AI has been cre­at­ing art since the 1970s: the evo­lu­tion of a para­dox

Gen­er­a­tive AI Mean­ing: Under­stand­ing the Basics

The AI artist can con­tin­u­ous­ly adapt to the pref­er­ences of its col­lec­tors, mod­i­fy­ing the aes­thet­ics of its works based on feed­back from its com­mu­ni­ty of over 5,000 par­tic­i­pants. To ensure gen­er­a­tive AI serves soci­ety with­out under­min­ing cre­ators, we need new legal and eth­i­cal frame­works that address these chal­lenges head-on. Only by evolv­ing beyond tra­di­tion­al fair use can we strike a bal­ance between inno­va­tion and pro­tect­ing the rights of those who fuel cre­ativ­i­ty. The fair use doc­trine was designed for spe­cif­ic, lim­it­ed scenarios—not for the large-scale, auto­mat­ed con­sump­tion of copy­right­ed mate­r­i­al by gen­er­a­tive AI.

Over the past few decades, advances in infor­ma­tion tech­nolo­gies have allowed firms to move from deci­sion-mak­ing on the basis of intu­ition and expe­ri­ence to more auto­mat­ed and data-dri­ven meth­ods. As a result, busi­ness­es have seen effi­cien­cy gains, sub­stan­tial cost reduc­tions, and improved cus­tomer ser­vice. For one project, our artists drew the main char­ac­ter from every sin­gle pose and angle, a hand­ful of back­ground char­ac­ters and four build­ings. Then we can go and make a whole city out of that, and it retains the artist’s style,” said Tril­lo. “It allows us to do this world build­ing and iter­at­ing faster, rather than hav­ing the artists do each and every thing.” This isn’t over­ly shock­ing when you real­ize that most of these datasets are craft­ed by using AI or some relat­ed online tool.

Prompt Engi­neer­ing And Per­sonas

The per­son devis­ing the dataset tells the AI or tool to gen­er­ate tons and tons of per­sonas and store them in a dataset. The sur­prise for many is that the num­ber of AI per­sonas in these datasets is usu­al­ly in mil­lions or bil­lions of instances. You don’t have to be dog­mat­ic about using the AI per­sonas strict­ly as spec­i­fied in the datasets. When AI-gen­er­at­ed con­tent com­petes with human cre­ators, courts are unlike­ly to view its use of copy­right­ed mate­r­i­al as fair. This process turns a chaot­ic data ecosys­tem into some­thing that can be queried with pre­ci­sion.

Why does AI art screw up hands and fin­gers? — Bri­tan­ni­ca

Why does AI art screw up hands and fin­gers?.

Post­ed: Wed, 15 Jan 2025 08:00:00 GMT [source]

You can invoke mul­ti­ple AI per­sonas and use just the one from the dataset as the core base­line. Anoth­er equal­ly fine approach con­sists of describ­ing the over­all nature of a per­sona that you want to have invoked. On one side, it invites us to cel­e­brate inno­va­tion and the expan­sion of cre­ativ­i­ty; on the oth­er, it forces us to con­front the lim­its of our def­i­n­i­tion of what cre­ation itself means. Per­haps it’s not about deter­min­ing whether all this is good or bad but about learn­ing to live with a future where these ques­tions will remain open.

And last­ly, the biggest con­cern is that some fear that gen­er­a­tive AI might replace human jobs in cre­ative fields. A com­mon­ly ref­er­enced method of cus­tom-mod­el train­ing is cre­at­ing LoRAs, which refers to low-rank adap­ta­tion. Sources sug­gest­ed that an IP or spe­cif­ic project could involve cre­at­ing and apply­ing a set of dis­tinct LoRAs, such as one for a spe­cif­ic char­ac­ter and anoth­er for the ani­ma­tion style. I am going to look at one called FinePer­sonas and anoth­er dataset known as Per­son­aHub. The datasets that pro­vide AI per­sonas are pret­ty much all rel­a­tive­ly sim­i­lar. The typ­i­cal for­mat is a spread­sheet-like struc­ture that hous­es the AI per­sona descrip­tions.

Devis­ing From Scratch Or From Dataset

In the film and gam­ing indus­tries, gen­er­a­tive AI cre­ates real­is­tic char­ac­ters, land­scapes, and ani­ma­tions. AI-gen­er­at­ed music is also used for back­ground scores and sound­tracks. Gen­er­a­tive AI mean­ing can be defined as a type of arti­fi­cial intel­li­gence that is used to cre­ate con­tent. It dif­fers from tra­di­tion­al AI mod­els, which are typ­i­cal­ly used to recog­nise pat­terns or make pre­dic­tions.

