Deepfakes — The Good, The Bad & The Dirty

The Top-3 positive & negative uses of deepfakes — through a futuristic lens

Andrey Gizdov
8 min readSep 13, 2022

You’ve probably seen those face-swaps that people do with filters on Snapchat & Instagram. While funny to watch, in most cases you can still tell that the swapped face is not real. Well, deepfakes are changing this very rapidly:

Tom Cruise

The video of Tom Cruise you just saw is not real and has been created by such a “deepfake” program — Tom Cruise never did what’s on the video. Would you have guessed that? I personally wouldn’t have…

However, besides viral videos, what else might this technology bring to the future? Without further ado, let’s find out.

In a nutshell — what is it?

Essentially, a deepfake program remembers the facial features of two people — Person A & Person B — and is then able to mimic Person A using the facial features of Person B. Let’s give an example.

Suppose that President Barack Obama is Person B & Jordan Peele is Person A:

Jordan Peele (Person A) & Barack Obama (Person B)

A deepfake can make the facial movement of Person B (President Obama) strictly follow that of Person A (Jordan). The same applies to voice –words told by Jordan, are mimicked by Obama, in Obama’s own voice:

The Limitation

The only thing those programs need in order to produce a realistic fake, is data — a lot of it. We are talking hundreds of facial images at various angles & good lighting. Otherwise, the results could turn out not as impressively as one would wish:

Unsuccessful deepfake (Harry Potter)

This limitation is both a blessing and a curse — if there wasn’t a need for hundreds of facial images, one could easily put your face on whatever they want, and it would look legit! What came to your mind first when I said that?

Starting to sense how this technology can be a lot more impactful than just viral videos on social media? You’re on the right path — it can be used for a lot of things. Let’s see what exactly.

Applications: The Good (Top 3)

#1. Voice recreation

Sometimes, people lose their voice. Amyotrophic Lateral Sclerosis (ALS) & chronic laryngitis are a few conditions that can lead to this.

If this happened to you, would you want to sound like Siri for the rest of your life?

Siri

Don’t get me wrong, Siri doesn’t sound bad (especially the Australian one), but wouldn’t it be better if you could just sound like you?

This is what Team Gleason are doing:

Using deepfakes to create a virtual image of your voice — over 70,000 people are living with ALS so this could certainly impact humanity for the better.

The Future

My take for this particular application of deepfakes is that it will take a while before it becomes widely available. In its current form, is quite expensive and only accessible in more developed countries.

It is also not a solution for everyone. To have your voice recreated by a deepfake, you need to have been recorded speaking multiple times — something called voice banking. Most people would not have had this in mind before they lose their voice. Perhaps it’s time to backup your voice, just in case?

#2. Education

Imagine you were a teacher trying to engage you students in the current history class. You are talking about the opening of the Dallas Trade Mart — at the time, one of the world’s largest trade centers — and want to make sure that your students remember the lesson well.

While an engaging lecture can be very memorable, a combination with multimedia — movies, voice recording, images — is even more so. For this purpose, perhaps it would be useful to play the speech which President John F. Kennedy was supposed to deliver at the Trade Mart back in the 1960-s.

The problem?

President Kennedy was assassinated on the way there — the speech never took place.

However, CereProc collected hundreds of voice-recordings of Kennedy, processed them & created a deepfake of his voice. 60 years later, they managed to deliver the best speech never made, in the President’s own voice:

“Reviving” historical figures for the purpose of education and entertainment is certainly not impossible! Here is another amazing example by the Dalí Museum:

The Future

Unfortunately, education and museums are not one of the most profitable businesses in existence. While this application of deepfakes is very impressive & useful, it is — in its current form — very expensive to implement.

The team which produced the deepfake of Kennedy’s voice above spent 1000+ hours of manual labor to collect & process data.

My take is that this particular application of deepfakes — at least for the foreseeable future — will be used very occasionally and for very famous historical figures only.

#3. Movie Production

When Paul Walker died during the shooting of Fast and Furious 7, the producers had to hire a very expensive FVX studio to recreate his character.

The replacement of Paul Walker costed the producers as much as $50m! That’s a lot of money.

