In this article I will dive deeper into Geospatial 2.0. I plan to discuss in more depth artificial intelligence (AI). This is at the core of Geospatial 2.0. Let’s explore the importance of AI to Geospatial 2.0.
Academic discussions and papers I have read recently motivated this conversation.
Prediction is now cheap
In the next two sections I will summarise the thoughts of two academics at the University of Toronto and Rotman School of Management: Avi Goldfarb & Ajay Agrawal.
I share their belief that we are in the midst of a once in a generation moment.
Let’s start by providing some important background and context. Specifically, let’s first focus on cost. The cost of arithmetic and predictions in relation to computers and artificial intelligence (AI) respectively.
Computers do arithmetic accurately and quickly. As computers became cheaper, so naturally did arithmetic. That meant we started looking for arithmetic problems to solve. We started with the obvious: accounting and simply counting numbers for example. From there we looked for less obvious uses of arithmetic, and it turned out there were many: gaming, music, images, movies etc. Computers moved from counting machines to solving a vast array of arithmetic problems.
AI is really good at predictions. Over the last 5 years, AI has got cheaper. Again, that has meant we have started looking for prediction-based problems to solve.
Prediction is using information you have, to fill in information you don’t have.
Prediction could be about the future, but could also be about the present or the past. As with computers and arithmetic we have been looking for and finding more and more applications for prediction. Starting with the obvious: for example predicting loan defaults (can you pay us back) in insurance. From there we moved to the less obvious: medical diagnosis, object identification, autonomous driving etc.
Prediction is interesting. Take for example autonomous driving. This is focused on predicting what a good human driver would do, and creating vehicles which drive like good human drivers. That is a re-framing on this prediction problem.
The key to re-framing is to find processes which need information filled in.
Also Read: Geospatial 2.0 Companies: New Direction. New Conversations
Decision-Making
Prediction is useful since it is an input into decision-making. That makes prediction foundational. But decision-making is not prediction. Prediction is a component of decision-making. Since prediction is now cheap, that makes the other components of decision-making more valuable.
So what are those other components of decision-making? These include data, action and judgement.
- Data – In geospatial we talk much about the new location data now being collected (IOT sensor data, satellite imagery, aerial LiDAR etc), but that data has become ever more valuable since prediction (which uses that data) is now cheap.
- Action is important because prediction is useless unless you do something with it. We can now apply actions to higher fidelity predictions.
- Judgement is knowing which predictions to make, and what to do with those predictions once you have them.
Think of AI as a prediction task. A workflow converts an input into an output. Workflows are made up of tasks. Anybody starting with AI needs to estimate the ROI of AI completing individual tasks in a workflow. List, rank and start with the tasks at the top of that list ie. those with the highest ROI.
Also Read: Western U.S. smoke from fires stretching across the country
Geospatial 2.0
Okay, enough AI theory. Let’s move this discussion to Geospatial 2.0. Now we have discussed that AI is best applied in conditions of uncertainty. And that it is really good at predictions.
Data, prediction, judgement and action are all parts of decision-making. The promise of Geospatial 2.0 is fast, high confidence decision-making leveraging:
- Aggregated static and dynamic location data
- AI powered high fidelity predictions
- Insight-driven actions
Let’s colour the picture with two examples. It is important to remember prediction means filling in information gaps.
a) Retail and Influencing customer behaviour.
Jon just drove up to Starbucks. Based on past behaviour, we know he usually orders coffee only. We also know at lunchtimes, on hot days, he has on occasion ordered a chocolate bar or peanuts. We do not know what he will do today (missing information). The data tells us today is hot, Jon is at the Starbucks on ABC Street and, though a little after 1 pm, it is still lunchtime. Based on historic data, AI tells us there is a 60% chance Jon will buy nuts today, and a 30% chance chocolate.
Prediction: Jon is most likely to buy nuts.
Judgement: Nudge Jon with a nuts promotion.
Action: Send SMS message to Jon’s phone with 30% discount coupon for nuts.
b) Risk management and California Fires
The Californian wildfires now cover over 1.25 million acres, an area larger than the Grand Canyon. They are proving an enormous challenge to get under control. Geospatial 1.0 has been providing maps of where the fire was. But this is a fluid, rapidly changing event, threatening both life and property: a real-time, location-based challenge. Data about the past is of little use for decision-making. The promise of Geospatial 2.0 is real-time insights for geo-aware decision-making. Critical information for those on the ground includes: where is the fire now, and where might it be in the next hour (missing information)?
There is complexity here. But, simplifying for our example, here are our decision-making components:
Data: Terrain, vegetation etc. Sources: satellite, drone, climate forecasts and sensors.
Prediction: There is a 60% chance the fire will start moving due north in the next 30 minutes.
Judgement: Prepare town ABC, which is 5 miles north of current fire perimeter, for evacuation.
Action: Mobilise emergency teams for ABC town evacuation preparation.
Closing thoughts
I share the same belief as Avi Goldfarb & Ajay Agrawal, that this is a once in a generation moment. Cheap prediction, a tsunami of location-based data, and a pandemic which is forcing change, focusing organizations to use data to help drive higher confidence fast decision-making. In some ways the perfect storm.
I’ll finish this article with a piece of advice for my geospatial colleagues. Think expansively. Start to lead and not be led. A very bright future awaits you.
