You can also watch it on YouTube.
Below is the essay version of this video. It is not a transcript as it differs in the parts where a raw transcript would just be outright nonsense or clumsy as hell.
How would you even start a video?
How do you open the first real video in a series?
This particular video took too takes before I arrived at something, so I just went with this one.
It all made sense when I was planning this.
It’s the norm today for people to rely on video as their primary source of information.
It doesn’t matter whether it’s TikTok, YouTube, or some other form of video. This is the medium of our age.
Even people who do read for pleasure or for information, they quite often discover which books they want to read through video.
So, if I want to promote my work-- and I have things to promote – it makes sense to be doing video in some form.
I’ve written books. I’d like you to know about those books in due course.
I do consulting, I’d like you to know about the consulting in due course.
I have a newsletter, I’d love for you to subscribe.
But if I’m going to reach people on video, I need to actually do video, and not procrastinate on it forever.
That’s where the tricky part came up. This is not something I’m used to doing.
So I did what I always do. I did research. I figured out how to do the typical talking-head well-lit YouTube video at a minimal expense.
I used one of the better microphones I own – you know, the one that sounds like radio. I had the entire script written out in detail in advance. I had an interesting backdrop, and I was posed standing, because standing is best for your voice.
And it was just awful. It did not work at all. It just was awkward and stilted, and I froze several times.
I seemed to find it difficult to figure out what to say, even when the text was right there in front of me.
I did what any person would do under these circumstances.
The logical next path.
I called my mother and asked her for her device.
Now, this is not quite as unusual as you’d expect, although mothers often do have good advice. But my mother in particular used to work in TV and radio before she retired, and she worked specifically as a TV reporter for the Icelandic National Broadcasting Service. Her experience is quite extensive.
And she also has, like most of us in my family, the personality quirk, where we do our info-dumps and thoughtlessly offer unfiltered and unvarnished criticism of whatever work is being done, which is fine.
She did that when I was growing up and read my school essays, which is awesome.
Because actionable advice is always fantastic.
She came up with a list of solid pointers.
I need to think about what the audience wants, not just what I want.
I had the desire to look like a professional, to look like somebody who was a pro YouTuber who had this polished look to everything he did, and that was not based on anything that the audience is likely to want.
The only reason to listen to a video by, you know, me, is for my expertise – for me to convey what’s in my head over to you so you can build on it or benefit from it.
The audience is only going to be here for my expertise, for the things that I know how to do well, that I know how to explain well.
Her first advice is that I need to drop all of the professional aesthetics and just focus on a video production setup where I could be myself.
Turns out to be a bit trickier than expected, because there’s this minor issue of crippling anxiety about being on camera and being recorded.
But it’s just something I need to cope with, hence the fact that this is something like the fifth take.
The first step is to figure out ways that let me be be more relaxed in front of the camera and the first way to be more relaxed in front of the camera, which I’m very bad at following, is to speak more slowly.
I am extremely bad at this because I automatically start to ramp up the speed of my speech as soon as I get onto something that I find interesting, because that’s the fun part!
But that is less relaxed and harder for the audience to keep up with.
More importantly than speaking slowly is to figure out ways for me to be natural.
As a reporter, my mother’s beat was science and healthcare, which required a lot of interviews of people with expertise, and finding ways of getting people with expertise to convey that knowledge in a natural manner.
That can be tricky, because academics tend to lock up and stiffen up as soon as you take them out of their environment where they’re most natural, like being in the lecture hall or their offices, or an archaeologist at the dig.
You shoot the interviewwhere they feel natural, where they’re most used to thinking about their work and their expertise. In my case, that would be my home office where I’m surrounded by work-related books, my typewriters and pens, and the photos I use to liven the place up. And, most importantly, a small statue of Gaston Lagaffe, the patron saint of absent-minded klutzes at work.
This is where I feel the most comfortable. It feels so much more normal to spout my nonsense in this environment than where I’d set up before.
Another issue she pointed out is that as soon as you start to stress about pauses and errors in your speech, things can go haywire very quickly, you basically go off the rails.
She has a theory about that.
One of the consistent results in research on retention and that is memory and comprehension is that audio books, on average, tend to score lower than printed books.
