthcare and climate change – in both cases there are lots of individuals and communities with ideas and enthusiasm, but who are hampered by socio-technical barriers: lack of data exchange standards, lack of appropriate organizational structures, lack of institutional support, lack of a suitable framework for exploratory software development, tools that ignore key domain concepts. It seems increasingly clear that typical governmental approaches to information systems will not solve these problems. You can’t just put out a call for tender and commission construction of an ultra-large scale system; you have to evolve it from multiple existing systems. Witness repeated failures of efforts around shared health records, carbon accounting systems, etc. But governments do need to create the technical infrastructure and nurture the coming together of inter-disciplinary communities to address these challenges, and strategic funding of trans-disciplinary research projects is a key element.
But what was the response at the workshop to these issues? The breakout groups presented their ideas back to the workshop plenary on the final afternoon, and the resulting discussion was seriously underwhelming. Several people (I could characterize them as the “old guard” in the software engineering research community) stood up to speak out against making the field more inter-disciplinary. They don’t want to see the “core” of the field diluted in any way. There were some (unconvincing) arguments that software engineering research has had a stronger impact than most people acknowledge. And a long discussion that the future of software engineering research lies in stronger ties between academic and industrial software engineering. Never mind that increasingly, software is developed outside the “software industry”: e.g. open source projects, scientific software, end-user programmers, community engagement, and of course college students building web tools that go on to take the internet world by storm. All this is irrelevant to the old guard – they want to keep on believing that the only software engineering that matters is that which can be built to a specification by a large software company.
I came away from the workshop with the feeling that this community is in the process of dooming itself to irrelevancy. But then, as was pointed out to me over lunch today, the people who have done the best under the existing system are unlikely to want to change it. Innovation in software research won’t come from the distinguished senior people in the field…