SDSW23: The future of data and technology from 1Spatial

tl;dr

Seb Lessware, Chief Technology Officer at 1Spatial, talks about the impact of AI on technology and the misconceptions surrounding it. He explains that while AI has gained hype and attention, it is not a new concept and has been evolving over the years. Seb emphasizes that AI should be seen as a tool to assist in tasks rather than a replacement for human capabilities, highlighting the importance of rules and data cleaning in AI applications and how they are used in products like 1Integrate and 1Data Gateway. Seb also mentions the need for continuous software maintenance, security, and cloud-enabled capabilities. Finally, he touches on the trend of data aggregation hubs and the varying levels of data maturity among different organizations.

Transcript

I'll talk a bit about what we see in terms of trends and futures and technology influences that affect us and how we build technology and what we build.

So I apologise, we will inevitably be talking about AI because everyone's talking about it. Everyone's got fear of missing out. Everyone's panicking about it. Don't panic. It's okay. It's not a problem. But I will talk about it because it has some impacts.

I had a look at the Gartner hype cycles. I had a look at this year's one. Looked at all of the things moving up the hype cycle curve. AI is probably sort of launching off the top of it right now. I looked at one random six years before. Completely different. A lot of the things on there don't even appear on the new one. I even looked at one just one year before and they're completely different as well because this is not necessarily about the technology implementation, it's about the hype. The hype for stuff can really come and go in the green box are some of the things that we've seen have an impact on our industry. And some of these things are day to day we do it all the time, it's just normal for us now, like cloud and SaaS and moving down there are things that have impacts and are used like the Metaverse. Remember that, everyone was talking about it last year, thanks Phil for even mentioning it. Disappeared off the radar completely. So you have to be careful about which trends you're seeing happening and how they're impacting the world and customers.

So we see what's out in the press, we hear the customers doing things, we look at sort of technology coming along. We assess all of these things. When it comes to AI, there is some absolute nonsense talked about it a lot of the time. Some of it is positive and some negative. So the negatives, you can broadly break down into four areas. We're all going to lose our jobs, is one of them. Bad people will do bad things more easily, that's another one. We're going to automate stupidity and bias and mistakes, or we'll have nations that will generate the wrong thing. And then the other one is the matrix is going to destroy us all. And even this week there's been a conference at Bletchley Park the UK Government is organising about AI. And this morning I think in the news Elon Musk talking about, you know, it's going to destroy us all.

Actually, two days ago in the news I think it's Andrew Ng who’s a professor at Stanford who worked on the Google Brain project, said actually a lot of big tech companies are stirring up the fear in the hope that legislation will come along and stop all their competitors, and they've got a head start and they're just raising the drawbridge. There's lots of nonsense talked about it. There's also lots of opportunities. People are really excited. Investors are throwing money around about it. One of the reasons it's become so prominent, as it's been happening for years. I went to university, God knows when 25 plus years ago and we were doing AI machine learning and we were learning about it as a part of our course. Things haven't massively changed and it's been evolution of the technology and mostly more data and hardware to run it on. But essentially it's the same sort of stuff.

The thing that made it really take off was Chat GPT, because that was easy for people to play around with. A simple interface, you could log in for free, have a type around. Openai who launched it was surprised by the impact it made. They let it out there as a, have a play with this, it's good because it's interactive, so it's not just one question. It’ll consider your previous answers. They weren't really prepared for everyone playing with it.

And the reason that the moguls get so excited about it is, it was trained on the data that's out there in the real world, in the Internet, so written words. And so it was used to generate essentially it's a large language model. Ok, the L stands for language. The language is text that's online that text that’s online, is mostly English. There’s other languages that are available as well. But some of the text that’s online is Python and C++ and programming languages. So what people found is actually it's learnt of all these symbols, that you can get a piece of text put into a long line, look at the order of the symbols, essentially a very powerful autocomplete.

What people do when they try it, is they try, let’s try a simple thing. Okay, let's summarise what my company does? Or can you write me a simple program to do X, Y, Z? And these are great examples because a lot of the stuff that's online is tutorials that has that information as examples very geared up to the sorts of things that you type in first. Now actually in reality the things so say our software engineers or developers, they don't often sit there and think, write me a simple program to do X, Y, Z. They tasked us to say, make this very complex change to very complex rule base without breaking everything else. And that's the sort of thing that's harder to do with AI. So these tools will become assistance to the things that we do. They'll have useful things like in the same way that I use the spell checker, hopefully correctly on my presentation. It was just an assistance tool for me to work on it.