Ebooks have historically been all over the place, it depends on the device, it depends on the toolkit, the machinery that surrounds the ebook, and the sample sizes in these studies tend to be so small that it’s next to impossible to get reliable results.
But in terms of comparing printed books with audio books, audio books tend to score lower, but there’s also some variability in both.
My mother’s favourite theory on the topic is that people tend to have an easier time comprehending and retaining audio information that feels like a natural dialogue, that feels like it’s something – it might not be a back and forth between the listener and the speaker, but it feels like that. It taps into the skill sets and the experience we have at conversation.
That would mean that some voice actors, some narrators, can do this just purely through skill, through the way they perform. Even when they’re just reading the text off the page, they make it feel like they’re talking to you, like this is a personal conversation between you and them.
That, in this theory, could improve retention. But from the perspective of a video like the one I’m doing, the pauses, the hesitation, the markers of natural language and natural conversation might make this thing more accessible than the paced, stilted reading off the screen.
I mean, it’s a theory, it doesn’t feel that implausible. I guess we’ll find out.
There’s also an aesthetic to professionalism. Not just in terms of the visuals, but also in terms of the visuals, acoustics, and voice style.
There’s a pro YouTube look. There’s a pro podcast sound. None of these represent actual professionalism. They aren’t what makes a podcast professional. They aren’t what makes a video professional. They are the trappings of professionalism – the acoustics and aesthetics of professionalism.
They’re a look that can be replicated without an understanding of the underlying professionalism that drove the decisions in the first place. Professionalism is practice, and the specific look that comes with a professional video is a consequence of an expert applying their practice to deliver that video.
If you replicate that result without the expertise, without the practice, you just end up with the aesthetics, even when it’s inappropriate, even when the look of it actually makes the end result less credible.
You can see this happening in, for example, documentaries, where everything is shot like it’s a corporate training video, and it immediately has a kind of of artificiality – an implied falsehood that you don’t get when you cover the same topic using a more run-and-gun dynamic and, let’s face it, uglier aesthetic.
But that’s the professional product. The professional understands the context, and understands the role that the end result is supposed to have – the job it’s doing for the audience. The same applies here.
I don’t think a more professional aesthetic would make this video more credible. It wouldn’t make my video more credible if it was lit – pro Youtube style – with a softbox and shot with a 6K open gate camera and colour-graded to match the YouTube default. It might even make it feel a little bit fake to a lot of people, understandably, because what does a photographer know about strategy or management or machine learning or software development or world development?
The video might benefit from better lighting or a better camera, but they would need to be applied in such as way as to not compromise the credibility or authenticity of the work.
The aesthetics of the video can betray a different expertise than what you’d expect and that makes it less credible.
Authenticity and credibility matter as they connect with one of the biggest issues of any media production today: AI.
Anything that is surface-level replicable, like a specific aesthetic, a specific style, a specific look, or a specific sound, can be replicated using machine learning techniques.
They are using fairly detailed statistical modelling of collections of similar material to replicate a new material that matches a common style or form. That means that when something is more broadly represented in the data set that they’re modelling, like video, that aesthetic becomes easier to replicate.
So the actual YouTube look, the standard softbox fill light, backlight for the hair, or for the bald head, the common colour-grading,that look with the standard sound and the podcast voice, all of that can be automatically replicated, at least in short segments, especially if they’re intercut – video essay-style – with clips and images from other sources.
If you don’t need to convey expertise, if you’re just conveying the clichés and the stereotypes of the day, and want to do so in a way that’s represented using the clichés and stereotypes of video of the day, then you’re competing directly with AI. That’s what it does extremely well.
The simplistic look in the video is, turns out, genuinely harder for AI to replicate. I move my head a little bit too much, it’s a problem when I use proper microphones and with those it makes my voice go in and out. That’s why I’ve got the lavalier microphone set up.
I use my hands too much. AI has problems with fingers for a reason, because they don’t actually model the anatomy of their subjects. They model the 2D plane that’s represented on the screen. So they don’t recognise the physical reality of what they’re replicating.
Sometimes the lo-fi effort – the imperfect and personal – is actually stronger, more convincing and harder to automatically replicate in the era of AI, is the first step to doing a project like this.
