Artificially intelligent, naturally efficient!?

Pro­fes­sional Arti­cle

AI – our daily com­pan­ion

Every day, clever voice assis­tants lip-​read all our wishes, sug­gest suit­able prod­ucts to us from web shops thanks to sophis­ti­cated algo­rithms, manoeu­vre our cars into the small­est park­ing spaces by them­selves and our inquiries are answered on var­i­ous web­sites by cour­te­ous chat­bots. The series of such exam­ples, which can be con­tin­ued almost indef­i­nitely, impres­sively shows that Arti­fi­cial Intel­li­gence (AI) has long since become an inte­gral part of our every­day lives. How­ever, we usu­ally asso­ciate the pop­u­lar buzz­word with com­plex machines or human-​like robots and over­look the fact that AI is also found in small appli­ca­tions and tools. This means that we use a vari­ety of AI-​based appli­ca­tions and ser­vices with­out being aware of it.

This raises the ques­tion of what is actu­ally behind the trend topic of Arti­fi­cial Intel­li­gence and what oppor­tu­ni­ties or risks arise from it. Despite the cur­rent hype, the topic is cer­tainly not new. Experts have been dis­cussing AI for more than 60 years. And in the sci­ence fic­tion genre, rel­e­vant visions of the future have an even longer tra­di­tion. For a long time, how­ever, no sig­nif­i­cant progress could be made in this field – mainly due to tech­ni­cal lim­i­ta­tions. In the recent past, how­ever, the com­put­ing power of our com­put­ers has increased con­sid­er­ably. Not least because of this, a deci­sive step has been taken in the devel­op­ment of AI and related appli­ca­tions in recent years.

The human brain as role model

Cur­rently, Arti­fi­cial Intel­li­gence is – once again – on everyone’s lips and the road map seems clear: AI will change our daily rou­tine, our lives, our work­ing world and soci­ety. But not all con­tem­po­raries wel­come this devel­op­ment. Dif­fuse ideas about what AI can and can­not do, or whether it is an oppor­tu­nity or a threat, lead to uncer­tain­ties and eth­i­cal con­cerns. Thus, the first ques­tion to be clar­i­fied is how Arti­fi­cial Intel­li­gence is defined at all. Until today, there is no gen­er­ally valid def­i­n­i­tion. In the broad­est sense, AI is under­stood as the automa­tion of intel­li­gent behav­iour and machine learn­ing. AI is sup­posed to enable machines to work on tasks that require human-​like intel­li­gence to solve them. The basic idea is there­fore to imi­tate the way the human brain func­tions.

In prin­ci­ple, a dis­tinc­tion can be made between “strong” and “weak” AI. “Weak” AI is what we often encounter – as in the exam­ples men­tioned at the begin­ning – in a sup­port­ive way in every­day life. It is lim­ited to a clearly defined area of appli­ca­tion, for exam­ple the recog­ni­tion and pro­cess­ing of speech. “Strong” AI, in con­trast, has intel­lec­tual abil­i­ties that are at least equal to those of the human brain. Its con­clu­sions should not be lim­ited to one area but should be trans­fer­able to any num­ber of other areas. Until today, it has not been pos­si­ble to develop a “strong” AI accord­ing to this under­stand­ing.

AI – but how?

A large part of the (“weak”) arti­fi­cially intel­li­gent sys­tems, assis­tants and tools we use every day are based on the prin­ci­ple of machine learn­ing. The aim of this extremely suc­cess­ful sub-​area of AI is to iden­tify con­nec­tions or pat­terns through the intel­li­gent link­ing or sta­tis­ti­cal analy­sis of data in order to draw con­clu­sions or make fore­casts. For this pur­pose, the learn­ing machine needs a suf­fi­ciently large data­base. In addi­tion, it must be enabled to process this data by pro­gram­ming spe­cial algo­rithms. Thus, human action is nec­es­sary before the machine can develop its arti­fi­cial intel­li­gence.

Machine learn­ing has become par­tic­u­larly pop­u­lar in appli­ca­tion sce­nar­ios in which huge amounts of data are to be sta­tis­ti­cally eval­u­ated and searched for pat­terns. AI-​based assis­tants, spam fil­ters, lan­guage trans­la­tions, weather fore­casts, autonomous dri­ving sys­tems, med­ical diag­nos­tics and invest­ment advice, for exam­ple, work accord­ing to this prin­ci­ple.

