Jadu Research & Development Engineer Mike Smart tells us about the part Artificial Intelligence could play in solving the nation's pothole problem.
Research and Development Engineer Mike Smart gave us a mind-blowing introduction to the all-new Amazon Lex chatbot and Jadu CXM integration proof of concept, at last year's London Academy.
Here we catch up with Mike about the part Artificial Intelligence could have to play on solving the nation's pothole problem.
Mike, you’re a research and development engineer, why are you looking at potholes?
Now there’s a question and one that I’ve asked myself a few times recently! But on that subject, potholes do really get to people. Have you seen the anger they provoke?
I’ve seen people on the news drawing inappropriate images around them so that councils have to act. There was this one chap putting dolls into water-filled potholes to create images of a sort of mini jacuzzi, which worked in creating some buzz and getting attention on the problem.
While this can be quite funny, the frustration is definitely real! Potholes can cause real damage and therefore it’s something people expect to be taken care of. And rightly so.
Councils are actually fixing more and more of them but spending a lot of money in the process. So, how sustainable is that level of expenditure? And, is there a better way technology can help in solving that problem?
VIEW: Videos From The Latest UK Jadu Academy
So what things are you looking into Mike?
Jadu and our partner Pitney Bowes have really elegant solutions already in place, but with regards to R&D (the stuff that’s a little further off), we’re teaching A.I. about potholes... that’s not something I ever thought I’d hear myself say.
Councils up and down the country use our software to manage service requests and we’ve been experimenting with robots - specifically using Amazon LEX, which is a service for building conversational interfaces using voice and text.
The technology is great at identifying natural language and intent but is not great at understanding descriptions of the likes of ‘potholes’...
The big limitation of conversational integration of robots is the contextual data that the intelligent service is trained on. Contextual information really matters! The technology is great at identifying natural language and intent but is not great at understanding descriptions of the likes of ‘potholes’.
Why is that?
Amazon has a ton of existing information they’re training their robots with. But I can’t see a time in the near future where the Amazon committee sits down and decides it's going to implement pothole descriptions into their Alexa framework.
However, they do provide a platform that allows us to train the robots ourselves.
You may ask, “Why is that important?” Well, it’s common to see demos of A.I. being great for taking orders for pizzas but you quickly realise that to take that and make it useful for the likes of potholes, you’re going to need 10,000 descriptions of potholes to teach the A.I. How many descriptions can you think of?
Errr, big, small… quite bad... I don’t know, it’s hard!
Hmm, exactly! I think I got to about five and stumbled. It’s quite difficult to be creative when describing potholes. But, we have those real-life descriptions, thousands of them that people have reported to their councils. So we're taking that data and training the Amazon service for our council customers.
Why is it important?
The automation is not about replacing humans but freeing them up for when human intervention is needed.
In the past year, a pothole has reportedly been fixed every 17 seconds. And that requires a lot of resources that could, perhaps, be better spent elsewhere.
"...if robots can free up the time of people who do know and who can address the big questions, then that can only be a good thing!"
For example, some believe that the sheer volume of potholes is down to short-term ‘patch and mend’ work and that attention should be focused on the resilience of road surfaces. I don’t know about that, I’m certainly no expert on roads and infrastructure. But if robots can free up the time of people who do know and who can address the big questions, then that can only be a good thing!
Plus, cost efficiencies are always welcomed by councils that are stretched and by residents that hate seeing reductions in services, and especially hate potholes.
So what next?
The acquisition and execution of this training data is an area we’re working hard on across the board. It’s applicable to many services and is, in my view, the critical factor in developing and deploying a truly conversational service.
We’re at varying stages of testing this stuff with councils, so keep your eyes peeled and we’ll keep you updated!