If you’re going to do video, don’t do whatever it is that AI can do easily.
It’s the same thing with writing, if you’re going to be writing today, avoid the twee LinkedIn-style professional and shallow writing style that large language models are so great at.
Avoid the mundane, the typical, the traditional structure. Be a little bit chaotic, but be yourself.
People will forgive chaos and a little bit of disorder if there’s a person behind it.
No matter what you think about TikTok, new social media, the generation today, or whatever age-related bugbear you might have – whether it’s against older people or younger people – you need to bear in mind that everybody’s doing this to connect with other people.
That’s literally what TikTok is for. That’s literally what social media is for, even YouTube. People are doing social media because they want a human connection. Even when they follow celebrities, that’s because they want to connect with them personally.
Nobody is doing any of this to follow a robot.
It doesn’t matter how polished the robot is, it doesn’t matter how polished the videos are that the robot makes, it’ll never be a person, and that’s what the audience wants. They want people.
This is my plan: I’m going do the whole personal thing and be a little bit… Error-prone. I’m going be trying a few different approaches.
The next video is not gonna be as meta as this one. I might touch up on the process, because it’s interesting when you’re starting but.
But this is the plan.
If my accent is a bit of a mess, and that’s because none of it is natural. I was born in Iceland, partially raised in Lancashire, in England, and my original English accent as a child was very Lancaster, and quite indecipherable.
When I moved back home to Iceland as a slightly older kid, I tried to get rid of the accent. So, I figured out how to speak English with an Icelandic accent when I want to, because that can be a useful talent. (There was a period I didn’t mention in the video where I spoke with an American accent, but that didn’t last long.)
But then when I moved to Bristol, 25 years ago, the Icelandic accent came to be a little embarrassing, because one of my friends said it sounded like I was Dutch.
“No, can’t have that.”
I began to consciously (and unconsciously) pick up on the accents of the people around me. The problem there is that this was at a university with a mixed bag of people from all over the UK, so my accent is all over the place.
You might have the occasional word that’s weirdly North English, and then you’ve got Bristolian, you’ve got South English, and an older mixture of things, and then the occasional Icelandic pronunciation slip up.
If you find the chaotic accent off-putting, then I’m sorry, there’s nothing I can do about it.
It’s just the result of literally more than 45 years of messy language.
Most people have noticed the… shenanigans that are affecting international trade at the moment.
But people might not necessarily appreciate how this affects small economies like Iceland.
Or, specifically, how this affects smaller currencies.
We’re still using our own Icelandic Krona. It’s not the most stable in the world, mostly because of the small size of our economy: it can easily be disrupted by changes in currency flows that would be considered minor in other countries.
Up until now, pricing in US Dollars has been the safest route to take for somebody like me, who is selling digital goods and services to both the US and Europe.
But this rubric has changed over the past couple of months. The dollar is more volatile and, because of how US politics are affecting international trade and tourism, there’s a very real risk that it might weaken even further in the future. Because the krona is such a small currency, its relationship with the USD is likely to be even more volatile than most other currencies.
US economic chaos also gives European businesses an incentive to migrate away from US-based services.
I’m not quite ready to do that with my payment provider, but an essential first step is to switch my pricing over to Euro.
This is effectively a price increase, my first since I started publishing.
But to offset that somewhat, I’m doing two things.
The first is that I’m switching to tax-inclusive pricing, which is a bit unfamiliar to many Americans but is the norm in Europe (indeed, it’s a legal requirement for consumer pricing). That means that even though a straight-up switch from $35 USD to €35 EUR could at the moment be effectively a 10% price hike, if you’re in a country that has a %10 VAT on ebooks, the switch is effectively a wash: the effective price is roughly the same.
But, to ease the transition, especially for those in countries with low to no sales tax or VAT on ebooks which might be hit more hard by this price change, I’m also offering a €5 EUR discount until the 19th of March.
Just use this discount code at checkout:
JACKPOTBEGINS
Or, use these direct purchase links that have the discount already included:
So, one of my biggest regrets over the past few months is the degree to which I missed the fact that the “mainstream” “AI” movement is a fascist project.
Sorry, let me step back a bit instead of jumping right in.
