Are artificial intelligence and human intelligence comparable?

Arti­fi­cial intel­li­gence (AI) is dis­rupt­ing the world as we know it. It is per­meat­ing every part of our lives, with more or less desir­able and ambi­tious goals. Inev­it­ably, AI and human intel­li­gence (HI) are being com­pared. Far from com­ing out of nowhere, this con­front­a­tion can be explained by his­tor­ic­al dynam­ics inscribed deep with­in the AI project.

AI and HI, as fields of study, have co-evolved. There have been two dis­tinct approaches since the early days of mod­ern com­put­ing: evol­u­tion by par­al­lel­ism or by dis­reg­ard. “The founders of AI were divided into two approaches. On the one hand, those who wanted to ana­lyse human men­tal pro­cesses and repro­duce them on a com­puter, in a mir­ror image, so that the two under­tak­ings would feed off each oth­er. On the oth­er, those who saw HI as a lim­it­a­tion rather than an inspir­a­tion. This trend was inter­ested in prob­lem solv­ing, in oth­er words in the res­ult and not the pro­cess”, recalls Daniel Andler.

Our tend­ency to com­pare AI and HI in numer­ous pub­lic­a­tions is there­fore not a cur­rent trend, but part of the his­tory of AI. What is symp­to­mat­ic of our times is the tend­ency to equate the entire digit­al world with AI: “Today, all com­put­ing is described as AI. You have to go back to the found­a­tions of the dis­cip­line to under­stand that AI is a spe­cif­ic tool, defined by the cal­cu­la­tion that is being made and the nature of the task it is solv­ing. If the task seems to involve human skills, we will be look­ing at the capa­city for intel­li­gence. That, in essence, is what AI is all about”, explains Maxime Amblard.

The two major trends men­tioned above have giv­en rise to two major cat­egor­ies of AI:

• sym­bol­ic AI, based on logic­al infer­ence rules, which has little to do with human cognition
• con­nec­tion­ist AI, based on neur­al net­works, which is inspired by human cognition.

Maxime Amblard takes us back to the con­text of the time: “In the middle of the 20th cen­tury, the com­put­ing capa­city of com­puters was tiny com­pared with today. So, we thought that to have intel­li­gent sys­tems, the cal­cu­la­tion would have to con­tain expert inform­a­tion that we had pre­vi­ously encoded in the form of rules and sym­bols. At the same time, oth­er research­ers were more inter­ested in how expert­ise could be gen­er­ated. The ques­tion then became: how can we con­struct a prob­ab­il­ity dis­tri­bu­tion that will provide a good explan­a­tion of how the world works? It’s easy to see why these approaches exploded when the avail­ab­il­ity of data and memory and com­put­ing capa­city increased radically”.

To illus­trate the his­tor­ic­al devel­op­ment of these two branches, Maxime Amblard uses the meta­phor of two skis advan­cing one after the oth­er: “Before com­put­ing power became avail­able, prob­ab­il­ist­ic mod­els were ignored in favour of sym­bol­ic mod­els. We are cur­rently exper­i­en­cing a peak in con­nec­tion­ist AI thanks to its revolu­tion­ary res­ults. Nev­er­the­less, the prob­lem of mak­ing the res­ults com­pre­hens­ible leaves the way open for hybrid sys­tems (con­nec­tion­ist and sym­bol­ic) to put know­ledge back into clas­sic prob­ab­il­ist­ic approaches”.

For her part, Anna­belle Blan­gero points out that today “there is some debate as to wheth­er expert sys­tems really cor­res­pond to AI, giv­en that there is a tend­ency to describe sys­tems that neces­sar­ily involve machine learn­ing as AI”. Nev­er­the­less, Daniel And­ler men­tions one of the lead­ing fig­ures in AI, Stu­art Rus­sell, who remains very attached to sym­bol­ic AI. Maxime Amblard also agrees: “Per­haps my vis­ion is too influ­enced by the his­tory and epi­stem­o­logy of AI, but I think that to describe some­thing as intel­li­gent, it is more import­ant to ask how what is pro­duced by the com­pu­ta­tion is able to change the world, rather than focus­ing on the nature of the tool used.”

