06.-Intelligence artificielle / Artificial intelligence — CHURCH OF FRANCE / États généraux de la bioéthique — States General on Bioethics

http://eglise.catholique.fr/sengager-dans-la-societe/eglise-et-bioethique/comprendre-les-enjeux/

Télécharg­er (PDF, 287KB)

Scientific and legal data

The lat­est report[1] on “Arti­fi­cial Intel­li­gence” (AI) of the Com­mis­sion Nationale Infor­ma­tique et Lib­ertés (CNIL) of Decem­ber 2017 under­lines that AI is the “great myth of our time” and that its def­i­n­i­tion remains very impre­cise. When we talk about AI, we think of tar­get­ed adver­tise­ments on the Inter­net, con­nect­ed objects and the Inter­net of Things, mas­sive and het­ero­ge­neous data pro­cess­ing out­put from a sur­vey, dig­i­tal machines and humanoid robots capa­ble of learn­ing and evo­lu­tion, autonomous vehi­cles, brain-machine inter­faces and algorithms.

Algo­rithms are in a way the “skele­tons of our com­put­er tools”. They are instruc­tion sys­tems that allow the dig­i­tal machine to pro­vide results from the data pro­vid­ed. They are at work when using an Inter­net search engine, or when propos­ing a med­ical diag­no­sis based on sta­tis­ti­cal data, but also when choos­ing a car route and select­ing infor­ma­tion on social net­works accord­ing to the tastes of friends’ net­works. Algo­rithms belong to the design­er and very often remain unknown to users. They become capa­ble of more and more com­plex tasks thanks to the expo­nen­tial­ly increas­ing com­put­ing pow­er and to auto­mat­ic learn­ing tech­niques (auto­mat­ic adjust­ment of the para­me­ters of an algo­rithm so that it pro­duces the expect­ed results from the data pro­vid­ed). In this way, “deep learn­ing” has met with many suc­cess­es. It already makes it pos­si­ble to rec­og­nize images and objects, to iden­ti­fy a face, to pilot an intel­li­gent robot…

The links between neu­ro­science and AI are the basis of the Euro­pean Human Brain Project, one of whose objec­tives is to sim­u­late the behav­iour of the human brain. AI can also be used to bet­ter under­stand neu­ronal dis­eases such as com­pul­sive dis­or­ders or depres­sion. It is there­fore a ques­tion of build­ing so-called intel­li­gent machines both to dri­ve evo­lu­tion­ary sys­tems and to par­tic­i­pate in the under­stand­ing of the human brain.

The Data Pro­tec­tion Act of 6 Jan­u­ary 1978, amend­ed reg­u­lar­ly since, states that “data pro­cess­ing must be at the ser­vice of every cit­i­zen… must not infringe human iden­ti­ty, human rights, pri­va­cy or indi­vid­ual or pub­lic free­doms”. In par­tic­u­lar, it defines the prin­ci­ples to be respect­ed for the col­lec­tion, pro­cess­ing and stor­age of per­son­al data. It spec­i­fies the pow­ers and sanc­tion­ing capac­i­ties of the CNIL. The new Euro­pean reg­u­la­tion on the pro­tec­tion of per­son­al data (RGDP), adopt­ed on 27 April 2016, takes effect on 25 May 2018 in the Mem­ber States of the Euro­pean Union to strength­en legal reg­u­la­tion. That is why a bill (No. 490) was intro­duced in the Nation­al Assem­bly on 13 Decem­ber 2017.

For many, the AI is a tremen­dous oppor­tu­ni­ty in terms of the knowl­edge econ­o­my. Its con­tri­bu­tions in the fields of med­i­cine, robot­ics, learn­ing and sci­ence in par­tic­u­lar are already con­sid­er­able. But how to tame the AI so that it is tru­ly at the ser­vice of all?

Questions this raises

Among the fears and risks most often expressed are the prob­lems of job cuts with robots. There is also dis­trust, even a “loss of human­i­ty” in the face of the “black box” rep­re­sent­ed by the algo­rithms “that gov­ern us”, on the Inter­net, on social net­works, in e‑commerce and even in our pri­vate lives. But these algo­rithms could also gov­ern the doc­tor and the employ­er who would rely on them to make their deci­sions. Who con­trols what? “This is the ques­tion often raised. Thus one won­ders what are the “bias­es” by which judg­ments are made to recruit an employ­ee through AI, with sus­pi­cion of discrimination.

In addi­tion to the pro­tec­tion of per­son­al data, the main ques­tions of the CNIL report are:

• Faced with the pow­er of machines, how can we appre­hend the forms of dilu­tion of respon­si­bil­i­ties in decisions
• How far can we accept the “auton­o­my of the machines” that can decide for us?
• How to deal with the lack of trans­paren­cy of algo­rithms as to the bias­es they use to process data and “decide the results”?
• How to appre­hend this new class of objects that are humanoid robots like­ly to drive
  What sta­tus should so-called intel­li­gent robots be giv­en, and what legal con­se­quences in terms of lia­bil­i­ty in the event of a problem?

Faced with the risks of a pos­si­ble form of more or less invis­i­ble “dic­ta­tor­ship of dig­i­tal tech­nol­o­gy”, the CNIL report argues for two found­ing prin­ci­ples for the ethics of AI:

• col­lec­tive loy­al­ty (for trans­paren­cy and demo­c­ra­t­ic use of algo­rithms for example) ;
• vigilance/reflexivity with regard to the auton­o­my of machines and the bias­es they prop­a­gate or gen­er­ate, so that man does not “lose con­trol” over AI.

