The media is full of stories about Artificial Intelligence (AI) – both its huge promise and its possible negative effects on all aspects of our society. But what impact will it have on the AEC professions? Martyn Day explores
If we are to believe films like Blade Runner, 30 years from now Artificial Intelligence (AI) will have progressed so far that we can replicate ourselves, so precisely that we find it almost impossible to tell ‘them’ apart from ‘us’. As we jointly populate a polluted planet, where farmers harvest weevils for protein under acrid orange skies, surrounded by the decay of our concrete cities, we can look forward to a struggle for survival as ‘they’ seek to escape their weak and stupid creators. And running at 2hrs and 44 minutes, we can all turn against the director who clearly forgot how to use video editing software.
Dystopian futures have become the de facto standard when contemplating the potential fruits of our technical prowess and AI is possibly the most exciting, daunting and feared technology under development today. It is seen as the sword of Damocles, as while it has the potential to help us across a huge range of tasks, it could also make many of us redundant, irrespective of education level or type of job. The deployment of AI and automation is expected to impact doctors, lawyers, accountants, financiers, as well as possibly architects and engineers.
With today’s cloud infrastructure, all there needs to be is one algorithm per job function developed, having learnt from thousands, millions of previous legal cases, patients, buildings, construction sequencing and, with the power of the cloud, it can be flicked on and be everywhere – an overnight global job grab. Plus, the more it’s used, the more it learns. In this vision of the future, AI is more 2001: a space odyssey ‘HAL’ than Tyrell Corporation replicant, so you might want to put off going on that spacewalk; it might not let you back in!
Bringing matters closer to home, a recent Instagram video post by New York-based Chilean designer Sebastian Errazuriz, went for the jugular.
Over a warm beverage, Errazuriz delivered his warning, “I think it’s important that architects are warned as soon as possible, that 90% of their jobs are at risk. If you haven’t really realized that, you should be taking measures right now, as soon as possible.
The reasons are quite simple. When you happen to be in the type of project, the type of field in which experience requires two to three years, it’s almost impossible for you to compete with any kind of machine that can immediately have 10 times, 50 times, 100 times, a thousand extra million times your experience. “The reality is that we already have enormous quantities of data, enormous quantities of blueprints and models that hundreds of thousands, millions of houses that have already been developed. Why do I need a new one? They tend to be the same and can be packaged in very similar systems. Why shouldn’t I just go into an app, and choose what kind of household I want?
“Architecture as an artistic practice, is the only one that will survive, and it will be developed by a tiny elite. We’re talking five percent, one percent of architects, max. The rest. They are done. They’re doomed. They’re gone. Finito. This is the end. Muerto!” Errazuriz concluded.
There are a series of five video posts as Errazuriz reacts to comments from mainly outraged architects and well worth a watch.
Errazuriz’s post has certainly caused a commotion and is undoubtedly a simplistic view based on some core AI vectors. Removing the emotion for a second, within it there is some logic to potential future capabilities, but AI as the master builder is a long, long way from what is available today.
Errazuriz does talk about the exponential nature of AI learning and how with enough data, an expert system can easily assimilate all our knowledge, and this is undoubtedly happening. But architecture is essentially creative, shape, form, light, materials, systems and very personal. Conceptual origins of things that are very hard for computers to come up with.
While there are AI algorithms in research which have learnt techniques from Rembrandt, Canaletto and other great masters, and can do impressive party pieces, you have to question if an AI-based system could come up with Cubism, Pointillism, Brutalism, Gothic or Art Deco. Could we ever respect a branded AI that starts a new artistic wave, simply by being a bit random and remixing its data?
There will always be signature architects. However, as we look at meeting the needs of a global population of nine billion in housing and infrastructure, maybe we are going to need a bit of help to get this done efficiently.
From everything I have heard and from talking with developers, the AI systems we will see in our industry will augment our design abilities, not move humans away from the process of design. In the main, they are chiefly striving to automate the mundane tasks and provide oversight to complex processes and interactions.
Getting back to design
Racel Williams, AI development manager at Autodesk has spent the last year travelling the world, talking to Autodesk customers about what their biggest pain points are and what they are really struggling with. Unsurprisingly, these were mainly centred around design changes. So the team is researching Machine Learning (ML) and AI when applied to CAD, PDF and BIM, how the industry documents data and how these objects and components relate to one another.
Looking away from AI and ML in specific Autodesk products, Williams muses on the bigger picture, “There’s obviously lots of things that as an industry (not just Autodesk, but the whole CAD industry) we can do to work better together to figure out how we can be more strategic in what data we’re collecting and what we’re trying to do with that data later on. When we were doing Google Analytics in the early days, like any company out there we were just trawling through this lake of information; there was no strategy and companies were just collecting all this data. Now we realise that we need to be more strategic about what we’re doing with the data to potentially solve specific issues.