When I first began to look into Large Language Models (LLMs) and Diffusion Model back in 2022, it quickly became obvious that much of the rhetoric around LLMs was… weird.
Or, if we’re being plain-spoken, much of what the executives and engineers at the organisations making these systems were saying was outright weirdo cult shit, much of it horrendously racist and offensive.
After talking it through with the editor who had volunteered to help me out with The Intelligence Illusion I decided to de-emphasise in the book the fact that much of the field of “AI” is utterly detached from reality in ways that are simultaneously disturbing and offensive and focus instead on the ways the technology was risky for business users.
In her words: “these people are so unhinged that if you try to point it out, you’ll come off as an irrational conspiracy theorist. You’ll have more credibility if you don’t mention it.”
Even without mentioning their deluded worldview, I still managed to underestimate just how dysfunctional the tech industry is and just how little regard “AI” vendors have for actual productive outcomes for their software.
I also underestimated the outright dishonesty of many in the management class.
I had thought it would be obvious from the sheer cavalcade of risks presented in the first edition of the book that it was not suitable for use in business, not realising that many people would use it as a guide to sell “AI” to businesses:
They used my list of concrete business risks as a way to look “reasonable” and “realistic” to help sell “AI” to organisations.
This is why I published a second edition of the book. I needed to both correct my underestimation of the risks and dysfunctions of the industry and I needed to make it absolutely clear that deploying these systems in your business will harm it. You can mitigate the harm, but that’s like deliberately taking both the poison and an antidote. The poison has no benefit, so the sensible thing to do would have been to skip both and not poison yourself in the first place.
But the political and religious dimension to “AI” still remained unaddressed – for the most part – simply because it is neither my area of expertise nor the focus of the book. Providing a solid business rationale for avoiding the tech is my contribution, but others have been doing the hard work of outlining the ideological and political dimensions to “AI”.
There are others doing similar work, much of it led by The Distributed AI Research Institute – founded by Timnit Gebru – and its projects.
(Just because I haven’t mentioned others here doesn’t mean I’m not aware of their contributions or importance. Emily M. Bender mentions a few of them in this post.)
This picture of the field of “AI” is not a pretty one.
The “AI” field as a whole has bought into racist ideas on intelligence born out of eugenics:
To make matters worse, “Sparks of Artificial General Intelligence” is not the only AI paper to cite Linda Gottfredson’s work. After the news regarding Microsoft’s paper broke, I discovered that Shane Legg, co-founder of DeepMind (now part of Google) also cited her definition of intelligence in the introduction to his 2008 PhD thesis called “Machine Super Intelligence”. In it, he refers to it as “an especially interesting definition as it was given as part of a group statement signed by 52 experts in the field”. The controversy regarding Gottfredson’s racism and those of these “experts” was well established by 2008 as the statement was originally released in defense of Charles Murray’s 1994 book The Bell Curve. Murray’s book was immediately accused of supporting racist political policies upon its release and these accusations have followed the book into the 21st century… As such, I again have no clue how this oversight managed to get into a released paper.
AGI Researchers Stop Quoting White Supremacists Challenge (Impossible)
Now that US tech is firmly in bed with an authoritarian US administration, I think we do have a clue how this oversight happened. Or, at least, it doesn’t take much for us to imagine how.
Eugenics has always been a constant facet of fascism and Nazism specifically:
Negative eugenics is what justified restrictive immigration and anti-miscegenation laws throughout the twentieth century, as well as the forced sterilization programs implemented in states such as California. California’s eugenics program, which started in 1909, was subsequently adopted by the Nazis as a template for the “racial hygiene” policies that ultimately led to the Holocaust (Black, 2003; Stern, et al., 2017).
And…
In fact, leading figures in the TESCREAL community have approvingly cited, or expressed support for, the work of Charles Murray, known for his scientific racism, and worried about “dysgenic” pressures (the opposite of “eugenic”) (see Torres, 2023a). Bostrom himself identifies “‘dysgenic’ pressures” as one possible existential risk in his 2002 paper, alongside nuclear war and a superintelligence takeover. He wrote: “Currently it seems that there is a negative correlation in some places between intellectual achievement and fertility. If such selection were to operate over a long period of time, we might evolve into a less brainy but more fertile species, homo philoprogenitus (‘lover of many offspring’)” (Bostrom, 2002). More recently, Yudkowsky tweeted about IQs apparently dropping in Norway, although he added that the “effect appears within families, so it’s not due to immigration or dysgenic reproduction” — i.e., less intelligent foreigners immigrating to Norway or individuals with lower “intelligence” having more children [59].
