Armed forces are turning to generic technologies, like AI 

Eight months since the begin­ning of the con­flict in Ukraine, many lessons have been learned from this con­flict, notably the return to high inten­si­ty. One of the major issues aris­ing from this, which goes beyond the mil­i­tary field, is the ques­tion of inte­grat­ing new tech­nolo­gies to pro­vide an oper­a­tional advan­tage for the armed forces. 

Part­ner­ships, sup­port­ed by the US Depart­ment of Defense, with Microsoft or Ama­zon Web Ser­vices (AWS) to host Ukrain­ian data that could be sen­si­tive to Russ­ian cyber-attacks, or the use of Elon Musk’s Star­link mobile ter­mi­nals to con­nect to the Inter­net via satel­lite, have high­light­ed the impor­tance of the rapid adop­tion of gen­er­al-pur­pose tech­nolo­gies for mil­i­tary needs, and their effec­tive­ness on the battlefield.

In dri­ving the reform of its defence inno­va­tion ecosys­tem, the French Min­istry of Defence has, in recent years, tak­en on a more entre­pre­neur­ial and strate­gic role, with a more exper­i­men­tal approach to inno­va­tion pol­i­cy¹. Nev­er­the­less, the ques­tion remains as to the capac­i­ty of our defence indus­try and our armed forces to be able to rapid­ly adopt all these gen­er­al-pur­pose tech­nolo­gies (GPTs) – such as arti­fi­cial intel­li­gence or quan­tum tech­nolo­gies, at very dif­fer­ent lev­els of development.

Gen­er­al-pur­pose tech­nolo­gies can be defined as those capa­ble of con­tin­u­ous future improve­ments and with the poten­tial to serve as the basis for com­ple­men­tary inno­va­tions in relat­ed appli­ca­tion areas (Teece, 2018). As such, a gen­er­al-pur­pose tech­nol­o­gy has such an impact on over­all pro­duc­tiv­i­ty growth that it becomes ubiq­ui­tous, as was the case with the inter­net in the 2000s.

With more and more prac­ti­cal appli­ca­tions in all sec­tors fol­low­ing sig­nif­i­cant work in basic research, arti­fi­cial intel­li­gence (AI) has the poten­tial to become a gen­er­al-pur­pose tech­nol­o­gy and thus trans­form the econ­o­my, nation­al secu­ri­ty, and soci­ety as a whole.

Armed forces around the world are increas­ing­ly look­ing to AI as a key tech­nol­o­gy for their long-term strate­gies and plan­ning. With tech giants such as Google and Ama­zon at the fore­front of AI inno­va­tion, major US defence com­pa­nies are under pres­sure to step up their AI activ­i­ties to keep pace with the com­mer­cial market.

100 mil­lion a year in arti­fi­cial intel­li­gence for defence, this rep­re­sents only 0.6% of the French armed forces’ equip­ment bud­get, where­as the Amer­i­cans invest four times as much for an equiv­a­lent bud­get².

Beyond invest­ment, sev­er­al bar­ri­ers to the adop­tion of these gen­er­al-pur­pose tech­nolo­gies per­sist. From a tech­ni­cal point of view, the vul­ner­a­bil­i­ty of the sup­ply of real data to train the algo­rithms (Oso­ba and Welser, 2017) is often the cause of unsat­is­fac­to­ry per­for­mance. How­ev­er, the prob­lem of the con­fi­den­tial­i­ty of mil­i­tary data has often, at least in dis­course, been an argu­ment that can hide anoth­er prob­lem: that of the gov­er­nance of data, which is too often left to the indus­tri­al own­ers of plat­forms and weapon systems.

AI can have a crit­i­cal role in oper­a­tional plan­ning and decision-making.

Copié !

In terms of user aware­ness, there are still oth­er obsta­cles, often linked to the unequal knowl­edge of deci­sion-mak­ers of the uses of data and their added val­ue for mil­i­tary oper­a­tions. This can be explained by the fact that sys­tems inte­grat­ing AI do not all enter the tech­ni­cal reper­toire of the armed forces at the same time and with the same effi­cien­cy, but also by cul­tur­al and organ­i­sa­tion­al bar­ri­ers between the defence sec­tor and the civil­ian sec­tor. A strong dis­in­cen­tive to take risks, as well as pro­cure­ment process­es that are not adapt­ed to this type of tech­nol­o­gy, are often enough to dis­cour­age any ini­tia­tive. Ulti­mate­ly, accord­ing to the fig­ures pre­sent­ed by the Grand Défi focused on arti­fi­cial intel­li­gence, only 10 to 15% of AI-based proofs of con­cept (in the sense of a demon­stra­tor) in France are indus­tri­alised and go to scale. 

