I'm in a writing mood today so here's the entry on the broader implications of AI.

This is a big subject. I've given smattering of my thoughts before, but never synthesized them into 1 post. My thoughts are also still of yet incomplete. I believe no one can possibly have a complete picture, and indeed some things are simply unknowable until they happen.

I will breakdown my thoughts in 5 sections. First I will discuss the technology itself. Then I will discuss the economic implications, followed by the social and philosophical implications. Then I will follow with some future speculation. Lastly I will confirm or dispel some of the prevailing notions about AI.

Technology

I won't go in depth here, I'm not an AI expert. AI is based on some of the same structures found in the brain, hence the term "neural network".

Real nerds have recognized since the early 2010s that AI was going to be big, big big, unlike other tech trends like VR or blockchain. I think the tech itself is incredibly interesting. The math and science behind it just has that cool factor. The whole concept of training and inference is so clever and unintuitive, despite being literally in our brains.

The advances we've made in the last decade, in terms of hardware, computational methods, and high-level logic analytics, have been leaps and bounds. However, I think they may be limited or plateauing from this point on.

First, the hardware. Moore's law has already stopped scaling a decade ago. Minimum "fine" feature size (note this is different from advertised "node size" and minimum feature size) has already hit <10nm as of 2026. We are at or near the limits of physics. We're unlikely to get more than 1 order of magnitude improvement in terms of hardware compute out of traditional silicon based computers.

We're also fundamentally limited to print CPUs in 2D (with essentially constant number of layers). Brains are 3D.

There are some candidates to break out of this plateau. I would consider graphene-based CPUs, photonics, analog compute, quantum assisted compute, and grown computers (lol) to be some of the most notable directions of advancement, although I can't speculate on how important each will be.

Second, the software. I'm not knowledgeable as to what the frontier labs are cooking up. I've seen differing opinions on whether there are still major software breakthroughs possible in this field, or if it will be marginal gains from this point on.

I am fairly certain though, that available "free" training data has been exhausted.

Overall I think there's at least 1 order of magnitude improvement we can still gain out of compute factors, and another 2 order of magnitude (multiplicative) out of software efficacy -- pre and post training -- in the next decade or so. How much of that will actually translate to functional improvement is debatable.

Asymptotic self improving AI is possible, but singularity is a pipe dream. In real space, we have yet to create self-replicating machines. This is because replication requires constant energy and material input. Self replicating AI will require constant energy input, and constant data input. I have some confidence in my intuition, that as long as data ingress is limited, which it is, the singularity will not be achieved.

That is not to put a cap on it though. The human species itself can be considered a self-replicating, self-learning singularity, limited only by our inability to leave our planet. Maybe one day we will create a singularity that involves AI. But this is all getting too philosophical instead of technical, so let's move on.

Economic

AI is a tool. A naive reading of tools is that all tools have the ability to do at least nothing. It is only up to us what we use that tool for.

AI is useful for things like protein folding, image analysis, information synthesis, coding etc. It has the potential to be a great economic boon. It also has negative uses like hacking, AI scams, forgery etc. In this section I will weigh the positives and negatives from an economic perspective.

But before that let me just consideration the stock market real quick. "Is there a AI driven bubble in the stock market?" Yes. Yes there is. I think this is so apparent there is no need for me to provide further analysis.

As discussed in my last entry, I expect AI to replace a large proportion of white collar jobs, especially mine (software engineering). This is just like how automation replaced factory jobs. The final result is that there is more output for less input. The problem arises from 1) the inability of a human to change fast enough to adapt in their lifespan, and 2) wealth distribution.

Automation has replaced a huge proportion of manufacturing jobs. But it has happened over 100 years, giving time to adapt. The internet exploded over 10 years, but it has replaced comparatively few jobs (at least skilled jobs), instead mostly just adding jobs. AI, however, has both the depth of penetration of automation, AND the speed of the internet.

This speed, depth, and breadth of change can't be handled organically. The government must step in. Fortunately, neoliberal structure is well equipped to handle this. Unfortunately, there seem to be idiots in government all over the world.

Fiscal and monetary policy

I have break this down to fundamentals. AI as a tool has shifted the productivity further in favor of capital, as opposed to labor. The more productive capital becomes, the more important it becomes to distribute it publicly. This is because of the principle of marginal utility. As an absurd example, money (as a conceptual unit of value) is far more valuable when evenly distributed to all 8 billion people, as opposed to concentrated in the hands of 1 person. I know this sounds communist (it kind of is), but it's not incompatible with neoliberal policy.

The first neoliberal resolution to the overproductive capital is to use a sovereign fund. While sovereign funds are not inherently neoliberal, it is compatible. Just as example figures, let's say that the sovereign fund owns 15% of all public companies (public capitalization out value private capitalization by several factors, in the US). In addition, all companies have their profits taxed by 15%. Dividends, buybacks, and profit taxes are directly redistributed to each adult citizen. In an overproductive capital scenario, at least (but probably a lot more) $900+ billion will be redistributed to 270 million adults in the US. That's about $3300 per person.

The second neoliberal resolution to the overproductive capital is to simply print money. Overproductive capital simultaneously increases productivity, and drives down the cost of labor, which lowers wages, which creates deflation. The government can print money equivalent to the amount necessary to achieve healthy inflation, and directly distribute it to citizens. Let's say deflation is 0.5% and target inflation is 2%. Inflation is roughly correlated with M2 increase, and M2 is roughly correlated with GDP. "2.5%" inflation gap can roughly be filled by printing money roughly equal to 2.5% of GDP, which is about $2300 per US adult.

The third neoliberal resolution to the overproductive capital is estate tax. This is functionally equivalent to a wealth tax, but when a person is dead they're much less likely to fight it. Wealth taxes are not neoliberal, unless the person you're collecting it from is dead, thus absolved of their relationship to this wealth. If loopholes are closed, up to 15% of all accumulated wealth can be collected over the course of an average lifespan. Going by US death rates, and accounting for deductions, this could be around 0.1% of all wealth in the US, which is about $600.

All together that's about $6200 in top line "UBI". Remember this is not meant to be a "living UBI". This is merely the supplement to suppressed labor wages due to overproductive capital. Another great thing about these levers is that they auto adjust according to the relative value of labor and capital.

Another way to think about this, is what if capital accounted for 100% of GDP (impossible and absurd)? For ease of calculation let's just say this capital is in the form of a company that prints useful money equivalent to GDP. 100% of that money is profit (this is a valid economic equivalence I'm pretty sure). 15% of gets taxed. 15% of it is owned by the government. This does not compound, it's additive, 30% gets redistributed. The company gets 15% of it taxed after 75 years, and gets sold back to itself at a 15x profit multiplier. This is equal to 7.5% of GDP per year. Plus 2% printed money. In total 39.5% of output gets redistributed to the people.

People policy

So what happens to people who suddenly find that their skills are useless? Tough luck I must say. I don't really have good solution to this one, other than artificially handicap AI development speed.

The natural resolution is that skilled people are usually skilled because they have some innate skill like smart or strong. And if they have that innate skill, they'll pick up something else relatively quickly. The problem is that these days we all have to have a 4 year degree to be a white collar worker, and spending another 4 years in school is a bit impractical for an adult who has 40 years of working age.

Perhaps we'll just have to accept that times change, and we lost the roll of the dice.

It's kind of late, I'll have to do a part 2 another day.