The world of artificial intelligence has been destined for several years to leave us with our mouths open from time to time.and the truth is, they do. From specific companies in the sector, such as OpenAI, to large technology companies, such as Google, we have seen in these times how artificial intelligence has learned to do everything from writing text on a specific topic that we suggest to scaling gaming. pictures in real life. time to improve the gaming experience.

A very interesting area of ​​this development of artificial intelligence can be found in image generation, in some cases purely by chance, in others on the basis of a description previously provided by the user. And in this last group we will find both specialized solutions for a specific type of image, such as GauGAN2 from NVIDIA or a popular face generator from Google, as well as those for general use.

Among the latter, so far the gold medal corresponded to the OpenAI DALL-E system, but this has changed with the presentation Image, a new system designed by Google to generate images from text descriptions. And this is because, as we can see on the presentation page of the project, some of the images generated by this artificial intelligence can pass perfectly, except that at least some of them reproduce not completely realistic motifs.

However, it is important to mention that as always in these cases Google will select the best results achieved, but we can assume that there were also some not very effective outputs. In addition, this is something we won’t be able to verify at least in the short term, as Google has decided not to make Image available to potential users / customers at least for now.

In this example, you can see that the result of the image is not always perfect.

There are two reasons for this. The first is the company is concerned about the possible harmful use of ImageFrom fake reports to illegal sexual content like CSAM, Google is common sense in this regard, because we can be sure that such use will occur almost from the moment this technology becomes available to everyone.

The second, and also very meaningful, is that, as with virtually all massive datasets used to saturate artificial intelligence learning processes, we can find in them some prejudices that should be corrected. The problem is that performing a detailed analysis of these datasets is a daunting task that makes companies the size of Google fade. Instead of approaching it this way, the approach is that the AI ​​itself will be able to learn about these distortions in order to correct them automatically.

Personally, I admit that I’m not sure if I find it fascinating or if it scares me, so maybe a little bit of both.