There are other things in large language models, so GPT three and lots of other technologies are taking things like images as an inputs, you can input image and text, you can get out images and text and that's great for particularly creative things, you know, generate me an image for my presentation that shows, you know, what does a 1Spatial event look like? And it will invent some sort of view of a conference, which is great and that's creative, but actually, here we're talking about data and real data about the real world and we describe that using words like geometry. Which means measuring the world and geodesy, which means dividing the world. And so we have to be careful about stuff. We can't just invent information we can, but it's not useful for our jobs.

And the other complexity with spatial data as well is it’s two or three dimensional. The way that text handle is a large language model, it's a long line of symbols, and the way that images are handled is you break this discrete image with edges into small blocks and you divide up these small blocks to create a long line of symbols, which is fine for a discrete image. But spatial data is two dimensional. It covers potentially the whole world or wider. And so what we need to do is boil out the information from that and use that to deal with it, if we are going to pass into it into an AI model. I'll give some examples of that.

What impressed people about these things say Chat GPT is it was automating things they didn't think was automatable. Actually, that's something that we've been doing for years now for a long time. That was a thing we do. We saw 1Streeworks, that was something that's been tried in the past and was considered unautomatable. But by using in this case, mostly rules, explicit rules and an expert system, we could automate that thing that was very complex. And that's what we do a lot of the time. Where we see things like AI techniques coming in is a support to those explicit tools. So there is a rule book written and we have to follow the rulebook when we're generating these traffic management plans.

Other examples are inferring missing bits of the sewer network. We've got all these assets. We don't know where they are. We need to infer the missing ones. If we speak to an engineer, they'll tell you, I reckon I can guess where they sewer pipes are because look at the houses, look at the age, they’ll probably be going around through the back garden. That's something that we can turn into rules and generate. Now within that we might want some AI techniques to help some of those decision making processes. So when you're laying out your cones on your road, you might need to decide is that widening of the road? Is that a parking bay or is that another lane? We need to work that out. And that's where some specific AI techniques can help to make that sort of judgement. That's called neuro symbolic AI. That's what they call it, neuro symbolic AI, a combination of AI decision making processes and explicit rules written by humans. And that's where that's where it comes in.

So we see a number of places where we've used or we are using or we are experimenting with using AI tools, but it's just a support for these automated rules. And in order to get data that's not biased into the system, you need to clean it up using rules. Once it spits data out, so if one of these examples is is generating structured data from images, which is a classic thing that people have been doing for a long time and trying to improve, if someone said today, you don't start from an empty sheet of paper, you've got data. Most people have data they're trying to maintain and enhance. So spitting data out of an imagery tool that turns it into vectors. A, that's normally quite messy and noisy, and B, you then have to work out how to integrate into existing data. Is that a change or is that the same thing just represented slightly differently? And that's where again we use the rules to get that data in and integrate it into your system.

So these rules are used for cleaning data, checking it, validating it, integrating it, transforming it. So our product suite and our focus is on different ways of applying these rules. And we've seen examples and some examples out in the demos of 1Integrate, which is the Rules engine and 1Data Gateway, which is the portal into the rules engine. We've got some tools that are used out in the field, so 1Capture people on a mobile device capturing data as say they're burying an asset. 1Edit, which is on a Windows tablet or on a desktop. Surveyors out in the field capturing data. They are just ways of having the data, having the rules running literally as you capture the data. So it's not just checked when you send it back in, it’s checked out in the field as you do it. The 1Spatial Management Suite, we have some organisations using that which implements the workflow to make sure the rules get applied. And if the rules fail, what's the flow, what’s the feedback loop to get that checked before the data goes in.

We also have new things for 1Data Gateway, for example, which we're demoing upstairs as well, which is an ArcGIS Pro add-in which directly calls 1Data Gateway. So people don't need to go to the website and upload their data, they can press a button within ArcGIS Pro and have it sent automatically into the server side. So these are all rules based products. And we were talking earlier about maintaining the data. Data is not just about capturing it, like David was saying from Ordnance Survey. The effort is maintaining that data, making it up to date and keeping it up to date. And it's the same with software. So when we talk about innovation, it's not just new features and new types of data and new capabilities.

There's a lot of stuff that happens that's hidden, but is really necessary. So for example, user experience and user enhancement capabilities make it easier to use that something that’s visible and happens as well. And we keep doing that with our products, we’ve recently launched version 4.0 of 1Integrate. There's also things in the background that are less visible. So for example, security and keeping things up to date. All software uses all sorts of third party libraries and integrates with third party systems, and they're always evolving and keeping up to date and have security patches. So you have this constantly moving set of tools and capabilities that we need to keep up to date and then pin test our software to make sure it's secure, as part of the the sort of security requirements.

