Under the sub heading of ‘Opportunities & Challenges for Geospatial to 2030 – the countdown has started’, a panel consisting of contributors to the Association for Geographic Information’s (AGI) Foresight 2030 Report came together at GEO Business. They explored themes from the report, and discussed how the narrative has already changed in six months since the report was published.
Key Takeaways:
- We need a different approach if we are going to create data that solves local problems
- AI is nothing without data. Without data there is no AI
- There is a stronger risk from data invisibility than from poor data
- GIS needs to move away from being a technical, specialist kind of discipline, and become more of an enablement discipline
- Did you become a geo professional to draw a map or do you really want to be at the centre of wider organisational discussions?
- Garbage in garbage out will always be a truism, however what we need to recognise is that AI systems have no sense of smell. They don’t know its garbage and that is something we have to address, full on now. We need to make sure we have data hygiene.
- Let AI do what it does well which is to create ‘glue technology’
- Advice for early careers professionals – the 3 (or 4) ‘C’s; the ability to communicate, the ability to collaborate, and the ability to think critically, as well as be curious
The AGI Foresight Project was a 12-month exercise which sought to engage with both the mainstream geospatial industry in the UK and beyond, and also with those on the fringes of the sector. Building on previous work undertaken in 2010, and then repeated in 2015, the AGI extracted key opinions and insights from the geocommunity in the form of an online survey, engagement with partner organisations, and extensive in-depth, one-to-one interviews. The result is an extensive report providing a clear view of the opportunities and challenges facing the geospatial community, alongside a road map of recommendations.
Six months on from publication, the six themes identified within the Foresight Report 2030; Data, Artificial Intelligence, Interoperability & Infrastructure, Skills, Collaboration, and Earth Systems, are continuing to have impact, and are already evolving.
Under the banner ‘Who ate my lunch? The AGI Foresight Report 2030 – six months on’, a panel, chaired by one of the driving forces behind the report, AGI Director and CEO of Map Impact, Richard Flemmings, came together at GEO Business to discuss and debate some of the ideas and concepts raised within the report.
Featuring contributors to the Report, Rebecca Firth, Executive Director of the Humanitarian OpenStreetMap Team, Mia Dibe, Senior Data & AI Strategy Consultant at Arup, Ed Parsons, Digital Geographer and Consultant, and Will Cadell, Founder and CEO of Spark Geo, some of the topics introduced within the report were discussed:
Minimum viable data vs minimal viable governance
Described as the minimum information needed to solve a problem, the concept of minimal viable data challenges professionals to identify, create, and even use data that is fit for purpose, rather than fixating on data for data’s sake and the continued focus on perfection and non-scalable standards. Citing examples and experiences from the humanitarian sector, the ethos of local people, using local tools, to create open knowledge, was proposed. Whilst many may assume this means personal mobile devices, in areas where organisations such as HOT operate, even rulers are not considered local tech so using fingers to measure flooding from storm surges means that data can be captured quickly and used to inform mitigation. In the words of Rebecca Firth, ‘There isn’t a business model that applies to the settings we care about, so we need a different approach if we are going to create data that solves local problems.”
She also challenged the wider geocommunity to consider this approach. Rather than focusing on what does good data look like and instead identifying the problem and potential solutions, she argues that this concept could save time and money whilst creating usable and scalable outputs and processes.
Aligned with minimal viable data, was the idea of fitness for purpose for a specific problem. Frameworks, such as those proposed by the Data Management Association (DMA) are gaining traction within both the geospatial and AEC sectors fuelling the transition from obsessing over good quality data by pairing with perfect data maturity. Introducing the idea of minimal viable governance, it was suggested that this would allow for the concept of minimal viable data to flourish by designing trusts in data with recognition of data ownership and stewardship.
The idea of trust, especially of data around the built environment, was enforced with appreciation of its importance for transferability, interoperability, usability, together with the importance of understanding how the data was created, managed, stored so that it can be selected and used correctly.
Data, governance, and the impact of AI
A third of the way into the discussion and surprisingly it is the first time that AI is mentioned! The idea that data forms a deep horizontal across multiple, if not all, sectors was suggested and not for the first time. It was proposed that there is something almost human about data as it develops a character often reflecting how it was captured and by whom.