AI can process large vol­umes of com­plex data to speed up deci­sion mak­ing at all lev­els of com­mand, act­ing as a force mul­ti­pli­er for com­mand – par­tic­u­lar­ly with the emerg­ing require­ments of mul­ti-domain oper­a­tions. While human resources can cur­rent­ly process, at best, 20% of the infor­ma­tion pro­duced today, this per­cent­age may drop to as lit­tle as 2% in the face of the explo­sion in data pro­duc­tion³.

As a sign that this use case is of major inter­est to the armed forces, the Min­istry of the Armed Forces has just signed a defence and secu­ri­ty con­tract worth €240 mil­lion, over sev­en years, with a non-tra­di­tion­al play­er in the defence indus­try: a pure play­er in defence AI in France. The aim of this con­tract is to accel­er­ate the intel­li­gence cycle and to process the tsuna­mi of data com­ing from these sov­er­eign CSO opti­cal imag­ing satellites. 

In addi­tion to han­dling the expo­nen­tial growth of data, arti­fi­cial intel­li­gence can also improve the qual­i­ty of deci­sion-mak­ing process­es and become a key play­er in the plan­ning and con­duct of oper­a­tions con­tin­u­um. Indeed, the aggre­ga­tion of many data sources, and their analy­sis by machine learn­ing algo­rithms, can help deter­mine the best pos­si­ble geo­graph­ic dis­tri­b­u­tion of forces based on the mis­sion, a unit’s capa­bil­i­ties, con­di­tions in the area of inter­est, resup­ply require­ments and infor­ma­tion derived from the analy­sis of all intel­li­gence sources.

Every the­atre of oper­a­tions offers an infi­nite num­ber of force con­fig­u­ra­tions. To get an idea of the mag­ni­tude of this num­ber, con­sid­er a game of chess. After each play­er has made four moves, there are 988 mil­lion dif­fer­ent pos­si­ble con­fig­u­ra­tions. Com­bin­ing mas­sive sources of rel­e­vant infor­ma­tion (such as loca­tion data, weapon ranges, intel­li­gence, etc.) gives AI a crit­i­cal role in oper­a­tional plan­ning and decision-making.

AI can go beyond human capa­bil­i­ties and dra­mat­i­cal­ly reduce the response times of defence sys­tems to attacks by fast-act­ing weapons sys­tems (hyper­son­ic mis­siles, cyber-attacks or direct­ed ener­gy weapons) to bring about a rad­i­cal change in capa­bil­i­ties. Again, there are non-tra­di­tion­al play­ers in the defence indus­try, such as the US com­pa­ny Anduril, one of the few US start-ups to have suc­cess­ful­ly scaled up. Its algo­rithms on board drones for tar­get­ing pur­pos­es enable the US army to accel­er­ate the deci­sion loop at a tac­ti­cal lev­el and to get clos­er to real time to counter ene­my manoeu­vres much more effectively.

On so-called broad-spec­trum effects, AI can also help to obtain a broad­er knowl­edge and under­stand­ing of the bat­tle­field, and for exam­ple antic­i­pate attempts to manip­u­late cit­i­zens. Indeed, AI-dri­ven analy­sis of the effect on pub­lic sen­ti­ment of a kinet­ic strike on infra­struc­ture, or AI-based solu­tions to detect dis­in­for­ma­tion cam­paigns, are among the most advanced intel­li­gence ser­vice tools for dig­i­tal influ­ence and infor­ma­tion­al warfare. 

Con­crete appli­ca­tions of AI for the armed forces do exist, and not all use cas­es of this “enabling tech­nol­o­gy⁴” have yet emerged at this stage. Non-tra­di­tion­al play­ers in the defence sec­tor stand out as sup­pli­ers of these AI-based solu­tions, thanks to their abil­i­ty to deal with spe­cif­ic “user” needs and to pro­duce demon­stra­tors in “agile” mode based on exper­i­ments. How­ev­er, these play­ers are faced with the more gen­er­al dilem­ma of adopt­ing gen­er­al-pur­pose tech­nolo­gies: to what extent and at what speed do these play­ers suc­ceed in chang­ing the tra­di­tion­al forms of organ­i­sa­tion of the armed forces to enable them to ben­e­fit from the advances pro­vid­ed by gen­er­al-pur­pose technologies?