I attended the inaugural Hexagon Defence Industry Advisory Council at HxGN LIVE expertly facilitated by Ken Chadder and Hexagon’s own Steve Duplessis. The session raised some very interesting ethical issues in the use of AI for data collection in some military use cases and cautioned that it’s not always the promised land of milk and honey claimed by technology developers.
Whilst a PriceWaterhouseCoopers 2017 report estimated that “artificial intelligence technologies could increase global GDP by $15.7 trillion, a full 14% by 2030,” its obvious the AI is a technology that is transforming every walk of life because it enables people to rethink how we integrate information, analyse data and use the resulting insights to improve decision making. Being a late 1980’s and self-confessed map anorak, I’m interested in how AI can impact geospatial data collection to create foundation feature databases.
Sure, AI can save time to an unprecedented level only dreamed about 10 years ago, but what happens to data accountability? Who becomes responsible for the integrity of data and can they vouch for its provenance? This is especially pertinent when lives might be at stake, civilian casualties for example. And tough questions rightly need to be asked in the event that an artificially made decision in data collection results in any unnecessary fatality. Who or what is accountable? And how do we get answers for those questions when they lie in the implementation of an algorithm. Does the coder bear responsibility, or their employer? Or is accountability absolved?
Geospatial Insight develops machine learning tools that extract intelligent attributed building databases, principally for the insurance underwriting and claims industry, so we are aware of the value of enriched databases in understanding risk better, but there’s a parallel content requirement in defence mapping. The MGCP programme is now evolving into producing larger and larger scales with products such as Urban Tactical Planner (UTP) and Urban Feature Data (UFD) so the prospect of applying our ‘PropertyView’ solution to MGCP looms large. Unlike the insurance industry however, there isn’t the same spectre of accountability, but we also recognise that to undertake 100% check on all data is self-defeating. The solution lies in the specific use case. Geospatial Insight can produce digital polygons of building footprints direct from satellite imagery and also attribute those polygons with AGL height, roof geometry type, roof construction material and more – all automatically. The confidence we hold in those attributes is a function of the quality and quantity of training data used in creating the all-critical decision rules. We’d advocate however that where safety of life is an issue that man-in-the-loop is introduced, so in that instance AI becomes a means to an end, not the end in itself. And likewise, meta data needs to ensure that with the advent of AI based data collection, standards such as ISO19115 also evolve to ensure that ‘data source’ is captured against each attribute.
AI is most definitely a force for good, but we mustn’t take that for granted and understand that the face value might not always be what is seems – or promises. The case continues…
