In contrast to the common images of horror with which artificial intelligence (AI) is often outlined, in the construction industry it is by all means a useful and helping hand. It is often used where humans have reached their limits. For example, in combination with elevated camera systems that monitor construction progress and safety, or an application that examines 10,000 live streamed data sets per second for anomalies to alert to possible accidents or damage. This is where AI systems provide valuable support and will continue to acquire more task areas in the future. The question now is: How easy is it to establish an AI system on a construction site?
The AI itself needs a lot of helping hands to be able to fulfil the tasks set for it. If AI systems are put into operation directly on the construction site, the system must be self-sufficient, so that it knows how to deal with IT-unfriendly circumstances (especially at the beginning of a construction project), such as a missing or unstable power and network infrastructure. If a stable power supply exists, the question of external communication then arises. Certain adversities can usually be circumvented in the short term by enabling the AI system to evaluate the data on site. However, at the latest after the evaluation, the generated information should be passed on to a reporting or monitoring system in order to be able to make decisions in case of an emergency. This makes it clear that AI systems need communication networks as a strong partner in order to use their generated added value profitably. These communication networks can be used in a variety of ways. AI systems that have been commissioned in a cloud system require a stable power supply at the latest for the transmission of data. WiFi technologies are plentiful, whereby the consideration must always be made here as to whether I want high range or high data volume. Especially in the initial stages, such as site set-up and the shell construction phase, both can only be achieved at considerable additional cost, especially if you have to rely on battery or rechargeable battery operation.
Once the communication line is in place, care must be taken in advance to ensure that a suitable AI has been selected for the use case that also knows how to handle the data.
An AI that can recognise objects in a camera image will not be able to reasonably interpret a temperature measurement of a sensor consisting of a text message and vice versa. I also like to talk about artificial island intelligence here. No doubt AI systems have their advantages, but they have limited intelligence and their results will always depend on a human mentor as well as the quality of data they come into contact with. They are specialists in recognising objects on a construction site, deriving geometries from point clouds or creating operational profiles of construction vehicles, but all because a human has written a very well-defined brief in a way that machines can understand. And they don’t lift a finger beyond this order. Unless they are taught to.
Assuming you have now decided on a suitable AI system, the legitimate question that AI could well ask us now is:
- How good is the data that is to be collected and analysed on the construction site?
- Does a human collect and generate the data?
- Are several humans involved in the data acquisition?
- Do they know what the required data should look like?
- Does the data come from a sensor?
- Was the temperature sensor placed in the sun on a black background?
- How much dirt is in front of or on the camera lens?
- Is the camera in focus at all?
- Does the camera system fit the application?
- Can the AI read the format at all?
Data will only ever be as good as the recording quality allows. Of course there are methods to detect outliers or to refine the data. But systematic errors will strongly influence the results of AI in the long run. Neural networks & Co are only a small part of a sewage treatment plant. As you can see, the decision alone is not enough to install an AI system on a construction site. For the AI to deliver the hoped-for added value, the working conditions must also be right. It won’t complain about working 24/7, but it does have its habits, which have to be dealt with.
I almost mean AIs on the construction sites are the new guest workers from an unknown country like Communistic Nation of Neuronalia or from the Random Forest. You have to get used to each other first. The language is unfamiliar and one has to work on communication to find out what conditions the new workers need. But once these are known or the area of responsibility is clearly defined, it will do its job reliably like a good tool and support the digitisation of the construction site considerably. Once properly set up, it will not be necessary to ask when the AI will start its daily service on the construction site. Holidays, coffee or tea breaks are just as unknown to it. It plays its results continuously and without prompting into the defined interfaces in order to communicate its findings. Here and there, however, it is good to check up on the AI. Perhaps, at a more advanced stage of construction, one or the other data pipeline can be optimised in order to obtain even more data, which, in a later project, can make the AI smarter through new insights and thus improve the quality of predictions or classifications.
Valuable experience has also been gained in which area which AI, which system has collected meaningful data and insights in order to use them even better or improved in the next construction project. This draws a recurring cycle that reflects the core philosophy of artificial intelligence and (rational) humans – learning from experience. We will grow together and complement each other until we can no longer imagine the construction site without AI systems as a helping hand.
Author: Paul Arzberger, Analyst CONTAKT GmbH