Anthropological and Ethical Aims

To ask the ques­tion of the sta­tus of robots is sig­ni­fy­ing a “dis­or­der” intro­duced by AI con­cern­ing the rela­tion­ship that man main­tains with his “learn­ing machines”. A Euro­pean Par­lia­ment res­o­lu­tion encour­ages research on grant­i­ng “elec­tron­ic per­son” sta­tus to cer­tain robots[2].

This legal expres­sion would rel­a­tivize the notion of per­son, which is root­ed in the dig­ni­ty of the human being[3]. The term “cog­ni­tive robot”, for exam­ple, would be prefer­able. How­ev­er, how close can the capa­bil­i­ties of machines be to those of humans, and then exceed them? Some extreme tran­shu­man­ists await this moment when AI will sur­pass human intel­li­gence, a sort of “sin­gu­lar­i­ty” from which a man-machine fusion will con­sti­tute a “cyborg” that will take over from homo sapiens!

With­out enter­ing into such fan­tasies, celebri­ties like Stephen Hawk­ing, Bill Gates and Elon Musk have repeat­ed­ly expressed con­cerns about AI[4]. They express their fear that learn­ing machines will con­trol us, because they will have sta­tis­ti­cal and com­bi­na­to­r­i­al skills far supe­ri­or to ours, as well as access to gigan­tic data­bas­es that man can­not process direct­ly. It is essen­tial­ly on the com­put­ing aspect that the pow­er of dig­i­tal machines is appre­hend­ed today. Only this form of intel­li­gence is at stake, where­as man has many forms of intel­li­gence (ratio­nal, emo­tion­al, artis­tic, rela­tion­al, etc.). Of course, we under­stand that the pow­er­ful cal­cu­la­tors allow the machine to find the com­bi­na­tions to beat the cham­pi­ons of the game of Go. How­ev­er, the AI is now in the sim­u­la­tion field. How­ev­er, there is a thresh­old between “sim­u­lat­ing” an emo­tion and expe­ri­enc­ing it. Emo­tion, with its com­mu­nica­tive dimen­sion, leads the man who expe­ri­ences it to attribute a val­ue to things from which he makes choic­es in dai­ly life. Emo­tion express­es the wealth of the vul­ner­a­ble man. The learn­ing machine isn’t there yet! Does AI human­ize? It is indeed a “pow­er” that must be sub­ject­ed to dis­cern­ment in the face of fragili­ty and vul­ner­a­bil­i­ty as sources of human­iza­tion. Sim­i­lar­ly, it is impos­si­ble to com­pare human con­scious­ness (exis­ten­tial, psy­cho­log­i­cal and moral) with a pos­si­ble machine consciousness[5].

When it comes to AI, the idea that “think­ing is cal­cu­lat­ing” is often preva­lent. This leads to a lot of con­fu­sion. Man, endowed with an intel­li­gence made for truth and wis­dom, has anoth­er reg­is­ter of thought much more var­ied, vast and subtle[6]. Our con­scious­ness is sit­u­at­ed in a body shaped by mil­lions of years of evo­lu­tion, with beau­ti­ful capac­i­ties of rea­son, cre­ation, psy­chic life and spir­i­tu­al depth, which go far beyond the most sophis­ti­cat­ed combinatories.

Some point out that instead of point­ing out the risks of com­pu­ta­tion­al and com­bi­na­to­r­i­al intel­li­gence of machines, it is more urgent to make pub­lic the val­ues that algo­rithm design­ers intro­duce into their soft­ware. The trans­paren­cy of algo­rithms is a real issue of sub­stance. Does their design always aim to improve the care and ser­vice of human dignity?

BIBLIOGRAPHICAL REFERENCES TO CONTINUE THE WORK

IA, Promis­es and Per­il, Cahi­er du Monde n. 22696, 31 Decem­ber 2017–2 Jan­u­ary 2018.

Frédéric Alexan­dre and Serge Tis­seron, “Where are the real dan­gers of AI? “in Les robots en quête d’hu­man­ité, Pour la Sci­ence, n. 87, April-June 2015, p.102–107.

Milad Douei­hi and Frédéric Louzeau, Du matéri­al­isme numérique, Her­mann, 2017.

Serge Abite­boul and Gilles Dowek, Le temps des algo­rithmes, Le Pom­mi­er, 2017.

Feb­ru­ary 2, 2018

———————————————————-

[1] Com­ment per­me­t­tre à l’homme de garder la main ?, Rap­port de la CNIL, pub­lié le 15 décem­bre 2017.

[2] http://www.europarl.europa.eu/sides/getDoc.do?pubRef=-//EP//NONSGML+TA+P8-TA-2017–0051+0+DOC+PDF+V0//FR

[3] La notion de « per­son­ne morale » ne per­met pas d’ambiguïté. Par ailleurs, elle n’est pas recon­nue par toutes les tra­di­tions juridiques.

[4] Voir Alexan­dre Picard, « L’intelligence arti­fi­cielle, star inquié­tante du web sum­mit à Lis­bonne », Le monde économie, 10 novem­bre 1017.

[5] Voir par exem­ple Meh­di Khamas­si et Raja Chati­la, « La con­science d’une machine », in Les robots en quête d’humanité, Pour la Sci­ence, n° 87, avril-juin 2015. Cf. Vat­i­can II, con­sti­tu­tion Gaudi­um et spes, n. 16 ; Déc­la­ra­tion Dig­ni­tatis humanae, n. 1–3 ; Jean-Paul II, ency­clique La splen­deur de la vérité, 6 août 1993, n. 31–34.

[6] Cf. Vat­i­can II, con­sti­tu­tion Gaudi­um et spes, n. 15 ; Jean-Paul II, ency­clique Foi et rai­son, 14 sep­tem­bre 1998, n. 16–33.