Wikipedia:Articles for deletion/Media synthesis (AI)


 * The following discussion is an archived debate of the proposed deletion of the article below. Please do not modify it. Subsequent comments should be made on the appropriate discussion page (such as the article's talk page or in a deletion review).  No further edits should be made to this page.

The result was Draft:Media synthesis (AI). The article has been moved to the draft space. Thanks everyone for participating and assuming good faith! Missvain (talk) 16:31, 29 January 2020 (UTC)

Media synthesis (AI)

 * – ( View AfD View log  Stats )

A NEOLOGISM grouping a number of concepts that does not appear in multiple reliable independent in-depth sources to satisfy GNG. The only source that I see that uses the term "media synthesis" is a blog post written by the author of the article. In fact, the only search hits appear to be posts by the author. To me, this fails NOTESSAY.

The article is severely-BOMBARDed by sources that are only relevant to the topic because the article claims so. In other words, it is SYNTHesis of sources grouping various concepts under the label, even though none of the sources for those concepts deliberately use the term "media synthesis". As it appears, it's an invented concept by the author (and likely falls under WP:CSD).

The author also created Category:Media synthesis. They also appear to have gone through other articles and added mentions of the term and category (e.g. ), none sourced as far as I can see. This has gone a little beyond WP:BRD and brings up FAITACCOMPLI issues. However, the resolution of this AfD should hopefully also resolve whether category or these other mentions are suitable (I do not believe they are, but I rather not revert en masse).

The article was moved out of draft space, so not re-incubating it. (The author did not resubmit the draft, so there was no chance to review it.) I assume CSD/PROD would be contested (especially since there are so many sources at a glance), so going with full AfD. — HELL KNOWZ   ▎TALK 11:55, 22 January 2020 (UTC)
 * Note: This discussion has been included in the list of Technology-related deletion discussions. —  HELL KNOWZ   ▎TALK 11:55, 22 January 2020 (UTC)
 * Note: This discussion has been included in the list of Computing-related deletion discussions. —  HELL KNOWZ   ▎TALK 11:55, 22 January 2020 (UTC)

The primary issue with the Media Synthesis article is that I did indeed create the current term before a more widely used and very highly similar term, "synthetic media", was coined. If you may, I'll provide at least 50+ links to articles (including official government documents) that use this term and even define it much the same way. Most notable is the Congressional Deepfake Report Act. My primary reasoning for using the term "media synthesis" was more due to grammatical rules. If it is more prudent to rename this article "Synthetic Media (AI)", it very well should and ought to be done.

As the page is still largely under construction, a number of these sources are to be added as time goes on and the article is further cleaned up.

In regards to moving the page from draft to mainspace, I've been told multiple times by others that this is something available to a user who is autoconfirmed, that I need not wait for administrative approval if the article is already complete. If this is the wrong move, please correct me. I did indeed resubmit the draft, at least over a month ago, and had been patiently waiting for its approval. It was only recently when I was (perhaps erroneously?) informed that I need not wait in the first place if I was not an anonymous user.

Rather than deletion, my personal final verdict is to rename Media Synthesis (AI) to Synthetic Media (AI) and add the proper sources to justify the name. --Yuli Ban (talk) 14:25, 22 January 2020 (UTC)

In regards to further instances of the usage of "synthetic media", I'll simply post ten links just to not clutter up the page. Hopefully this re-establishes the argument towards renaming the article:
 * Witness Media Lab
 * New Yorker
 * CMS Wire
 * Genetic Literacy Project
 * Washington Post
 * Forbes
 * Medium, as part of the Wall Street Journal
 * Fast Company
 * The Hill
 * Google AI Blog

All of these articles both mention the term "synthetic media" as well as define it in some capacity, with the most basic being "AI-generated media" (hence the alternate titles for the article). They also refer to "synthetic media" as having multiple branches, of which the likes of deepfakes and human image synthesis are individual branches. Thus, it can be ascertained that at least most of the central argument of the Media Synthesis (AI) page is still applicable.

Also, my reasoning for using multiple tangentially-related sources derives from the central reasoning that "media synthesis" refers to a family of processes of applied computational creativity and algorithmic generation; therefore, image synthesis and procedural generation alike qualify. --Yuli Ban (talk) 14:39, 22 January 2020 (UTC)


 * If you want to draft or create an article Synthetic media using sources that actually use the term and stay within their scope, then that's a different subject and scope. I scanned through these and the best I can see is that they are talking about "field of deepfakes, or “synthetic media”". But you have equated this "Media synthesis" with all the subjects -- synthetic media/deepfakes, AI-generated content, procedural generation, speech synthesis, etc. And the source for this grouping is a blog post you wrote yourself. That's WP:OR and not acceptable. Gathering sources to surround a topic with related topics based on such grouping is thus equally WP:OR and WP:SYNTH. — HELL KNOWZ   ▎TALK 15:19, 22 January 2020 (UTC)

A few more points of contention:

1: Synthetic media is media synthesis. As mentioned, I created the latter term a good bit of time before the former became widely used. Both refer to the exact same thing. In fact, when I first created the term "media synthesis," calling it "synthetic media" was one of my options, but I decided against it due to it being a noun adjunct. Now I see that "synthetic media" is the more widely used term anyway.

