Does AI plagiarize and take credit for work of others ?

In some respects I think it's fair for the user to do that research, i.e. make sure the AI output is valid and functioning, and isn't using anyone's code that may require permission or attribution.

Personally, I prefer to examine each line of AI-suggested code, to see if I can do it better, or tailor it to my style, and that ends up with some re-write, and guides the AI for subsequent lines of code too.

Except that no one is doing this, and I've already seen my own code being output verbatim by Copilot, even with the typo I made at the time.

Sure, that code is under a permissive license in terms of usage, but that license clearly requires citing me and the other contributors, which is NOT happening here.

How's that different from pre-AI though? If people want to abuse license terms, they would have done so pre-AI too.

I think those that respect license terms, will be taking care as they were before. Personally, I do that by not accepting large chunks of code, without going through each line, and guiding the AI on a per-line basis.

The main difference is that before AI "pumped" all that code, one had to get to it and it was hardly possible to miss the comment at the top or the LICENSE file describing the conditions. Sure, one could always ignore it and I'm sure it happened.

But with AI, the code is given without any context, without citing any reference, even sometimes giving the impression this is its own creation "out of the blue".

I see you have a good discipline about that, but I can assure you this is a rare situation.

Some hobbyists might just take code perhaps, but I would have thought almost every engineer from junior level upwards will have been trained about not taking source code without considering the license. It's drilled into them because of the liability if orgs use stolen code.

There are disreputable firms that probably don't care all too much I guess. 

If we set the responsibilities of individual users aside for a moment it does seem from examples provided by obones that the AI algorithms are playing fast and loose with intellectual property and stripping out the credits or conditions for using a specific code snippet... and that doesn't quite sit right in my opinion.

I agree it's not ideal, I just don't think it's as big an issue as some people think, because engineers still need to consider their sources, and they needed to do that pre-AI too. If it was such a big issue, people wouldn't be using it.

I acknowledge the limitations, but equally I believe people should and do work around it in a moral way, same as engineers did pre-AI (well, at least the normal engineers; there are the immoral ones like those responsible for emissions scandal, dodgy firms that won't care about laws regardless, and so on).

Has life become harder for an engineer to discover the sources? Maybe, but that's a decision the engineer has completely under their own control, because they are not obligated to use AI output, or they could choose to restrict their AI usage to (say) tasks that are not going to form part of their product, and their are plenty of those sorts of tasks to benefit from using AI as well.

Today the responsibility (and liability) lies with the engineer and the firm that accepted the code to a large extent. Would that responsibility change even if AI provided a thorough list of sources? Probably not, for two reasons:

Firstly, because AI is not 100% accurate, the engineer would be expected to check regardless.

Secondly, practically speaking, I cannot imagine governments allowing individuals and firms to sue AI companies for unbounded amounts of mistakes, because AI companies cannot be expected to anticipate all repercussions. An analogy is one's mobile phone data connection going offline. If that occurs when someone wanted to make a stock transaction and they lose millions, they could try taking the phone company to court, and the case would get thrown out, because it's completely impractical for commerce if the phone company could be liable for this. Today it's much easier for the user to "do something" to reduce risk, than it is for the AI firm or phone company respectively.

I agree it's not ideal, I just don't think it's as big an issue as some people think, because engineers still need to consider their sources, and they needed to do that pre-AI too. If it was such a big issue, people wouldn't be using it.

I acknowledge the limitations, but equally I believe people should and do work around it in a moral way, same as engineers did pre-AI (well, at least the normal engineers; there are the immoral ones like those responsible for emissions scandal, dodgy firms that won't care about laws regardless, and so on).

Has life become harder for an engineer to discover the sources? Maybe, but that's a decision the engineer has completely under their own control, because they are not obligated to use AI output, or they could choose to restrict their AI usage to (say) tasks that are not going to form part of their product, and their are plenty of those sorts of tasks to benefit from using AI as well.

Today the responsibility (and liability) lies with the engineer and the firm that accepted the code to a large extent. Would that responsibility change even if AI provided a thorough list of sources? Probably not, for two reasons:

Firstly, because AI is not 100% accurate, the engineer would be expected to check regardless.

Secondly, practically speaking, I cannot imagine governments allowing individuals and firms to sue AI companies for unbounded amounts of mistakes, because AI companies cannot be expected to anticipate all repercussions. An analogy is one's mobile phone data connection going offline. If that occurs when someone wanted to make a stock transaction and they lose millions, they could try taking the phone company to court, and the case would get thrown out, because it's completely impractical for commerce if the phone company could be liable for this. Today it's much easier for the user to "do something" to reduce risk, than it is for the AI firm or phone company respectively.

Although I began my last response asking that we set aside the responsibilities of the individual user aside for a moment, I do agree with you that the user does have a responsibility for due diligence... and yes, I think for the most part big tech shouldn't be liable for any negative outcomes from search results or AI generated code.... but I still don't think they should be let off the hook for violating intellectual property.

Some of the arguments that suggest IP violation are unreasonable, because some of the AIs do not regurgitate verbatim; they instead try to generate output that suits your own coding style and function prototypes. I think it's fair game to learn from publicly available content if it's retold in different words/syntax. Humans do that each time they read a copyrighted article in a newspaper and then tell their friends about what's going on. No-one suggests they are violating IP.

It may well turn out that some AI's will never be able to cite all sources for their generative output. Maybe we need to change our perspective on things, because it's pragmatic that some of that responsibility will still be in the user's hands for now, not AI.

I won't be able to persuade you further... but I leave you with this... "It's unfair to say the student cheated on the entire exam.  After all he didn't need to peer over anyone's shoulder when he was writing his name."

In your example, the student chose not to copy the name.

What if AI had printed all names/sources alongside the generated content? A user could still decide to ignore the license terms at those sources. In other words, those who want to plagiarize will do it anyway.

There really is no difference. Those same people would have done it pre-AI and post-AI, so why the hang-up that a user needs to make some effort to locate the sources if he/she makes the decision to use content?

The decision and repercussions (and clearly liability!) always rests with them and the firms that accepted the content into their codebase, no-one is suggesting anyone can be excused for getting it wrong, just because AI might have not provided sources (or might have provided incorrect sources).