Well, we already saw that Salvador Dali can be revived using deep fakes. The same principle can be applied for substituting actors in movies. Recently, there was a wave of people using deepfakes to add Nicolas Cage to random movies. Here’s a snippet of him playing Lois Lane in Man of Steel:

The Future:

While education is not one the most profitable businesses, the movie industry certainly is! Currently, the main obstacle stopping deepfakes from entering the movie scene is image size.

Deepfakes are currently limited to a resolution of 1024×1024 — lower than the 2048×2048 silver screen standard. However, Disney Studio Research is investing in changing this.

My take here is that we’ll very likely see a movie production using a deepfake in the next few years — for a short scene to start with. What an exciting time to live in!

The Bad & The Dirty (Top-3)

#1. Political Propaganda:

Another use of deepfakes with large potential is definitely politics. Check out this video of President Obama:

Obama speech — deepfake

Keep in mind that this video is from 2018. Since then, deepfakes have gotten a lot more realistic — as you saw at the beginning of this article.

The real danger here is that, particularly in election time, spreading misinformation about political candidates can influence election results. Let’s see how plausible this actually is in reality.

The Future

In my estimates, this particular use of deepfakes will likely not have the bad effect everyone anticipates it to have. Why?

Political material goes through the utmost degree of scrutiny in proving its authenticity — for understandable reasons. While one could get away with using a deepfake in a movie — that is me & you won’t notice immediately — political material will get checked tens of times by experts from all sorts of institutions.

My take is that even if attempts to produce unauthentic political propaganda start circulating through the web, Facebook & other tech giants would already be prepared to identify them. In fact, this is a very recent topic of research by Facebook AI.

#2. Porn:

As you could probably guess, one of the highest-demand uses of deepfakes is porn production — celebrity porn in particular. In fact, over 90% of the existing deepfakes are porn.

The degree to which those artificially-produced movies can be realistic is staggering. There are entire forums dedicated to the topic — everything from sharing productions to tips & tricks on how to make one.

The Future:

The real problem here is not so much porn of celebrities. It is porn of normal people like you and me.

Indeed, with enough pictures & motivation to embarrass you, someone could put your face on a porn actor and make a movie which seems shockingly realistic — also called revenge porn.

Whether we like it or not, porn is one of the most profitable multi media industries out there — valued at about $100b. My take is that deepfake porn is not going away anytime soon.

The good news is that currently porn videos produced by deepfakes are flagged — so people know they’re not real. At the very least, this prevents some of the reputational damage that comes with having yourself in a porn movie.

#3. Financial Fraud:

We’ve all heard of phony schemes where a scammer calls you & asks for money — to save a relative who’s supposedly been in accident, for example. Most people identify those quickly — if someone really needs you to send money, the request will not come in an informal phone call by a random number.

However, deepfakes can take such scams to an entirely new level.

Imagine that the person calling you to send money in fact does sound like your relative… This is indeed what a deepfake can do with enough voice recordings of the particular person. What is to be done then? How do you know that the person calling isn’t your relative?

In fact, in 2018, using this method scammers managed to steal 243,000$ from a UK company by calling its CEO while pretending to be the Chief Executive of the parent-company.

The Future

In my estimates, this particular use of deepfakes is the most dangerous one. It wasn’t long ago when older citizens were getting scammed frequently in phone calls by people who did not even sound like their relatives.

Can you imagine if they did? Certainly, many people would fall pray.

Of course, if such crimes become wide-spread, technology to detect fake voices will emerge. However, a good few years of scams will pass before this problem gets picked up by mass media — especially in developing countries.

My take is that we still have not seen the worst from this use of deepfakes, and we very likely will within the next decade.

The good news is that criminals are usually not very clever and there aren’t very many publicly accessible voice recordings of most people — the main ingredient for faking someone’s voice. However, this can certainly be done to someone with a large social media presence.

Conclusions

There you go — some food for thought.

Impersonating actors, reviving historical figures & recreating a person’s voice — I think what we’ve seen so far from deepfakes is a net-positive for humanity, despite the odd malicious intent here and there.

What do you think — are deepfakes a good thing or a bad thing? If you have any suggestions, critiques or feedback, please let me know if the comments bellow — it helps me tremendously!

As always;

StAI smart!

--

--

Andrey Gizdov

Computer Vision Researcher @ Weizmann AI Center | Aspiring tech-entrepreneur | Sport enthusiast