2: Synthetic media is not perfectly synonymous with deepfakes. To use even more sources, it describes everything you've mentioned as an unbrella term for all these different sources. This is how it is usually used: separately from deepfakes, often placed over it to describe a family of different but related techniques (including speech synthesis, natural language generation, music synthesis, etc.)

See more:
 * Media Post's article "What is Synthetic Media?"
 * Psychology Today, using synthetic media to describe artificial humans & speech
 * Securing Democracy's definition of synthetic media (Including this helpful quote: "Experts generally prefer the more general term synthetic media, which refers to the generation or manipulation of media—video, audio, imagery, and potentially text—that would have been difficult and expensive to create prior to recent technological advances. This relies primarily, but not exclusively, on artificial intelligence.")

My final assertation remains: Media Synthesis (AI) can be moved directly to Synthetic Media (AI) with added sources, but should not be deleted or greatly culled. Yuli Ban (talk) 15:50, 22 January 2020 (UTC)

I do thank you for your assistance, however. All I really wanted this entire time was added help in clarifying the fundamental concepts and finding any flaws, holes, or poorly-expanded research, especially to assist in establishing a clearer centralized terminology for all related technologies deemed "synthetic media." This entire time, I worried that the lack of usage of "media synthesis" in the sources would prove to be a problem right up until I discovered that "synthetic media" was used in increasing amounts. By that time, however, the original article was already mostly complete. To expand upon my reasoning, the intention was that references to sub-branches of synthetic media (e.g. speech synthesis, natural-language generation, procedural generation, human-image synthesis, etc.) were enough to be used. Of course, now I see that there is a rule against synthesized sources. That the term "media synthesis" was my own and I was the primary source for it was also a worry when originally drafting the article last year, and in fact was the reason why it was not drafted sooner: the intention was to write it as early as the fall of 2018, but I feared self-authorship would lead to deletion. It was only when constant efforts to get others to take the initiative failed to materialize anything tangible that I decided to act. As a result, I've had very few to no other persons actually helping me prune or refine details. Considering the subject matter, urgency of the debate, short and long-term implications, and rapid development of the technology, it would be very unfortunate for the article to be outright deleted, especially since the primary issue is entirely the name of the article itself and that rectifying it (plus adding the proper sources for the terminology) would clear it of any further issues. Indeed, perhaps the #1 reason for the article's creation was to help raise awareness for this burgeoning field of technology (which is why it was soon turned into a category, with another intention being to use it as a basis for navboxes and portals in the future).

Yuli Ban (talk) 20:09, 22 January 2020 (UTC)


 * Comment - I am trying to understand reasons for deletion here. We have Deepfake as an article, and synthetic media appears notable - although I have not yet checked the reliability of all sources. Nevertheless it is a widely used term and I think is almost certainly notable. My question is whether Deepfake and Media synthesis (AI) are sufficiently differentiated for two articles. What is the argument for there being both? -- Sirfurboy (talk) 21:23, 22 January 2020 (UTC)
 * I'll be honest that I am now more confused after author's explanation than I was before nominating. But here's what I think I understand. "Media synthesis" is a term the author created (they say as much above and the only source to support the term is their own blog post). It fails GNG. "Synthetic media" is an actual term used by sources -- a similar but less encompassing topic. It might pass GNG. (The terms happen to sound very similar, so it's confusing at a glance.) The author wanted to promote their term "media synthesis" as a broader concept and wrote the whole article that way, collecting a lot of related (but with unsourced connection) concepts along the way. The article's intro "also known as" is based on author's own blog post. — HELL KNOWZ   ▎TALK 22:00, 22 January 2020 (UTC)
 * Thank you. Yes, agreed about the explanation, but I don't think he has wholly created that term. It is used in this article and indeed in this paper just as a couple of instances. It is not entirely his invention (or if it is, then it is an invention of his that some others are using). Yet you are correct that "synthetic media" is the more common term. I just did a "google poll" using each search term, coupled with "deepfake" (to weed out a lot of pages using the term in different contexts). Media Synthesis only had 1,950 hits and synthetic media had 49,500. So it is clear which term has wider usage in the context. Yet that only tells us something about what the page title should be. To qualify for deletion, this has to fail GNG. I don't think it fails GNG as a concept.