“From our side, the industry needs collaborate together to try to figure out what those standards are so we can actually solve bigger problems. As a former architect, I had to do a lot of manual tasks, and that wasn’t what I signed up to do! Generative Design, AI, Machine Learning can help bring people back to where they can, and do what they signed up to do in the first place, which is design awesome things.”
AI accuracy
Keith Bentley, EVP, CTO of Bentley Systems, has a frank assessment, “You will not be a competitive software company unless you’re a good AI company. Five years from now our entire conference [Year In Infrastructure] will be about different types of machine learning. Algorithms are helping with the process of either designing, operating or building.”
Bentley is one of our go-to guys to understand the impact of new design technology. In 2018, the company acquired AIworx, a machine learning and Internet of Things (IoT) developer, to specifically bring in house a team of AI programmers to boost the company’s Digital Twin advancements. Bentley highlighted areas where the AIworx team has been adding its AI know how, recognising information in context, via image recognition (finding cracks in highways and bridges or rust on towers). The second area for AI attention is recognising patterns in BIM models.
Bentley explains that, unfortunately, a lot of what people create today and what they believe is an accurate representation of things is, in fact, incomplete. When data is collated and mapped within an iModel, it’s often the case that there is inaccurate data, incomplete data, inconsistent or redundant data. In the future, Bentley’s software aims to do a much better job of sorting out consistency, and checking for standards, with algorithms. Bentley sees solutions in offering smart project predictors and many other tasks.
Autodesk’s Racel Williams agrees that there is an issue with model quality checks and rules-based solvers, “We don’t want people to do that for the rest of their lives!” she notes, “We are trying to understand those relationships and work out a way those people don’t have to write rules again and again. We are finding the quality of drawings and the quality of modelling isn’t consistent across the board, but first we need some understanding of what ‘good’ is like, what makes for a ‘good’ model?”
AI in BIM
One of the more recent and interesting examples of the application of artificial intelligence comes from Belgian developer Bricsys. Having developed a competent AutoCAD compatible product, the company has turned its eye to creating a new BIM tool. Instead of modelling with an array of pre-configured components – Lego CAD, as it were – BricsCAD BIM allows the designer to work with solid geometry, crafting the shape and punching holes for windows and doors. Once happy with the form, the designer simply types BIMify into the command line, then AI goes through the model automatically identifying building components such as walls, doors, columns, floors and assigning IFC tags. This is the antithesis of current working methodologies and once more allows architects to experiment with form at the conceptual phase, knowing intelligence can be added afterwards.
BricsCAD BIM now supports Rhino and Grasshopper, so it’s possible to bring in mesh geometry from pretty much any design or visualisation system. It means one could BIMify dumb geometry from products such as SketchUp, Blender, Unreal, Unity or AutoCAD etc.
The next challenge for Bricsys is to apply this AI to the scanning world. Working with Leica and HOK, the plan is to scan buildings internally and externally and let the AI interpolate between the external and internal meshes, to interpolate the voids where the scanners can’t penetrate. It’s possible to ‘infer’ the walls, windows, floors, doors and columns. Scan-to-BIM may actually become a reality.
As Graphisoft has been rewriting the core features of its flagship BIM tool over the last few years. ArchiCAD has seen some subtle uses of AI in features such as its stair design tool, for instance, which now uses predictive technology to automate the generation of complex forms, whilst taking into account hundreds of design codes. Also, Graphisoft has deployed AI behind the scenes, to optimise software performance with a self-learning multithread- balancing algorithm to optimise navigation predicting what to cache and where a user is likely to move in a view.
Akos Pfemeter, vice president, marketing told AEC Magazine, “We believe it will be mission critical for the future of our industry but, in our opinion, we have still yet to see a breakthrough practical application that would move the needle for our industry by and large. In development terms, the company’s next focus for applying AI will be automation of mundane tasks such as annotation and layout.”
Limiting risk
From the conversations we’ve had in researching this article, it’s clear that only a small number of AI-enabled features and applications have found their way into our industry’s core products. As the biggest player in the market, it is perhaps to be expected that Autodesk has the widest AI footprint but even here, it’s still a niche technology that is appearing in a number of its acquired web service products, namely Construction IQ, which helps construction project teams manage risk and improve performance, and Building Connected, which is used for online bid management.
Construction IQ is Autodesk’s most mature AI solution. Manu Venugopal, senior product manager, building information, explained the aims of using AI in the product, “We are trying to make our tools more assistive when they are dealing with a lot of information and data. We use AI and Machine Learning to learn from the data, both past and present in design and construction, and assist in that decision-making process. We have more than 1,500 active projects now and are working with many companies, including BAM, AECOM, and US-based firms PARIC, Swinerton and Danis.” Venugopal added that Construction IQ has a database of over 30,000 construction projects, which had 150 million issues and related inspections from which to learn.
“With recent developments we have been taking this intelligence slightly upstream into the design and construction space, ‘the design risk management aspect of construction’. We found that many of the issues that Construction IQ is finding, there is some link back to the design and preconstruction phase. Customers want to apply the technology even earlier.