After the his­tor­ic­al and defin­i­tion­al diver­sions, the fol­low­ing ques­tion arises: are AI and HI two sides of the same coin? Before we can come up with an answer, we need to look at the meth­od­o­lo­gic­al frame­work that makes this com­par­is­on pos­sible. For Daniel And­ler, “func­tion­al­ism is the frame­work par excel­lence with­in which the ques­tion of com­par­is­on arises, provided that we call ‘intel­li­gence’ the com­bined res­ult of cog­nit­ive func­tions”. How­ever, some­thing is almost cer­tainly miss­ing if we are to get as close as pos­sible to human intel­li­gence, situ­ated in time and space. “His­tor­ic­ally, it was John Hauge­land who developed this idea of a miss­ing ingredi­ent in AI. We often think of con­scious­ness, inten­tion­al­ity, autonomy, emo­tions or even the body”, Daniel And­ler explains.

Con­scious­ness and the asso­ci­ated men­tal states seem to be miss­ing from AI. For Anna­belle Blan­gero, this miss­ing ingredi­ent is simply a ques­tion of tech­nic­al means: “I come from a school of thought in neur­os­cience where we con­sider that con­scious­ness emerges from the con­stant eval­u­ation of the envir­on­ment and asso­ci­ated sens­ory-motor reac­tions. Based on this prin­ciple, repro­du­cing human mul­timod­al­ity in a robot should bring out the same char­ac­ter­ist­ics. Today, the archi­tec­ture of con­nec­tion­ist sys­tems repro­duces fairly closely what hap­pens in the human brain. What’s more, sim­il­ar meas­ures of activ­ity are used in bio­lo­gic­al and arti­fi­cial neur­al networks”.

Nev­er­the­less, as Daniel And­ler points out, “Today, there is no single the­ory to account for con­scious­ness in humans. The ques­tion of its emer­gence is wide open and the sub­ject of much debate in the sci­entif­ic-philo­soph­ic­al com­munity”. For Maxime Amblard, the fun­da­ment­al dif­fer­ence lies in the desire to make sense. “Humans con­struct explan­at­ory mod­els for what they per­ceive. We are ver­it­able mean­ing-mak­ing machines.”

Des­pite this well-argued devel­op­ment, the ques­tion of bring­ing AI and HI togeth­er remains unanswered. In fact, the prob­lem is primar­ily con­cep­tu­al and con­cerns the way in which we define intelligence.

A clas­sic defin­i­tion would describe intel­li­gence as the set of abil­it­ies that enable us to solve prob­lems. In his recent book, Intel­li­gence arti­fi­ci­elle, intel­li­gence humaine: la double énigme, Daniel And­ler pro­poses an altern­at­ive, eleg­ant defin­i­tion: “anim­als (human or non-human) deploy the abil­ity to adapt to situ­ations. They learn to solve prob­lems that con­cern them, in time and space. They couldn’t care less about solv­ing gen­er­al, decon­tex­tu­al­ised problems”.

This defin­i­tion, which is open to debate, has the mer­it of pla­cing intel­li­gence in con­text and not mak­ing it an invari­ant concept. The math­em­atician and philo­soph­er also reminds us of the nature of the concept of intel­li­gence. “Intel­li­gence is what we call a dense concept: it is both descript­ive and object­ive, appre­ci­at­ive and sub­ject­ive. Although in prac­tice we can quickly come to a con­clu­sion about a person’s intel­li­gence in a giv­en situ­ation, in prin­ciple it’s always open to debate”.

In the end, the issue of com­par­is­on seems irrel­ev­ant if we are look­ing for a con­crete answer. It is of great­er interest if we seek to under­stand the intel­lec­tu­al path we have trav­elled, the pro­cess. This reflec­tion high­lights some cru­cial ques­tions: what do we want to give AI? To what end? What do we want for the future of our societies?

These are essen­tial ques­tions that revive the eth­ic­al, eco­nom­ic, legis­lat­ive, and social chal­lenges that need to be taken up by the play­ers in the world of AI and by gov­ern­ments and cit­izens the world over. At the end of the day, there is no point in know­ing wheth­er AI is or will be like us. The only import­ant ques­tion is what do we want to do with it and why?

Julien Hernandez