 * The other reason I would go with deletion and redirect, however, is if there were no difference between the intended scope of this article and the Deepfake article. As deepfake is just a popular name for synthetic media, it is not clear to me that this article is required. Redirect to deepfake would appear to make more sense (or even rename Deepfake to Synthetic Media and leave in a redirect from there). I think it is for the author to explain (succinctly please) how the scope of this article is not entirely encompassed by Deepfake. -- Sirfurboy (talk) 22:24, 22 January 2020 (UTC)
 * The nomination is for the topic "media synthesis" (and I think those two mentions are coincidental and not trying to coin or use an established term). I agree that "synthetic media" may be notable. But I don't agree that this article can be renamed and fixed without a WP:TNT. There is nothing in this article directly supporting what synthetic media is. To me, all content is just author's OR and SYNTH of tangential sources. Lead doesn't support it. "Media" doesn't even mention it. "History" just discusses content generation. "Branches" has no main sources and just starts listing fields. And so it goes on mentioning individual fields, but never actually citing sources that conclusively say that "field X is part of synthetic media and here are some facts about it". The whole article is akin to a (decently-written) academic essay, but the opposite of the OR principles of Wikipedia. Anyway, that's how I see it. I'm expressing it very diplomatically, but I honestly don't think it even passes CSD#A11. — HELL KNOWZ   ▎TALK 23:08, 22 January 2020 (UTC)
 * Yep, but deletion is not cleanup. Although maybe WP:TNT is required. To come to a view on that I am going to have to sit down and read the whole article properly, so I won't come back on this until tomorrow now. In the meantime, I will be interested on the author's comments regarding overlap with deepfake. Thanks. -- Sirfurboy (talk) 23:15, 22 January 2020 (UTC)

Author here: The fundamental issue is that "deepfake" in its current form describes a method of face-to-face style transfer, which is indeed a kind of synthetic media, but doesn't describe the full gamut of synthetic media itself. Think of it as being similar to the difference between subgenres of hard rock music, and how "death metal" is a kind of hard rock but wouldn't accurately suffice to describe the entirety of hard rock music. Or, perhaps more on topic, how "machine learning" describes a subfield of artificial intelligence but would not be sufficient to describe the entirety of the AI field. In terms of scientific research and data science, "deepfakes" does not refer to all sorts of algorithmic media generation. Even colloquially, it is only used as a source of comparison (i.e. "the deepfakes of [X]" or "deepfakes for [X]"), which implies that deepfakes is being used more to patch for a lack of a central terminology. There has also been some developments in making deepfakes work for voices. Thus, deepfakes describes swapping between two media, principally video and audio.

Synthetic media, thusly, is the parent "catch-all" term and principally refers to any sort of applied usage of AI-generated data/media. This means it includes deepfakes, but also things unrelated to deepfakes (such as text synthesis, music generation, speech synthesis, etc.) Since "deepfakes" is used predominantly to describe face-swapping, using it interchangeably with synthetic media tends to cause too much confusion, hence why publications often point out that deepfakes are merely a kind of synthetic media or that deepfakes is "the most prominent form of synthetic media. This implies that there is more to synthetic media than just deepfakes. Indeed, they will usually reiterate what I have said: synthetic media refers to any "AI-generated, enhanced, or manipulated media or data." It is often mentioned that this goes for video, image, text, and voice; the implication is that any media whatsoever can be generated. "Synthetic media" encompasses the same scope as "media synthesis," perhaps losing Procedural generation only due to a lack of sources currently combining the terms.

There are some more papers and articles that go a bit more in-depth. If the article remains (but is renamed), Synthetic media (AI) would become the basis for a navbox, and Deepfake would be a link within in a "Video" or "Image" cell, separate from something like Pop music automation and natural language generation.

Or to use an active example, my voice and face being swapped for Sirfurboy's for use by HELL KNOWZ is a deepfake. This, as well as Sirfurboy generating a new, novel face via human-image synthesis and then generating a short story to go along with it via OpenAI's GPT-2 transformer, would all qualify as "synthetic media." This new person who does not exist could then be made to sing via MIDI files or waveform manipulation (set to an AI-generated song), and this quasi-music video could be made interactive if a neural network generated video game assets and the proper coding instructions (not unlike NVIDIA's work here). All of such, too, is synthetic media. Swapping the AI-generated person for another AI-generated person via autoencoding methods, however, would qualify as a deepfake. Generating a new person would not be considered a deepfake in technical terms, but might be referred to it in the media by some publications. Or it would be compared to deepfakes, but formally considered something different.

At this point, it seems the primary issue at hand is more whether to rename the article to Synthetic Media; we've all established well that the term does exist, and I have been trying to explain why it is not synonymous with deepfakes. I'm fully in support of doing this and editing all the outgoing links & category page.

