Focus of the letter 46

I described to MultiMatte what I wanted to keep in an image. A few seconds later, it was cropped, which usually requires editing software and practice. Then I looked at the edges, not as “clean” as they looked in some places. Nothing on the screen says if it’s successful, you need “an eye”.
Kario does the same thing for a video. One instruction, and you get a motion design cut into scenes, with text, shapes, images, and audio on separate, editable layers. This is the structure that an editor works on without you having to learn it.
Sharper AI makes complex productions accessible by generating reports, data in a spreadsheet and presentations from connected documents.

It is a real broadening of possibilities. It is also fractures that are superimposed: that of skills, that of access or that of the ability to judge results.

One has to do with skills. You have to know how to express an intention, structure a request, iterate, contextualize. Sometimes, you also often need English: this week, MultiMatte and Instrument Playground only accept this language. Finally, you have to know how to judge the result that the machine delivers.
Knowing how to ask, knowing how to do, knowing how to judge, knowing how to take back control: these things can be learned, and not everyone has access to them, or not in the same way.

Another touch on access: hardware, connection, and then very quickly the subscription that the freemium model tries to impose. Sharper AI offers 300 credits per week in the “Lite” version, and my tests consumed 160. Beyond that, expertise becomes payable. Switching to another tool or installing an open-source model is possible, but it still requires skills and a sufficiently powerful computer, and therefore expensive.

This week, these applications are removing one barrier without removing the other. MultiMatte is open-source, without an account, and seems unlimited, but it requires English and “an eye” to judge. Kario doesn’t have to know how to edit, but he accounts his credits.

So there remains the ability to check who brings a new fracture. Kario stopped dead in the middle of the test, before becoming available again the next day. Was what he had already generated usable? Knowing if a clipping is clean at the edges, if the rhythm of an animation holds, if a report is true: this judgment does not come by simply pressing a button.
This reflex is in line with this week’s readings on “IApocalypse”: check rather than take word for it, whether it is a result or a speech, especially the one pushed by the owners of large models. LaborIA says it on the scale of social work: AI for the form, not for the substance, and the time saved should be used in part to check and reflect.

The same reflex applies well beyond the image: in the readings, an announcement by OpenAI on the Navier-Stokes problem offers a very good example. Two readings are possible: either expertise has become democratized, or only its results, and not for everyone. Can someone who has never retouched an image spot a failed clipping, and does he even have enough to redo one without the available tool? These tools do not eliminate the need for expertise, they shift part of the manufacturing to critical control. And the universal access they promise can also become a factor of social differentiation itself.
Education, sharing, support, and critical thinking are certainly more necessary than ever. Without them, this universal access to tools remains an illusion that accentuates inequalities.
When the tools give everyone the impression that they can produce like an expert, how do we still learn to recognize what really falls under expertise?

Echo:

• Letter 36, “What is left when the tool has done the job?“,
• Letter 28, “From “wow” to “oh yes?”: making with the machine and understanding what we are making“.