Now that you know what A.I is, and how it works (if not, check the list of articles down below), I’m sure you may be thinking: “If A.I is trained by views that exist in this world, then how can you make ethical or fair A.I art?” and the truth is - we can’t! In the machine learning process, where neural networks break down information, computer scientists can’t translate “fairness” into conceptually blocky data (Coleman & Allin 2021, or refer to the article What is A.I art? ). Computer scientists as well as many other people are involved in the critical decision of how they can translate “fairness”. It’s highly difficult because bias is baked into the training of A.I into the classification patterns or properties of symbols - and there is no way to remove this(Coleman & Allin).

 

In machine learning, this primarily means that it weighs features in a hierarchy, but the process of classification itself is neither good nor bad (Allin). Rather, when it comes to training and decision-making, it can produce outcomes that are understood as discriminatory or unfair to the world (Coleman & Allin).

 

 

 

​

​

​

​

 

 

 

 

 

 

​

​

 

Machine learning is a mirror that can reflect our creativity and ability to make decisions (McManus 2021). However, it cannot reflect our common sense and passion as humans (McManus). So how do we solve this issue? The answer lies in where it all began: the data and the people who train the algorithm. 

 

Arti’s art generator runs with a database compiled of free stock images, open-access art and original artwork produced by a team of artists our company works with. We also work on an open donation system where anyone, regardless of artistic expertise can donate their artwork to improve the machine learning process. Moderators at Arti make sure to filter out any unethical or inappropriate images, such as those that promote racist, homophobic, transphobic, or sexist ideals as well as images that contain nudity or faces of actual people before it gets accepted into the algorithm. In case the generator produces an inappropriate result, users can report the image to be reviewed by an employee at Arti.

 

When it comes to art styles Arti does not want the A.I to perfectly match a specific artist’s style. The diversity of the training data is meant to ensure this, alongside blocking the ability to type specific names of artists into the search bar (artists who are active and are not affiliated with Arti). At the end of every generated result, the influences of artists that the A.I drew from will be listed. These will link to either their account on Arti where you can find more of their contact details, information and art, or their preferred method of support (such as PayPal or Patreon).