Yuli Ban (talk) 23:15, 22 January 2020 (UTC)
 * This is original research. You keep describing things using words like "thusly" and "hence" and "implies". That is not how Wikipedia works. You have to show sources that directly say those things. Sources that say things like "Deepfake is a field of synthetic media". Not blog posts by unknown authors with no editorial oversight. Not churnalism tangential articles that don't define their terms. — HELL KNOWZ   ▎TALK 23:31, 22 January 2020 (UTC)
 * Don't misunderstand me; I see exactly what you're saying. My point is that the fundamental problem comes down to the name of the article itself; if it were renamed, I'd be able to fix the majority of the problems. There are plenty of articles that state almost exactly or some variation of "deepfake is a field/type/branch of synthetic media," and similarly for image synthesis, speech synthesis, audio synthesis, natural-language generation, and video synthesis. Synthetic media as a term is either referenced or defined in many of them, and the definition invariably comes down to a fuzziness that starts with "AI/algorithmically-generated media/data" or "automated art." I chose to use "AI-generated media" as a direct synonym to "synthetic media" since they are almost always used in the same article or paper to refer to the same thing.
 * With "media synthesis," there are vastly fewer references to or definitions of this, even though it's the exact same concept with the words swapped. And because I had been unaware of the prevalence of "synthetic media" as a term describing this field of technology before very recently (as most searches came up with the biological meaning), I wound up having to infer the relationship between the many subfields to have a fully-established article. This + overzealousness on the topic led to it becoming largely original research. — Yuli Ban (talk) 00:09, 23 January 2020 (UTC)

I've decided to go ahead and begin work on Synthetic Media (AI) as a replacement article, reworking most aspects from the ground up to better tie everything directly to "synthetic media." Since I've saved/downloaded the page, media synthesis can be nuked. There are thus only three hang-ups:

1) Synonyms. "Synthetic media" and "AI-generated media" are perfectly synonymous (the definition of synthetic media is that of autonomous algorithmic generation of data/media), so the question here would be whether or not a report and paper on "AI-generated media" that nevertheless does not contain the term "synthetic media" would qualify as a suitable source. It seems obvious and uncontroversial to me that this would be the case, especially since the term was largely unused by the media before 2019, but the exact same principle was described via "AI-generated [i.e. synthetic] media". This is likely the only instance where I'll see any feedback from others and would like some alternative input. This would greatly ease finding sources, but it's also a very fuzzy area that really comes down only to the minutia of language. No source I've found, from blogs to scientific papers to government documents, lists any separation between "AI-generated media" and "synthetic media" and indeed often very explicitly state the latter is the former.

2) Applications, concerns, and impacts. Admittedly, this could use culling to remove some of the more personal takes on long-term implications to keep it purely down to the facts, but a primary issue with all of these is that, since they all describe sub-branches of synthetic media even as defined by many other outlets, at least some ought to remain.

3) Existence as a category. Seeing as synthetic media is supposed to describe a very wide range of different technologies (and as the term becomes more prominent, it will almost certainly be used by articles, papers, and documents to describe whatever currently "doesn't fit" the term as per this page's discussion thus far, so the article will eventually refill to the same level as the previous one), its existence as a category seems obvious to me, but again, I'd like alternative takes before making a decision.Yuli Ban (talk) 05:31, 23 January 2020 (UTC)


 * Draftify - It is clear this is not ready for mainspace despite a good deal of effort by the creator. In draft it can also be renamed to Synthetic Media (AI), and reworked to remove the OR and focus the page more on the subject. The history is a history of AI rather than the subject and the whole page could be focused much better, although the lead summary does define it as a subject that is notable and broader than the existing Deepfake page. There is no reason for deletion per se, except WP:TNT, and sending to draft will allow the creator to work on it based on comments here. -- Sirfurboy (talk) 08:02, 23 January 2020 (UTC)
 * Also, just to add, per the nomination, this was moved out of draft without review. Will the creator agree to submitting to review before moving out of draft in future? If not, then my view would change to delete, under WP:TNT. -- Sirfurboy (talk) 10:11, 23 January 2020 (UTC)


 * Draftify - I've read over this page and the article being discussed. Seeing the various sources which Yuli linked earlier (those which use synthetic media as a term), I think it's definitely justifiable to have a wikipedia article about the subject. However, I've also noticed that the sources which Yuli linked earlier aren't actually used within the wikipedia article itself, and the article instead uses various sources that are related to, but don't explicitly mention, synthetic media. I also want to note how the page was moved into the mainspace without being approved by a reviewer, which in and of itself justifies making the page a draft again. While I see the point in having a page on the subject, the one which Yuli has moved into the mainspace has real problems and needs work to be done on it. Apathetizer (talk) 02:43, 25 January 2020 (UTC)


 * The above discussion is preserved as an archive of the debate. Please do not modify it. Subsequent comments should be made on the appropriate discussion page (such as the article's talk page or in a deletion review). No further edits should be made to this page.