Harnessing the potential of AI for housing renovation

Arti­fi­cial intel­li­gence (AI) is increas­ingly used in renov­a­tion pro­jects. As such, it reminds us that renov­a­tion is an area of innov­a­tion for which we can dis­tin­guish four sep­ar­ate areas. 

The first is to clas­si­fy to pri­or­it­ise. AI is integ­rated into large-mesh dia­gnost­ic tools. For example, it will make it pos­sible to identi­fy – based on energy bills in rela­tion to their size – the 1,000 or 10,000 pub­lic build­ings out of 500,000, which will have the greatest impact per euro inves­ted in terms of redu­cing CO2 emis­sions or energy consumption. 

The second area is the cre­ation of optim­ised renov­a­tion plans, upstream of a build­ing site: they make it pos­sible to improve the lay­out of a flat, the renov­a­tion of a build­ing or the renew­al of an urb­an area. The third area, at the moment of imple­ment­a­tion, is the census. Com­puter visu­al­isa­tion tech­niques, which rely on images from vari­ous sources (scans, drones etc.), make it pos­sible to take stock of everything on a site, but also to determ­ine the per­cent­age of pro­gress or to identi­fy risks.

The fourth area is everything that hap­pens when the build­ing exists: pre­dict­ive main­ten­ance and intel­li­gent man­age­ment. Based on vari­ous sensors, AI makes it pos­sible to anti­cip­ate break­downs or to bet­ter under­stand the pref­er­ences of the build­ing’s users. Some smart build­ing sys­tems go fur­ther by allow­ing inter­ac­tion between users (or inhab­it­ants) of the build­ing with con­ver­sa­tion­al agents, or by mak­ing pre­dic­tions about the needs of res­id­ents and thus gen­er­at­ing pro­pos­als for per­son­al­ised services.

In terms of renov­a­tion, it is cur­rently the second area, design, that is see­ing the most prom­ising advances. 

At the build­ing level, tech­nic­al stand­ards are mastered by pro­fes­sion­als and the con­tri­bu­tion of AI is not yet major. How­ever, it will allow for energy per­form­ance, envir­on­ment­al impact, or cli­mate resi­li­ence sim­u­la­tions accord­ing to vari­ous object­ives and con­straints. This will con­trib­ute to optim­ising renov­a­tion solu­tions from an eco­nom­ic and envir­on­ment­al point of view. In a chan­ging envir­on­ment, where energy and mater­i­al prices are evolving rap­idly as they are now, the use of sim­u­la­tion allows us to reopen the field of pos­sib­il­it­ies and to find ways bey­ond pro­fes­sion­als’ cur­rent habits. More gen­er­ally, auto­ma­tion also allows work to be car­ried out more quickly.

On a neigh­bour­hood scale (more rarely on the scale of an entire city), where things become more com­plic­ated, AI plays a more import­ant role. Both for ques­tions of resources: the indus­tri­al trades involved are prac­tised by power­ful eco­nom­ic play­ers, often at the fore­front of tech­no­logy; and for ques­tions of needs: it is mainly at this scale that new vari­ables appear, which were not pre­vi­ously integ­rated and have no obvi­ous “tech­nic­al solu­tions”, trans­posed into practices.

These issues include the envir­on­ment­al impact of renov­a­tion activ­it­ies and the resi­li­ence and adapt­a­tion of infra­struc­tures to cli­mate change. They are added to the more tra­di­tion­al ones that have gov­erned decision-mak­ing until now: cost of the work, com­fort, energy con­sump­tion in the eco­nom­ic sense (pur­chas­ing power of house­holds, costs for man­agers), energy con­sump­tion in the eco­lo­gic­al sense (emis­sions). A renov­a­tion plan is now a very com­plex matter!