The so-​called Deep Learn­ing, con­versely, is a sub­field or a fur­ther devel­op­ment of machine learn­ing and could become the basis of “strong” AI in the future. It is based on arti­fi­cial neural net­works whose struc­ture and func­tion is mod­elled on the neu­rons of the human brain. With a large infor­ma­tion base and the struc­ture of the neu­ronal net­works, a sys­tem based on deep learn­ing should be able to link learned con­tent with new con­tent inde­pen­dently and in the long term also across depart­ments. In this way, new learn­ing processes would be con­stantly ini­ti­ated. The aim is that the machine does not sim­ply learn, but that it learns to learn. The more mean­ing­ful data such sys­tems process, the more reli­able they become. In the ideal case, the “learn­ing effect” there­fore leads to con­tin­u­ous self-​optimization.

Machines becom­ing a threat?

Con­ven­tional machine learn­ing requires human inter­ven­tion to clas­sify con­clu­sions as “right” or “wrong” and make appro­pri­ate adjust­ments. Deep learn­ing algo­rithms, on the other hand, are sup­posed to per­form this step them­selves. In con­trast to clas­si­cal machine learn­ing, it is hardly pos­si­ble to under­stand how deep learn­ing sys­tems achieve their results. Their higher pre­ci­sion in pat­tern recog­ni­tion is there­fore at the expense of method­olog­i­cal trans­parency. There is a risk that deep learn­ing sys­tems draw wrong con­clu­sions with­out rec­og­niz­ing or cor­rect­ing their own errors.

At the lat­est since the Hol­ly­wood clas­sic “Ter­mi­na­tor” (1984) showed in apoc­a­lyp­tic images what could hap­pen when machines were given too much power, many con­tem­po­raries may have approached the sub­ject of AI with a cer­tain scep­ti­cism. In the film, the machines of a tech­nol­ogy cor­po­ra­tion become inde­pen­dent and have noth­ing less in mind than to seize world dom­i­na­tion. In real­ity, how­ever, such a threat sce­nario hardly seems con­ceiv­able. Although machines are sup­posed to be able to per­form tasks through AI for which human-​like intel­li­gence per­for­mance is assumed, this does not lead to their devel­op­ing their own con­scious­ness. It is also debat­able whether machines can ever develop some­thing like intu­ition or spon­ta­neous cre­ativ­ity. A deci­sive limit would thus remain.

How can AI be used sen­si­bly?

On the basis of the exam­ples men­tioned at the begin­ning, a few prac­ti­cal fields of appli­ca­tion for arti­fi­cial intel­li­gence have already been out­lined. The spread of AI-​based appli­ca­tions affects numer­ous areas of life. While pri­vate users are prob­a­bly mainly happy to receive gad­gets that make their every­day lives eas­ier, doc­tors can use AI to make more reli­able diag­noses and politi­cians can obtain com­pre­hen­sive analy­ses of com­plex rela­tion­ships in no time at all. Exten­sive sta­tis­tics and data vol­umes can be eval­u­ated in a very short time. For exam­ple, AI could be used to make more pre­cise fore­casts of key eco­nomic fig­ures, such as gross domes­tic prod­uct, unem­ploy­ment fig­ures or the infla­tion rate. In addi­tion, intel­li­gent algo­rithms using micro­tar­get­ing enable tar­geted com­mu­ni­ca­tion with spe­cific groups of vot­ers.

Com­pa­nies can also ben­e­fit from the tar­geted use of AI. Although AI-​based tech­nolo­gies have so far only been used spo­rad­i­cally in the econ­omy, cur­rent stud­ies indi­cate that they are set to become mas­sively more impor­tant in the com­ing years. In the dia­logue with cus­tomers, the advan­tages of imple­ment­ing AI for com­pa­nies can already be seen today. For exam­ple, the intel­li­gent rout­ing of processes in the Con­tact Cen­ter ensures a fair dis­tri­b­u­tion of the work­load. Or through chat­bots, which cur­rently rep­re­sent one of the largest areas of appli­ca­tion for AI in com­pa­nies and – just like voice por­tals – lead to a sig­nif­i­cant improve­ment in the ser­vice offer­ing with man­age­able invest­ments. How­ever, it is impor­tant to ensure that processes are well thought out and cus­tomers are not fright­ened off by imma­ture appli­ca­tions. It is there­fore rec­om­mended to imple­ment AI solu­tions in coop­er­a­tion with expe­ri­enced experts, for exam­ple a spe­cial­ized ser­vice provider.

The con­cern that peo­ple will be replaced by a bot in the course of automat­ing the cus­tomer dia­logue can be quickly dis­pelled. In the cus­tomer dia­logue, the tech­nol­ogy aims to pro­vide the best pos­si­ble sup­port for ser­vice staff. For exam­ple, by allow­ing bots to take over time-​consuming rou­tine work, employ­ees can focus on value-​adding activ­i­ties. Arti­fi­cial Intel­li­gence in cus­tomer ser­vice, for exam­ple, only unfolds its full effect through the com­bi­na­tion of state-​of-​the-​art tech­nol­ogy and human skills. Instead of a clash, the path to the future leads through the inter­ac­tion of man and machine.