Don’t worry about deepfakes
The hysteria around deepfakes has been overblown
Unless you have been living under a rock, you have surely heard about deepfakes. You also may think that they are a “threat to democracy”, or a “powerful [new tool] for misinformation”, and you may be right about those things. But I’m here to tell you the introduction of deepfakes doesn’t bring us into uncharted waters, it brings us back to something that closely resembles the first 75,000 years of human existence.
Deepfakes are videos created by AI that resemble a real person. In essence, the goal is to make someone believe that a particular person said or did a particular thing. Here is a deepfake of Obama.
And for something more current, here is a video of Zelenskyy telling troops to surrender. It’s wild stuff.
70,000 BC — 10,000 BC
We have lived in a world of low information verifiability for almost the entirety of human existence. Prior to the 20th century, the only way you could be objectively sure something was true was if you saw it with your own eyes. But constant self-verification leads to societal inefficiencies that would have kept us in the stone age.
Of course, we don’t only believe things we see ourselves; we trust others to help us update our worldview. You may not have seen something personally, but you trust your spouse is telling you the truth. Friends and family are phenomenal transmitters of truth because in theory they have no reason to lie to you. Which is why lying to one’s spouse or family member is such a big deal; a core tenant of trust transmission has been broken.
This idea of trust holds society together. If everyone had to verify all facts themselves, nothing would work. And in small tribes, trust among friends and families is enough, but a problem arose as we shifted to a sedentary agrarian society. Having trust among a family or groups of families is one thing but forming trust across an entire city or nation is an entirely different beast.
To quote Bowling Alone by Robert Putnam (bolding my own):
A society characterized by generalized reciprocity is more efficient than a distrustful society, for the same reason that money is more efficient than barter. If we don’t have to balance every exchange instantly, we can get a lot more accomplished. Trustworthiness lubricates social life.
10,000 BC — 1900 AD
As we started planting barley and arguing about property rights, we required larger institutions to create societal trust. In other words, we needed to scale our trust factors because we had greatly exceeded Dunbar’s number. Enter: governments and organized religion.
One could say that hunter gatherer societies were already religious (this is unproven, but religious “practices” have been traced back to 100,000 BC ) — meaning that we didn’t invent religion right when we started farming. While this is true, what I am talking about is organized religion, we have always been spiritual animals, but it didn’t form a core tenant of trust until we began farming.
Below is a chart where I qualitatively map our truth transmitters on their trustlessness vs their scalability level. Of course, religion and government slide up and down depending on the time and place, with religion probably being above everything during some periods (usually inversely related to government institutions, AKA the dark ages).
It’s worth discussing this chart for a moment, starting with scalability. Peers, religion, and government are each more scalable than the previous. I think eyes and peers are self-explanatory on this chart, but why is religion less scalable than government? This chart is from a modern perspective, because organized religion used to be much more scalable than government, especially after the fall of the Roman Empire. However, religion has a tendency to break apart when it gets too large, if you want evidence of this, just look at the dozens of major Christian denominations. Governments historically also fracture if they get too large, but in modern times governments have been able to scale very successfully, mostly due to advancements in communication and transportation.
Ok now the y-axis. What is trustlessness and is it better if something is “more trustless”? This scale is how likely you are to believe something coming from one of these groups. Additionally, it is how likely one is to believe a member of each respective group, only based on their membership to that group. For example, a plausible order of believability would be yourself > friends & family > member of your church > member of your city/nation.
You will also notice that “one’s own eyes” are a different color (pun intended) than the rest of the methods on the chart. The reason for this is that seeing something yourself was the only way one could objectively verify the truth, without relying on any other party for verification.
Said another way, anything in red on the chart represents third parties that you must trust to verify the truth. We did not have anything that was scalable, trustless, AND objective, that sits right in the “sweet spot” (again caveat for some religious devotees and ultra-nationalists).
Some might wonder about the impact of the printing press. The printing press greatly allowed the ability of god and gov to scale, but it didn’t help much as far as trustlessness goes. I am no more likely to trust my bible than I am my priest, in fact I am probably less likely.
I swear this ties back to deepfakes.
1900 AD — 1990 AD
The world changed at the turn of the 18th century with an invention that would alter the way humans transmit truth for more than one hundred years: the camera.
For the first time, these inventions provided us something that is arguably just as objective as seeing something with our own eyes!
But the real value of photos and videos came in their transmissibility and scalability. We were now able to transmit objective truth through a medium that was not the word of some third party.
1990–2022
But objective truth transmissibility be damned, 1990 marked the beginning of the end of the trustlessness of photographs: Photoshop 1.0 was released in February of that year.
Photos were always doctorable to some degree (which is why I never put them quite as high as self-verification), but the introduction of photoshop made this process orders of magnitude easier. I was never alive before the invention of photoshop, so I don’t really know what it was like to look at a picture without skepticism, particularly digital photos. I have never been able to look at a digital photo, without knowing the source, and know with a high degree of confidence that it is accurately representing reality. These days we have fallen back to a new type of third party that hasn’t been mentioned yet, corporations (sometimes called fact checkers), to verify the truthfulness of something.
Despite this, I was always able to believe a video. It’s almost impossible for an average person to modify (or fake) a video on their own. The amount of time it would take, to a degree that wouldn’t be obviously noticeable to someone, made it such that only the blurriest of videos were fakable.
But as I was writing this paper, I noticed a video trending of Tom Cruise, that looked as real as anything I have ever seen. Yet, it wasn’t.
2022 AD to ?
Alas, we are now back to where we started, with the only way for us to verify something without seeing it ourselves is to trust someone else. However, I don’t think this is necessarily a bad thing. We were able to make it over 75,000 years like this, we just got spoiled after 100 years with something that was a historical anomaly.
But if technology has shown us anything, it’s that there is a constant tug of war between different innovations. One innovation allows you to preserve an image and pass it around (photography & video), while another allows you to fake it (photoshop & AI).
It’s reasonable then to believe that technology will again provide us with something that gets us back to the sweet spot of scalable trustlessness. Or perhaps that thing was already invented, by a Japanese man sitting at his desk on a cold night in October, writing about how we should go about chaining blocks together. Whatever that means.