Gov­ern­ments and orga­ni­za­tions will like­ly estab­lish reg­u­la­tions to address eth­i­cal and legal con­cerns. The term “gen­er­a­tive” comes from the word “gen­er­a­tion,” mean­ing the cre­ation or pro­duc­tion of some­thing. Essen­tial­ly, gen­er­a­tive AI enables machines to sim­u­late cre­ativ­i­ty and pro­duce out­puts that close­ly resem­ble human-made con­tent. Com­pa­nies face a vari­ety of com­plex chal­lenges in design­ing and opti­miz­ing their sup­ply chains. Increas­ing their resilience, reduc­ing costs, and improv­ing the qual­i­ty of their plan­ning are just a few of them.

AUG­MENT­ED HUMANS: “AI, CHECK MY GRAM­MAR”

Con­ven­tion­al spread­sheet skills are usu­al­ly all that you need to know. While fair use—a legal frame­work allow­ing lim­it­ed use of copy­right­ed mate­r­i­al with­out permission—has long been a pil­lar of cre­ativ­i­ty and inno­va­tion, apply­ing it to gen­er­a­tive AI is fraught with legal and eth­i­cal chal­lenges. We can use retrieval + gen­er­a­tive tech­nol­o­gy; ground­ed on our ontolo­gies and known pri­or knowl­edge, to assist in this inter­ro­ga­tion. We can begin to iden­ti­fy gaps in our knowl­edge, areas of con­tra­dic­tion, or cre­ate focus and reduce unnec­es­sary dupli­ca­tion.

This tech­nol­o­gy can help syn­the­sise infor­ma­tion into insights you can use, mak­ing sense of your data, con­nect­ing dots and high­light­ing pat­terns that would be impos­si­ble for humans to iden­ti­fy alone. Data Engi­neer­ing is the dis­ci­pline that takes raw, unstruc­tured data and trans­forms it into action­able, high-val­ue insights. With­out a strong data foun­da­tion, the $10M aver­age that 1 in 3 enter­pris­es are spend­ing on AI projects next year alone, are set­ting them­selves up for fail­ure. Gen­er­a­tive AI is a new and cut­ting-edge tech­nol­o­gy that is chang­ing the way we cre­ate and con­sume con­tent.

Fair use tra­di­tion­al­ly applies to spe­cif­ic, lim­it­ed uses—not whole­sale inges­tion of copy­right­ed con­tent on a glob­al scale. Yet even with the pos­i­tives described above, fine-tun­ing for con­tent cre­ation still holds a plau­si­ble degree of eth­i­cal and legal risk for stu­dios. Like­wise, even as a few AI stu­dios and inde­pen­dent cre­ators pur­sue new meth­ods, sources told VIP+ the major tra­di­tion­al stu­dios still see legal and con­sumer back­lash risks as rea­sons not to use AI for con­sumer-fac­ing con­tent. These stu­dio teams see fine-tun­ing as a way of exe­cut­ing on orig­i­nal IP devel­oped in-house. Sources reflect­ed that train­ing cus­tom mod­els speed­ed and scaled artis­tic out­put while remain­ing visu­al­ly con­sis­tent with the orig­i­nal IP or project.

• On one side, it invites us to cel­e­brate inno­va­tion and the expan­sion of cre­ativ­i­ty; on the oth­er, it forces us to con­front the lim­its of our def­i­n­i­tion of what cre­ation itself means.
• You don’t have to be dog­mat­ic about using the AI per­sonas strict­ly as spec­i­fied in the datasets.
• How­ev­er, some artists have gone fur­ther, involv­ing AI not as a mere pas­sive tool but as an active sub­ject in the cre­ative process.
• It is also used to cre­ate syn­thet­ic med­ical data for research pur­pos­es.

Sources described this process being done and seen as cre­ative­ly viable for ani­ma­tion. In-house artists or ani­ma­tors devel­op a “core set” of orig­i­nal con­cept art rep­re­sen­ta­tive of the orig­i­nal char­ac­ter or project. These assets form the dataset used to train any foun­da­tion image or video mod­el the stu­dio prefers (e.g., Sta­ble Dif­fu­sion). The result­ing fine-tuned mod­el can then be used to dri­ve sub­se­quent con­tent cre­ation, whether pro­duc­ing out­puts that repli­cate the studio’s spe­cif­ic char­ac­ters or an aes­thet­ic style present in the art assets. Gen­er­a­tive AI is pow­ered by advanced algo­rithms and machine learn­ing tech­niques.

