AI Based Governance For Software Development
Until very recently, in traditional software development, ‘testing’ continued to be seen as a separate process – detachable from development and deployment – a loosely coupled working model sometimes forced and justified as an ‘independent third opinionator’ in spite of ‘lacking epistemic testability’. Resulting from many such unfortunate scenarios, in spite of being widely accepted as a serious validator in most areas, many traditional testing models have been blamed for lacking in adaptability and governance. These are tell-tale signs that ‘testing’ will have to quickly mutate into a mightier avatar – through a suitable genetic upgrade – which will also prevent it from being blamed for wrong reasons. The role of TCoE - in developing AI based tools - is paramount, to bring in this transformation, especially when the right ingredients are just around the corner. Such a merger was bound to happen. A London based company 'DiffBlue' received huge sum of funding just for doing that.
Through empirical process control development models like Agile Scrum – being both iterative and incremental - many of these scaling discomforts have been squashed down within short ‘time-boxes’ - which promoted 'early feedback' over 'requirement analysis' - based on ‘traditional QC models applied on smaller cycles’ and ‘post-fixed business driven retrospection enabling course-correction – a heavily misunderstood and misused feature of Agile Scrum, an aspect which used to be handled earlier by traditional QA’. A shift-left approach (BDD, TDD or DDD) driven by TCoE can prevent the danger of common QA aspects reinvented and pushed later into the cycle, while letting the 'agile retrospective' handle rather unique and complex scenarios.
A sudden pragmatic shift in development process – similar to one described above, crippled traditional TCoEs to some extent, with no available ‘scope & speed’ to generalize over newly emerging domain specific testing needs and models. Even though, a few individuals got well-groomed within a vertical – they too became too indispensable to be shared with another vertical, let alone with any testing cost centre. This ridge was destined to grow. In fact, this natural phenomenon can be explained as a ‘cognitive misfire’ - in any exponentially growing field obeying Moore’s law. On top of that, what if, the ‘sprint durations’ and ‘life-cycles’ get shorter and heavier – and show up symptoms of ‘seven Vs’? Doug Laney, an analyst for Gartner, once described Big Data as consisting of the three dimensions of high volume, high velocity and high variety. Researchers discovered more dimensions and found similar dimensions governing other areas like IoT, Machine Learning and even human-mind. Today, applied CTM (Computational Theory of Mind) and Artificial Neural or connectionist networks used for machine learning accommodate thousands of similar dimensions – while there aren’t enough V-words to name them. Some of these can’t even be mapped to any physical reality.
Testing has also been blamed for being a redundant confounding variable – mainly when they were administered in a decentralized fashion – because of the way it could negatively impact both outcome and development process [Handbook Of The Philosophy Of Science, vol 13, Philosophy of Economics, Uskali M?ki (Series Volume Editor), 2012 ]. In Agile Scrum, probably for this reason, testing was prescribed to be operated like an indispensable but interchangeable role – not to attach any authority or person with it, but more like a feedback system insourced from our own bicameral mind concerned only with smaller survival goals [The Origin of Consciousness in the Breakdown of the Bicameral Mind, Julian Jaynes, 1976]. Who knows, if - like a ‘dichotomically self-aware third eye’, both generalized and specialized at the same time – ‘stripped off’ of any persona - is where testing is going to land?
Certain aspects from Analytics, Lifecycle Management and DevOps - are trying to cut across ‘testing’ in mutually debilitating ways – which will apparently not let it function anymore as an independent external entity. Or rather, as something more fundamental and personal, it probably came of age to exist as a self-aware organ to be retrofitted within an organism, i.e. customized for an individual to stay embedded as a self-aware component – to fulfil the unique survival needs of its unique host. In other words, with all of these, testing is destined to become a superior driving force - in a way how ‘mathematics’ is neither a vertical nor a horizontal – but something more pervasive. Next-gen TCoEs should bounce back taking up this huge opportunity, as platform providers - to foster this change.
A small metaphor from nature will help appreciate the rationale behind how self-aware organs can lead to exponential leap. Take our ‘eye’ for an example. Eye can see and manage by itself. We don’t explicitly manage its intricate inner functions. We rather take this gift for granted and take complex real-time decisions in infinite other areas by mixing the visuals with past experience in real time. Testing, like an eye, can evolve into this state within a very short span of time – say less than a quarter decade. If this is not true, then IoT and Digital revolution are also not true. It is like running a bullet train in a high-seismic zone and not expecting an earthquake. Contemporary testing will quickly evolve into a more powerful ecosystem – which will need more testers than ever – with ever-growing expectation from newer skill-sets.
Terms like ‘pervasive’, ‘omniscient’, ‘ubiquitous’ sound so much ‘Freudian crowd control’-ish – re-emerged via IoT, Social Media or Cloud (The Engineering of Consent, Edward L Bernays, University of Oklahoma Press, 1955). A close retrospection reveals, these are rather catalysts fuelling the emergence and evolution of a language-like symbol system to bridge the gap between human and machines - to be able to communicate at the speed of thought – by adding self-aware automation layers underneath - wherever we are inefficient compared to machines. TCoEs can not only access this layer but also can manage and become this layer.
The best rule for managing a complex ecosystem is ‘not to specify many complex rules’ but rather letting it evolve itself through a set of simple rules. This is known as ‘Instrumental conditioning’ [The Behavior of Organisms, B.F. Skinner, 1938]. This proves why, a ‘Hand up’ approach preferred over ‘Hand out’ is probably the simplest way to achieve more powerful result – not only in socio-economic reform but also in AI and genetics [Dr Craig Venter, creator of artificial life nicknamed Synthia 3.0, 2010]. Reinforcement Learning has applied it with huge success. The AlphaGo team at Google's Deep Mind even called it ‘general purpose AI’ when they decided to let it learn by a simple set of rules pertaining to reward and punishment.
One might need more experimental feedback to further justify the placement and timeliness for self-aware and self-improving organ-like testing-components embedded within the lifecycle of development and deployment. But the path is clear, with milestones already laid. The way machine learning and its recent development in last two decades came out into applied territory - with the great tools and ecosystems made widely accessible - than just being a subject matter of ‘research, experimentation or fiction’, we can happily identify our current generation as – ‘the era of applied AI’. What was achievable two years back - by ‘a large team, a research lab and a lofty budget’ - can be achieved today by a good programmer sitting in a café – if given access to the right data. The abundance of platforms have exposed us to overwhelming opportunities. ‘Spoilt for choice’, we, the team at HG Health Grid have decided to focus on AI based testing & governance in healthcare – probably the first of its kind; and will soon demonstrate how the same can be inorganically extended into other domains like finance, e-commerce, supply chain and retail.
Incubated in 2014, the team at HG Health Grid has heavily researched and decided upon a generic schema by interacting with various companies, domain experts and people related to testing and AI. With real and synthetic data used for training three different prototypes, we are closely working with an US based healthcare analytics company, two leading CRE (Commercial Real Estate) MNCs, 5 large AI vendors and four universities. Due to lack of access to sizeable data and scenarios, we are eagerly looking forward to shift to a larger base – preferably a vantage point of an MNC - known to take up large projects.
Good Reads :
1. Facing the future of software testing one change at a time
2. Testing in 2020 - The future of Software Testing
Architect, System Engineering, Embedded, AI and Digital space
8 年Nice refresh read.
We're hiring!
8 年Excellent read Ripan.