It was agreed that with the approaching AI future, the importance of data cannot be underestimated. Data is a prerequisite of AI, so robust data standards and governance to protect data and correctly put it to work are essential. The tools that can interrogate data are one thing, but the concept of minimal viable data and standards, require careful consideration.
Weighing into the conversation, Ed Parsons proposed, and asked the audience to reaffirm, “AI is nothing without data. Without data there is no AI”. Expanding on this he suggested that all of the intelligence that AI systems appear to display, is the result of the data on which it has been trained, and the importance of this in understanding the results and having trust in the systems is fundamentally dependent on data.
However, what happens when data that was produced to solve a specific problem, potentially to a minimum viable standard, is used to train an AI system? It becomes less transparent, less visible, and maybe even contributes to an hallucination in that system because track has been lost of why the system is making that decision. So perhaps more time needs to be spent making sure the data collected is accurate and is the best reflection of the information that we are trying to portray and to creating appropriate metadata that reassures not only human users but potentially machine users. An argument therefore against minimal viable data and one for greater quality and visibility?
Countering this was the argument for at least some data, with missing data having significant consequence, as the user of AI is not aware of the impact of data poverty. Rebecca Firth went further than that arguing that there is a stronger risk from data invisibility than from poor data.
Thinking about data providence and the chain of custody of data, especially in the context of earth observation in the low orbit, it was suggested that with many sensors in the sky actual change can be measured from a variety of sources so it is possible to create a robust opinion. However, event driven change, captured by only one or two sensors, may require an anti-doctoring capability, combining the notion of custody with the sense of fitness for purpose, to improve trust in AI systems and outputs.
Enabling trust in AI software for the geo future
Transparency, documentation, understanding of processes, are all well and good, however it needs to be recognised that, at least with the current generation of machine learning, of deep learning tools, that they are all probabilistic as opposed to deterministic. In other words, any particular run of a system will give outputs based on the probability, calculated at that point in time, from the data that has been used. This is a different way of working and means that users need to keep testing results using domain expertise to recognise what looks right, and as importantly, what doesn’t.
From the built environment perspective, thinking about things like digital twins, there is a transition of geospatial and GIS as a speciality. Looking at agile ways of working, it is not just about the day-to-day technical task, it’s about data, governance, and the recognition of what is happening in the wider industry. In other words, GIS needs to move away from being a technical, specialist kind of discipline, and become more of an enablement discipline.
More than ever, location is key. Whether that is within digital twins, global data platforms, or GIS systems, there are different use cases and different applications, but location is embedded everywhere and so it is inevitable that the GIS specialist, the GIS professional, will need to start adopting a more holistic mindset as to how they operate in the wider built environment. The take being, that a spatial capability has become more accessible with AI, so the GI practitioner needs to question their motivations; did they become geo professionals to draw a map or do they really want to place themselves at the centre of these wider organisational discussions?
If you start adopting an enterprise architecture or business architecture perspective to problems that you know require data governance then, perhaps, you start realising that the role of the GIS professional is about embedding different elements; half domain expert, half data specialist, part connecting different workflows and part coming together to solve problems, because, fundamentally speaking, the problems being faced haven’t changed. The desire is still for efficiency, for productivity, for automation. In the words of Mia Dibe, “It is more about interrogating what the future of the GIS discipline and industry looks like, placing location at the core of the problems we are trying to solve.”
Looking at the future, inclusive of AI, from a couple of different directions, the discussion continued with Rebecca Firth suggesting that rather than a ‘human in the loop (of AI) it should be a human in the driving seat’. She cited examples of where a human reviewing AI output had failed to do the job probably with consequences ranging from a temporary renaming of New York, considered as hate speech, to the bombing of a school in Iran, incorrectly identified as a military base, with subsequent loss of life. She warned of an increasing risk of just accepting AI produced insights without full understanding or rigorous review.
The second direction introduced was that of a lack of admittance around the use of AI, suggesting that this should be an important part of the dialogue. Not only did it take the best part of 15 minutes before the topic of AI was introduced within the discussion, it was also proposed that this was the only session of the day so far that hadn’t used AI in its preparation. It was proposed that AI shouldn’t not be used, rather that its use should be transparent, even, perhaps jokingly, AI branded.
