To me it seems, the whole subject requires additional consideration in licensing. It is a little like applying telephone based law to the internet. It will not 100% fit.

If the creators interests are not clearly expressed anymore with a license, we need updates to the license texts.

Let's look at MIT:

____________________

"Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. [...]

____________________

From the license text alone, it would not be clear to me, why anyone could claim that the OpenAI codex or the Github Copilot would require attribution to any of the used MIT source code to generate the AI model. The AI model is simply not a copy of the source or of a portion thereof. It is essentially a mathematical / statistical analysis of it.

Now what about any generated new source? How similar does it need to be to any source to be a copy? At what size of the generated code it qualifies to be a copy instead of a snippet of industry best practice?

Where does the responsibility for attribution lie? Should we treat the AI code generation models like a copy & paste program? Usually you cannot really say where the copy came from 100% - how do you know what factors influenced it?