PEO­PLE MOVES

For oth­ers, if you are con­duct­ing a sub­ject-based study and want to have a swath of AI per­sonas, or if you are unsure of what AI per­sona you want to invoke, these datasets can be quite valu­able. Indeed, any kind of large-scale test­ing of AI or using AI to gen­er­ate lots of out­puts of syn­thet­ic data can be stream­lined by lever­ag­ing an AI per­sona dataset. That being said, I don’t want to seem­ing­ly dimin­ish the hero­ic and thank­ful effort of those who put togeth­er these datasets. There is admit­ted­ly more elbow grease and hard work that goes into estab­lish­ing a use­ful and usable per­sonas dataset.

The use cas­es for gen­er­a­tive range over var­i­ous top­ics, from writ­ing to art and mar­ket­ing to health­care. One impor­tant thing to keep in mind is that it must be used respon­si­bly, like any oth­er AI tool. We can make the most of gen­er­a­tive AI by under­stand­ing its mean­ing, work­ings, and impli­ca­tions. “No scraped data will be part of the pipeline once that becomes avail­able,” said Tril­lo.

Every­one is enam­oured with gen­er­a­tive AI and state-of-the-art mod­el releas­es, often over­look­ing that it’s the data foun­da­tion that will make or break your use case (& the rel­a­tive invest­ment you’ve made). In today’s col­umn, I show­case a nov­el twist on the prompt­ing of per­sonas when using gen­er­a­tive AI and large lan­guage mod­els (LLMs). You con­ven­tion­al­ly enter a prompt describ­ing the per­sona you want AI to pre­tend to be (it’s all just a com­pu­ta­tion­al sim­u­la­tion, not some­how sen­tience). Well, good news, you no longer need to con­coct a per­sona depic­tion out of thin air.

• Auto­mat­ed writ­ing tools might under­cut oppor­tu­ni­ties for pro­fes­sion­al writ­ers. • AI-gen­er­at­ed text might reor­ga­nize or para­phrase exist­ing con­tent with­out offer­ing unique insights or val­ue. While these fac­tors have worked well in tra­di­tion­al sce­nar­ios like crit­i­cism, par­o­dy or edu­ca­tion, gen­er­a­tive AI presents unique chal­lenges that stretch these bound­aries. Gen­er­a­tive AI has been mak­ing head­lines for it’s poten­tial to rev­o­lu­tionise the way we think,work and solve prob­lems, with McK­in­sey pro­ject­ing it will con­tribute up to $4.4 tril­lion dol­lars to the glob­al econ­o­my annu­al­ly.

• Though the AI appears to often con­vinc­ing­ly fake the nature of the per­son, it is all still a com­pu­ta­tion­al sim­u­la­tion.
• Sources sug­gest­ed that an IP or spe­cif­ic project could involve cre­at­ing and apply­ing a set of dis­tinct LoRAs, such as one for a spe­cif­ic char­ac­ter and anoth­er for the ani­ma­tion style.
• Gen­er­a­tive AI mod­els are trained on vast datasets, often con­tain­ing copy­right­ed mate­ri­als scraped from the inter­net, includ­ing books, arti­cles, music and art.
• All you need to do is search the dataset to find what you are inter­est­ed in as an AI per­sona.

Yet the prospect of using gen­er­a­tive AI for ani­ma­tion still pos­es big­ger-pic­ture eth­i­cal and legal chal­lenges for the indus­try. No need to derive AI per­sonas from scratch when you can leisure­ly and con­ve­nient­ly lean into an AI per­sona dataset. Of course, this is based sim­ply on the numer­ous speech­es, writ­ten mate­ri­als, and oth­er col­lect­ed writ­ings that sug­gest what he was like. The AI has pat­tern-matched com­pu­ta­tion­al­ly on those works and mim­ics what Lincoln’s tone and remarks might be.

In an amaz­ing flair, the AI seem­ing­ly responds as we assume Lin­coln might have respond­ed. These cas­es under­score the dif­fi­cul­ty of apply­ing tra­di­tion­al fair use prin­ci­ples to gen­er­a­tive AI’s large-scale, auto­mat­ed process­es. The answer depends on whether the AI’s use of copy­right­ed mate­r­i­al sat­is­fies the fair use cri­te­ria, and in most cas­es, it does not. • An AI art gen­er­a­tor might cre­ate an image resem­bling a copy­right­ed paint­ing. Gen­er­a­tive AI has emerged as a trans­for­ma­tive force in tech­nol­o­gy, cre­at­ing text, art, music and code that can rival human efforts.

Why AI art will always kind of suck — Vox.com

Why AI art will always kind of suck.

Post­ed: Thu, 23 May 2024 07:00:00 GMT [source]

In those two exam­ples, I used first a physics teacher and then an art teacher. I might want to run through a wider range of teach­ers that cov­er a vari­ety of aca­d­e­m­ic spe­cial­ties. I then used that text in a prompt and got AI to pre­tend to be that per­sona.