You Can’t Write an Algorithm for Uncertainty

You Can’t Write an Algorithm for Uncertainty
defence.gov.au

Why Advanced Analytics May Not Be the Solution to the Military ‘Big Data’ Challenge

X
Story Stream
recent articles

The proliferation of sensors and data sources available to a modern military like the ADF often swamps the ability of the analyst to find what’s truly relevant in the sea of information. The exponential increase in sensors and data sources hasn’t been matched by an increase in human resources to process them. That imbalance makes aspirations of ‘information superiority’ untenable, leaving militaries vulnerable to promises that they’ll have machine solutions for and certainty about what’s an inherently human and uncertain problem: war.

We must be careful about proclaiming a revolution in military analytics and be cognisant of the failed promises of the last ‘revolution’ that occupied Western military attention. ‘Advanced analytics’ is a bet on computers being able to process the data deluge in a meaningful way to support military decision-making. My concern is that we don’t fully understand how difficult that is to achieve, or the significant changes that such a gamble implies for the workforce charged with implementation.

The fallacy of smart computing. Computers are only as smart as we program them to be. In the absence of Skynet-level AI, they can’t interrogate data holdings to generate links between diffuse pieces of information to predict or assess the military actions of a thinking human adversary. Existing software can’t make sense of complex human interactions in the same way, or with the same time-sensitivity, that a well-trained analyst can or should. Much is made of the ability to assist with pattern recognition, and while analytics can certainly assist with that task, it still relies on someone programming the correct patterns to recognise. But understanding what those patterns might look like implies a degree of certainty about the tactical environment that rarely exists on the battlefield.

Workforce design. The quandary we face is whether to design intelligence architecture around unproven advanced analytics platforms to get the most out of the technology, or to design an architecture that supports the analysts to understand the environment in which they work. Currently, with the personnel and technical overheads required to give advanced analytics systems a fighting chance—particularly in the fields of data entry and algorithm development—those two concerns appear mutually exclusive.

Uniqueness of military data. Advanced analytics tools are seductive when designers conduct demonstrations using carefully calibrated data to show their theoretical capability. But military data is rarely clean and is inherently difficult to control. Analysts deal with everything from UAS feeds, to Facebook posts, to scraps of paper and everything in between. Those are unstructured data sources that are ill-suited to the needs of a platform designed to ingest and analyse structured data. Data standards are incredibly hard to control, and ‘cleaning’ data to make it usable is both time-consuming and takes an analyst away from trying to fuse their assessments across data sources. When the ‘data in’ is poor, the ‘data out’ will be wrong. Many of the analytics platforms marketed to the military were developed for finance and industry, where there are limited data sources and the data can be structured to suit the purpose of analysis. That isn’t the case in the land warfare domain, and it’s largely impossible at this point to write an algorithm that can bring order to the chaos that’s inherent to war.

Stovepiped development. Powerful software is available to exploit single-source sensors, but those tools are rarely linked into an all-source fusion tool. Many of those systems are also proprietary software, meaning they can’t be exported into more powerful fusion systems. A sensible approach might be to design the all-source fusion system first and have individual sensor requirements nested underneath. However, that implies a level of capability development and acquisition alignment far in advance of existing stovepiped practice. A continuing challenge will be to find tools that work across the myriad defence systems, classified and unclassified, to provide a unified data environment as part of an enterprise approach to intelligence.

The cognitive shift. Military analysts have traditionally relied on qualitative rather than quantitative skills. Their successes have mainly been based on forming judgements from scraps of disparate information, supported by the intuition that comes from hard-won experience. The skills and aptitude needed to operate advanced analytics are largely the opposite. They rely on programming and coding skills—a quantitative aptitude to order and synchronise data. Those skills are more advanced than simply being technology ‘savvy’. If those tools are to be the centrepiece of any future intelligence, surveillance and reconnaissance enterprise, they’ll require a significant cognitive shift from the intelligence workforce. The questions must be asked: What’s lost in the process? And for what measurable gain?

Tail to wag the dog. Advanced analytics platforms require enormous back-end support to make them work, and maintaining an army of dedicated contractors is beyond the scope of most militaries. Data scientists are the most in-demand profession in today’s job market, and there’s no guarantee the military can access them in sufficient numbers to ensure the functionality of a chosen system.

Militaries must decide what they need advanced analytics to achieve. Only when that understanding is reached can they partner with industry to design the tools to achieve it. This lack of translation between user need and provider solution is the biggest stumbling block to any meaningful progress in the short term. Ultimately, however, we need to understand whether it’s even feasible to expect computers to make sense of war’s inherent unpredictability. After significant work and investment, computers may be able to assist in ordering and sequencing data to make analysis more efficient, but I’ll wager that they’ll be unable to provide any greater certainty than a team of well-trained and experienced analysts who understand the true difficulty of creating order from chaos.

Comment
Show commentsHide Comments

Related Articles