And the other aspect is around making things cloud enabled and observable and deployable. So things like DevOps and automatically you can press a button and your software will be deployed and ready to use and we automatically tested and automatically shunted from development test to development to production. And then the observability when it's running, having the right capabilities that can be read in by an observability system and say yeah, things are looking healthy. You don't need to wait for a user to say, hang on, there's a problem with your cloud system. You can spot it and you can trap it and you can deal with it at the time. So there's always this background of enhancement and maintenance.

Many of the trends that we're seeing that our customers are doing or asking for or getting to and we've seen them talked about today these data aggregation hubs. So like the National Underground Asset Register Project or the US Census Bureau or High Speed Rail 2. They've all got common requirements which is, I’m trying to aggregate data, and it used to be in the olden days people were quite into it. I've got a department, they capture the data, no one else does. Now there's more integration, there’s data coming in from suppliers, from third parties. They're trying to manage that.

These organisations form quite a wide spectrum of maturity in terms of data. So at one end people like national mapping agencies, so say SDFI or Ordnance Survey have been dealing with data for a long time and data they understand the importance, it's a central asset they need to manage. At the other end of the spectrum, are organisations say in a construction environment, when you're building things and you're getting subcontractors to build stuff and part of the contract is now, don’t just build it, but give me the data at the end to describe what you've built because I want to manage that in an ongoing way. I want to build a digital twin, for example. And so, organisations like Ordnance Survey want to get data in and they've already got good data processes and they might be looking at well, now we're going to make that 3D data.

At the other end of the spectrum, people are saying, just tell me that a file is being handed over and it's got the right name and it's the right extension. I can't yet worry about what's in it. And that's Day 2 from Day 1, this is a big spectrum, but they've all got this common requirement, which is data aggregation hubs. The reason that this data is being captured is not just for the sake of having data. So some people, they're shipping that data out. But once you've got that data in and centralised, then you can do the intelligent analysis in order to make the decisions. And that's where the real value that David was talking about this morning, the value is boiled out of it. What is the reason for capturing it? And that might be AI projects, now AI is not a value, that's a technique you'd use, but some of the value downstream comes from that. It's going to be about analytics and decision making and budget setting. And these are the reasons why you do this, to get the data up to date and accurate. And so that's our focus.

What I will do now is I want to hand over to my colleague Sheila Steffenson. She's CEO of 1Spatial in the US, and she can talk about some real world customer examples to finish us off.

And I've been in the geospatial industry for over 35 years and I realised that this is a missing piece and it was very innovative, even though 1Spatial have been around for a while, it was still very innovative because the way it was architected and what it was able to do with very large sets of data, very expediently and very accurately, really just said to me, This is great innovation. And what I've really appreciated about the continuing innovation at 1Spatial, it's not focused on the hype. It's not focused on one of my pet peeves, which is people who want to demo something and say “Isn't this the coolest thing?” I wanna say “What value does it add?” It's not just cool to be cool, what value does it add?

And so we really depend upon our customers and the input that we glean from them to look at where do we develop on. And so when I first started it was just 1Integrate. And then we heard, well, you know, we really need something that's not as complex for users who may not be GIS experts, but they still need data validated. And so out of that spawned 1Data Gateway. And I am especially excited about the new plug in, and I'll tell you why. And I've talked to numerous of our clients. So like California Department of Transportation and all of our 911 emergency clients who just need things done very simply, very easily and without having to learn a whole new system.

So the plug in will sit in ArcGIS Pro and right in their session as they're going through and editing their data in the system they're used to working in, they are able to hit the 1Spatial button, submit the data for validation, and then receive the spatial report right back in that session that they're working in. And it happens very rapidly so they don't have to wait for a long time to get the information back. And then they can use that as a roadmap on how to clean their data. There are other options. It could also auto clean, but a lot of them, especially in our emergency management, you don't want it to be auto cleaned. You need people to look at it who understand what's really on the ground. So I have probably been bugging the development group a bit about this for several months now and I'm really happy to see it here today. I'd really encourage any of you who are Esri users to take a look at it.

And you know what the other real beauty of it is? To me it's a template because I know there's a lot of CAD users out there too, like using Autodesk or Micro Station or what have you. Same kind of little plugin can be built for those systems too. Again, simplifying and allowing the person to stay in the system they're accustomed to using and just keep using it the way they have been using it, but guided to where they may have missed the mark on some data. Thanks Seb.

We won't have questions right now because of time. So if you have any questions, please grab me afterwards.