Microsoft is currently running ads extoling the virtue of AI and IoT sensors in helping farmers produce more and better crops, with less waste and higher yields. Elsewhere in manufacturing, supply chain management is being transformed with digital maps of goods and services that reduce waste and logistical delays.

In Finland, a combination of AI and IoT is making life safer for pedestrians. The City of Tampere and Tieto built a pilot system that automatically detects when a pedestrian is planning to cross the street at an intersection. Cameras at intersections accessed algorithms trained to detect the shape of pedestrians with 99% accuracy then activated the traffic lights to stop traffic.

Low Latency, High Expectations

There is a common thread in all these examples; sensors at the edge are used to send data to algorithms in the cloud trigger a response. All the while the data is collected to improve the algorithms to extrapolate trends and improve future systems. These examples show that IOT and AI already works well it responds to pre-scripted events such as a pedestrian appearing near a crossing, or soil drying out. The machines have already learnt how to deal with situations that would be expected in their environment. They are not so much replacing human decision-making process but removing the chore of having to make the right decision. All good.

Low latency is essential in any AI and IOT application for industry or agriculture if the right response is to be sent promptly to the edge from an existing library of algorithms. But what if the edge devices had to learn very quickly how to deal with a situation they had not experienced before such as an out of control wildfire or unprecedented flooding on agricultural plains? Here latency is only part of the equation. The other is the potential availability at the edge of massive amounts of data needed to decide on what to do, but edge devices by their nature cannot typically store or process such levels of data.

IBM has written a research paper on how edge devices, in this case drones sent to monitor a wildfire, could perform a complex learning operation, simultaneously model, test and rank many algorithms before deciding on the appropriate analytics that will be deployed to the edge and allow firefighters to respond. This is much closer to a truly intelligent model of IoT deployment than our earlier examples.

In the IBM example, Cognitive Processing Elements (CPE) are used in sequence to assist in making the right decisions to help stop the fire spreading, and understand how wildfires behave in extremis – in itself a not well understood phenomenon. Therefore, can we create a hybrid IOT/AI/Cloud architecture that can intelligently process data at appropriate points in the system depending on circumstances? It’s not just in natural disasters it may help but in another great hope for Ai and IOT: the fully autonomous vehicle.

Who Goes First, Who Goes Second?

Currently, driverless cars are totally reliant on pre-existing algorithm and learnings – such as a red-light or the shape of a pedestrian in the headlights to make decisions. We remain a long way from fully autonomous vehicles, in fact some researchers are now sceptical of whether we will ever achieve that point. The reason is that human car drivers, already act like the intelligent drones featured in IBM’s research paper – but uber versions of such. They not only have access to massive levels of intelligence but can process it at the edge in real time to make decisions based on their experience, intelligence and, crucially, learnt social norms.

Consider the following example that occurs millions of time every day on Europe’s narrow, crowded suburban streets to see how this works. Cars will invariably be parked on both sides with only a gap for one to pass in the middle. What happens when two cars approach: one or the other must give way – but which one? And how many cars are let through once one driver takes the passive role? Somehow, in 99.9% of incidents, it just works. One day we may be able to say the same when two autonomous vehicles meet each other on a European street!