Zeer inter­es­sant artikel van Ted Chi­ang in The New York­er. Boorde­vol inter­es­sante quotes. Een selec­tie:

 

Art is noto­ri­ous­ly hard to define, and so are the dif­fer­ences between good art and bad art. But let me offer a gen­er­al­iza­tion: art is some­thing that results from mak­ing a lot of choic­es. […] If an A.I. gen­er­ates a ten-thou­sand-word sto­ry based on your prompt, it has to fill in for all of the choic­es that you are not mak­ing. There are var­i­ous ways it can do this. One is to take an aver­age of the choic­es that oth­er writ­ers have made, as rep­re­sent­ed by text found on the Inter­net; that aver­age is equiv­a­lent to the least inter­est­ing choic­es pos­si­ble, which is why A.I.-generated text is often real­ly bland. Anoth­er is to instruct the pro­gram to engage in style mim­ic­ry, emu­lat­ing the choic­es made by a spe­cif­ic writer, which pro­duces a high­ly deriv­a­tive sto­ry. In nei­ther case is it cre­at­ing inter­est­ing art.

 

A dog can com­mu­ni­cate that it is hap­py to see you, and so can a prelin­guis­tic child, even though both lack the capa­bil­i­ty to use words. Chat­G­PT feels noth­ing and desires noth­ing, and this lack of inten­tion is why Chat­G­PT is not actu­al­ly using lan­guage.

 

We are enter­ing an era where some­one might use a large lan­guage mod­el to gen­er­ate a doc­u­ment out of a bul­let­ed list, and send it to a per­son who will use a large lan­guage mod­el to con­dense that doc­u­ment into a bul­let­ed list. Can any­one seri­ous­ly argue that this is an improve­ment?

 

It’s not impos­si­ble that one day we will have com­put­er pro­grams that can do any­thing a human being can do, but, con­trary to the claims of the com­pa­nies pro­mot­ing A.I., that is not some­thing we’ll see in the next few years. Even in domains that have absolute­ly noth­ing to do with cre­ativ­i­ty, cur­rent A.I. pro­grams have pro­found lim­i­ta­tions that give us legit­i­mate rea­sons to ques­tion whether they deserve to be called intel­li­gent at all.

 

The task that gen­er­a­tive A.I. has been most suc­cess­ful at is low­er­ing our expec­ta­tions, both of the things we read and of our­selves when we write any­thing for oth­ers to read. It is a fun­da­men­tal­ly dehu­man­iz­ing tech­nol­o­gy because it treats us as less than what we are: cre­ators and appre­hen­ders of mean­ing. It reduces the amount of inten­tion in the world.

 

We are all prod­ucts of what has come before us, but it’s by liv­ing our lives in inter­ac­tion with oth­ers that we bring mean­ing into the world. That is some­thing that an auto-com­plete algo­rithm can nev­er do, and don’t let any­one tell you oth­er­wise.