Geo in the driving seat
Returning to the idea of geospatial as a deep horizontal, with everything in some way related to place, there have traditionally been technical barriers to accessing geographic knowledge, making it hard to ask robust questions around navigation or mapping. Google broke many of these barriers, and AI is eying up the rest, however it was suggested that without ‘safety barriers’ this comes with a risk.
Will Cadell recalled a conversation with a colleague from a notable tech company who claimed he could make a map using AI. Expecting disaster, Will described the output as ‘annoyingly adequate’, disappointing, but it contained the information needed, answered the question that had been asked. He did, however, enforce the opinion that geospatial needs to take ownership of data with the comparison of ‘investing in data is like buying a house, and investing in technology is like buying a car’. The more you invest in data, the more valuable it becomes, however once you add a UI it becomes meaningless, ephemeral. He challenged that, if as sector, we care deeply about the presentation of geographic data, we care about algorithmic rigour, then we should create content that at AI can consume. Only that was can you guarantee the rigour, protect the data access, ensure fitness for purpose, create providence, and just let the AI do what it does well which is to create ‘glue technology’. So, AI may have taken jobs, maybe even eaten lunches, but it you let it do what it does well – build the glue, then users can build one off user interfaces, they can answer a question or tell a story, and then they can move onto the next thing. In this way, AI can help address that deep horizontal, everyone can access a GIS, ask a question that they may not even realise is geographic, and be fed with meaningful data.
With AI as another ‘big horizontal’ thing with opportunities abounding for geographers, the similarity to the web 30 years ago was suggested with the world we live in today, with maps in the mainstream, only possible through the transformative effect of the web. It didn’t impact geography itself, the ability to understand the world, the science, understanding patterns, the distribution of resources, geopolitics, etc. that wasn’t directly impacted but, it was suggested, it will be impacted by AI. AI has the potential for a massive impact in terms of making geography more powerful, more relevant, more understandable of the processes behind the scene that are then illustrated by the map. AI can help recognise spatial inequalities in an economic sense, it can help identify the distribution of resources in climate change, it allows for the processing of data, making it more accessible, and that is the opportunity. But it will need geographers to be engaged, to be part of future models.
Models used today have been fundamentally trained on text, generic text found on the internet, a lot of basic geography has yet to be exposed to large language models, it is the job of geographers is to make sure these models understand the principles of geographic information.
Returning to the battle of invisible vs visible, it was suggested that with the visibility of data comes the invisibility of the underlying geospatial infrastructure and that AI has the potential to make GIS more abstract. That may not mean geo professionals are replaced, victims of big data, technology evolution, and big platforms, rather they become architects of the infrastructure. Arguably a nuanced change, as geographers are important, they are best suited to understand and interpret context and meaning, and they know how to ask the right questions, and how to translate the mapped answers so the question is are they ready to change? In the words of Mia Dibe, “We want the right people to ask the right questions with an embedded and overarching holistic understanding of what the invisible infrastructure is about.”.
Advice for early careers professionals
With lots of talk about the job crisis, about AI taking jobs, as discussion drew to a close the panel was asked what is the best single piece of advice you would give an early careers professional at the moment?
Rebecca Firth challenged those entering the workplace to ‘think for themselves’. Graduate positions may not be designed for that, they make focus on learning buttons, being an expert on buttons, but she advocated forming learned opinions, asking questions of answers, interpreting results.
Mia Dibe warned against ‘trusting the job description’. She spoke of a fundamental problem in the education system that fails to keep pace with technology suggesting that new recruits should be willing to unlearn what they know and relearn what they need to know in order to fulfil different roles.
Ed Parsons agreed with Rebecca on buttons suggesting a concentration of what makes and individual human. He offered the 3 c’s as his starting point – the ability to communicate, the ability to collaborate, and the ability to think critically. He also challenged job starters to hit the ground running, think fast and be flexible.
Finally, Will Cadell offered a fourth c – be curious. Ask questions about new tools, look at how things fit together, dial into the question asking space for example ‘why do we project data into a flat space?, ‘why are we still doing old processes?’. As a closing point he argued that we are in a place where we have tremendous opportunity which is being hampered by software. Yes, AI will build some of it but we need people to guide the AI.