The TESCREAL bundle: Eugenics and the promise of utopia through artificial general intelligence
These beliefs all focus on the idea that you can “rationally” attain an ideal future outcome for the selected inheritors of humanity by both making horrendous sacrifices in the present and by investing in those who are inherently superior and are destined to inherit.
Coupled with the inherent white supremacy of the eugenics movement, it amounts to accepting mass death in the present to make a better world for white people.
For example, in longtermism this white supremacy seems to be baked into the very foundation:
These theories treat as less urgent those anthropogenic hazards that won’t snuff out humanity altogether, and the theories’ adherents place the currently intensifying human-caused climate crisis squarely in this category, encouraging us to regard as morally less important the suffering and death it is occasioning. The harms in question are falling in dramatically lopsided fashion on racialised and Indigenous groups the world over, groups whose very vulnerability to these harms is a product of long histories of injustice. Such theory-induced callousness to losses and damages visited grossly unequally on racialised people licenses talk of a racist strain in longtermist thinking, and individual longtermists deepen this strain in specific ways. A well-placed young longtermist once argued that inhabitants of rich countries are generally more ‘innovative’ and ‘economically productive’ and that saving their lives is hence substantially more important for humanity’s future than saving lives in poor countries.
Another strand of the bundle are ideologies that involve a future escape from the limitations of the physical body and the material world using technology, Singularitarianism being the most popular expression of the idea:
Singularitarianism is just the idea that the “Singularity” — the moment when the pace of technological development exceeds our comprehension, perhaps driven by an “intelligence explosion” of self-improving AI — will play an integral role in bringing about the techno-utopian future mentioned above, plus a state of radical, post-scarcity abundance. In one popular version, the Singularity enables our posthuman digital descendants to colonize and “wake up” the universe. “The ‘dumb’ matter and mechanisms of the universe will be transformed into exquisitely sublime forms of intelligence,” writes TESCREAList Ray Kurzweil, a research scientist at Google who was personally hired by Larry Page, the company’s co-founder and an adherent of a version of TESCREALism called “digital utopianism.”
These ideas all offer some form of salvation, either from the limitations of our physical bodies or from the limitations of the material world. These are outright mystical and religious ideas that have no grounding in science or engineering.
Separately, the TESCREAL bundle and the algorithmic authoritarianism paint a picture of a divided tech industry, one part taken over by cults, the other by authoritarian oligarchs.
The political side is obvious. They believe that “AI” systems can keep society around them functional in the face of the massive “population losses” that are inevitable if their policies get implemented.
But, if you instead look at them together as a single movement, they bear a striking resemblance to an earlier 20th century political ideology that also mixed together pseudoscience, strange mystical ideas, and horrendous authoritarianism.
The work these researchers have done make it clear that the political authoritarianism and the cult-like religious zealotry are two sides to the same coin.
And it’s a coin with deep historical roots:
Fascism. Or, specifically Esoteric Neo-Nazism.
One of the reasons why the somewhat conflicting ideologies of the TESCREAL bundle and the ongoing fascist project in the US work so well together is that TESCREAL itself is just the latest iteration of of the mysticism-bound Neo-Nazi political movements that survived the fall of Nazi Germany.
A common facet to TESCREAL is that they take mystical ideas from the Neo-Nazi movement and flip it from being about a mythical past to a mythical future.
These “bonkers” ideologies are integral to the fascist project as a rationale for atrocities and destruction. They are a belief system that promises a bright future to the selected people and provides them with a systemic rationale for letting mass death happen as “AI” will replace the workers.
I believe that the current “AI” bubble is an outright Neo-Nazi project that cannot be separated from the thugs and fascists that seem to be taking over the US and indivisible from the 21st century iteration of Esoteric Neo-Nazi mysticism that is the TESCREAL bundle of ideologies.
If that is true, then there is simply no scope for fair or ethical use of these systems.