Basic­ally, we are in the busi­ness of decision sup­port. Let’s take an example. In France and in Europe, many renov­a­tion oper­a­tions con­cern neigh­bour­hoods or build­ings belong­ing to the ‘social’ stock, built for the most part in the thirty years fol­low­ing the end of the last World War. These are there­fore col­lect­ive decisions, involving pub­lic act­ors and, of course, pub­lic money. The pro­cesses are cum­ber­some, the decision-makers numer­ous and the stakes high, because the phys­ic­al deteri­or­a­tion of these neigh­bour­hoods goes hand in hand with social and some­times polit­ic­al prob­lems. Moreover, they are most often heat loss areas, and the poor qual­ity of the build­ings exposes the inhab­it­ants to energy shocks (increase in price, dis­com­fort). Finally, the design of the neigh­bour­hoods and build­ings makes them vul­ner­able to the con­sequences of cli­mate change, par­tic­u­larly heat waves and flood­ing. There are car parks in these neigh­bour­hoods that are heat pock­ets, and those on the banks of rivers may be sub­ject to a flood risk that was neg­li­gible when they were built.

Renov­a­tion is both an urgent and com­plex sub­ject, which raises new issues that are not easy for pub­lic decision-makers to master.

To sum up, renov­a­tion is both an urgent and com­plex sub­ject, which raises new issues that are not easy for pub­lic decision-makers to mas­ter. One of the essen­tial chal­lenges is there­fore to accel­er­ate and optim­ise the col­lect­ive decision. This is pre­cisely what AI allows.

How can it do so? By allow­ing, dur­ing the early stages of the dis­cus­sion, us to auto­mat­ic­ally gen­er­ate optim­ised rep­res­ent­a­tions, plans for example, which are not “the” solu­tion, but which allow the dis­cus­sion to pro­gress. These rep­res­ent­a­tions are what is known in man­age­ment sci­ence as “bound­ary objects”: they are suf­fi­ciently con­crete to con­sti­tute an inter­face between social worlds and act­ors with dif­fer­ent perspectives.

In a decision-mak­ing pro­cess that lasts a dozen months, if we take a worst-case scen­ario, AI could save one or two months. This is sig­ni­fic­ant, without being a rad­ic­al upheav­al. We can also assume that the qual­ity of the decision will be bet­ter: it is made in a way that allows every­one to under­stand the reas­on­ing of oth­ers, and on the basis of rep­res­ent­a­tions that are easi­er to dis­cuss than columns of fig­ures. AI, by mak­ing it easy to gen­er­ate pro­pos­als, is a facil­it­at­or and optim­iser of human decision-making.

Not in the sense that a com­puter would speak at the table… but a dis­cus­sion dimen­sion can be integ­rated into the design work that involves AI. The area that interests us in the Ren­ov­AIte R&D pro­gramme on which I am work­ing is what is called Adversori­al Resi­li­ence Learn­ing (ARL), a concept that enables arti­fi­cial neur­al net­works to be mod­elled and trained by put­ting them in com­pet­it­ive situ­ations. We take two vir­tu­al entit­ies, which will lit­er­ally throw them­selves against each oth­er. One will play the role of defend­er; it will be a mod­el of the urb­an com­plex based on BIM (Build­ing Inform­a­tion Mod­el­ing: digit­al mod­els that do not rep­res­ent geo­met­ric shapes, but objects) and oth­er known ele­ments. The oth­er will play the role of the attack­er: this can be any crisis (cold, heat, flood­ing, with risks on com­fort, on heat­ing sys­tems, risks of cracks or impreg­na­tion, budget­ary con­straints, etc.). After sev­er­al hours, this com­pet­i­tion will allow each of the entit­ies to devel­op the best strategies and come up with a set of optim­ised urb­an renew­al plan pro­pos­als for decision-makers. 

It is already work­ing: this concept has been used in Ger­many in the design of smart grids, to identi­fy mal­func­tions on the net­work and gen­er­ate the appro­pri­ate auto­mat­ic responses.

Interview by Richard Robert