“It turns out that over 70% of RFIs in construction have a root cause in design and documentation errors [from Autodesk data science team’s research] Our research also showed that 38% of all the litigation problems, came from the design and documentation phase [from Engineers Daily (2011)].”
In short, AI will be checking the meta data in submittals, together with components which have been marked-up in BIM 360 Docs. This new work was launched at Autodesk University in Las Vegas.
From Uber to dating
There’s nothing wrong with taking inspiration from how some of the most impactful web service providers have deployed AI. Autodesk’s Building Connected web services (available in North America only at the moment) use machine learning in two of its core services: Bid Board Pro, which helps subcontractors track projects they have be invited to bid on, and, in BC Pro, which helps general contractors find and invite subcontractors to work on projects.
Chelsea Hodge, product lead for Autodesk’s Building Connected, explains, “Typically sub-contractors are getting dozens of invitations to bid from different general contractors every week. Some are even getting hundreds of invitations each week, and this manual process can hugely painful.”
Bid Board is a centralised service that tracks and logs these bid invites and ensures they don’t get missed. Bid invite emails are forwarded to Building Connected, which parses each invite with natural language processing, irrespective of layout variation, and formats them consistently for system access. Any due dates or deadlines get logged and warning of impending deadlines appear. This is similar to the way Expensify (expensify.com) or Tripit (tripIt.com) work.
When it comes to match making contractors and sub-contractors, it’s a recommendation issue, Hodge explained, “We use statistical modelling to predict the likelihood of a given sub performing a given trade in any given location. And our inputs into the model include things like the sub’s previous actions on Building Connected, their trades and service areas in their profiles and other factors.” Hodge admitted they took inspiration from Uber passenger scores, combined with a bit of online date matching.
Conclusion
The topic of AI is so mired in societal fear of its implications that it clearly needs to be knocked off its pedestal. Most of the coverage in the media looks at the concerns of governments’ utilisation of face recognition, tracking and its application to big data from insurance companies and social media networks. While of course, those concerns may well be valid in the very long term, it somewhat obfuscates the positives that AI can bring to other areas. AI will save lives and it will enable better buildings and free up more time to do more design.
From talking to industry leaders, it seems the next five years is going to be a crucial time for the development of AI in the industry. AI will be in most applications; it will be on demand through cloud services. AI will be watching project management; construction sites and it will be deployed from concept through the whole product lifecycle. What we are seeing today is really only the start, as the major software developers are only just building out their AI teams and so, even before we see tools, software companies admit that they have to sell AI internally to their development teams, as well as learning from the customers what are the low hanging fruit that they should be tackling with this new capability. It’s a clear sign how far we have to travel before we have to worry about replicants, terminators and murderous mainframes, not least because Autodesk believes we have still yet to define what ‘a good’ drawing or model is.
My concern is not about job losses. The problem, as always, is human. As a child, I studied mathematics with my father and he used to get irate at me for being lazy and using a calculator. I thought I was so lucky that I lived in the time when we were allowed to use calculators in class and in exams. However, my father realised when checking my work, that while I would get the overall process usually correct, I would get the answer wrong, as I was literally accepting the result that the calculator was spitting out, without any contemplation as to whether the number was vaguely in the right ballpark. I learnt to at least mentally pre-calculate the target area and where the decimal point should be!
In an increasingly automated world, our AI work assistants can only go on the data that is input and usually that data will come from us. If we make incorrect assumptions, or put bad data in, there is a danger that we just accept the outcome and move the process on, underpinned with bad data. It is perhaps not unsurprising that many of the early AI developments are there to check quality and coherence of data. Mental atrophy from relying on computers could be a problem.
Always remember: garbage in = garbage out.
Now we have seen what AI is currently capable of and who are the main developers. In subsequent AEC articles, we will look deeper into AI, to what hopes developers and designers have for this technology.
Artificial Intelligence (AI) explained
According to Autodesk’s Kyle Bernhardt, AI logic can come mainly in two forms: Machine Learning (ML) and human-defined. “Machine learning is a particular technological approach that takes large scale datasets to train a set of algorithms to produce an output that would otherwise be very challenging to write with manual code, such as determining if there’s a dog in a picture. This is essentially brute force computing and especially useful in pattern recognition. It is a subset of AI. The more data it has, the better it should be. Bernhardt expands on the theme, “AI is really all about codifying advanced logic, the expertise of the human condition, in a technological way. We talk a lot about that idea of delivering more, in a better way, with less of an impact; it’s something we take really seriously. AI is all about enhancing the intense creativity of the creative professionals; giving them a superpower. Machine learning is one of our best vectors to deliver on that promise.”
Eleven AI start-ups to watch in AEC
Read our article in which we explore some of the emerging AEC focused software tools that are harnessing the power of Artificial Intelligence.
If you enjoyed this article, subscribe to our email newsletter or print / PDF magazine for FREE
