The Great Inversion
When Trump won his first term against Hillary, it was a huge shock to the nation.
But what it revealed was an underbelly of dissatisfied voters who had been left behind. Those who weren’t living in big cities, didn’t have college degrees, but still had to pay high prices for everything while their wages stagnated.
That was a slow and long pressure buildup over an extended period of time. The eight Obama years before that, and even the Bush years before that, and then it reached its tipping point.
Now, we are experiencing a similar shock from yesterday’s results.
What does this one reveal?
Not a long, slow buildup. But a recent, rapidly rising, fear.
Uncertainty about the future. Financial insecurity. ‘Will I have a job 2 years from now?’
Voters are watching AI advance at a beyond alarming pace, and they’re not so sure they’re going to be included or accounted for.
They could just be left out to rot.
And that fear, that’s what will cause this type of landslide.
It’s not about the gas prices. That’s just the material thing that people can point to, because they don’t have 30 minutes to fully express all their frustrations and concerns and worries.
But what did they vote for?
Democrat, yes.
But not progressive policies. There’s no Universal Healthcare, or Green New Deal.
They voted for a slowdown. For oversight. For accountability. For constraint.
And so, it is in this age of AI, that the Democratic party becomes the conservative party, and the Republican party takes on the progressive role.
‘Progressive’ used to mean challenging norms and trying new types of civil rights and personal expression against a conservative voter base that held to tradition and didn’t want things to change.
But where are the norms predominantly challenged today? Where’s the experimentation? AI
It’s the AI frontier explorers who are the new progressives, and they now fall on the Republican side.
And it’s the ‘let’s have oversight, guardrails, limits’ crowd who are now the new conservatives.
This is the great flippening.
The Great Inversion.
The two parties have inverted once before, in the 1960s.
But that took over two decades to play out, and there was no AI in the room.
Now there’s actually something real to disagree about.
The Great Divide
As we know, anything can get politicized.
Something as obvious as identifying yourself in order to vote became so politicized that ~60% of CA voters voted against it, which is mind-boggling. But it’s tribal at that point. You’re either against voter identification or you’re pro-Trump.
Isn’t it funny, by the way, that no one during this election was talking about banning guns or abortion rights? Even something that was politicized for decades can be dropped by the wayside for the next hot topic to prove membership to the tribe.
The next hot topic is already a hot topic: AI
We are almost sure to see AI become politicized, with the Dems wanting to tap the brakes and Repubs taking the gas pedal side.
These will become new tribal identities that will divide us yet again, just in a brand new way so it’ll feel different this time, yet again.
And what will the AI companies do?
They’ll be on team gas pedal, of course, that’s where the profits are (all the while declaring publicly and loudly that they are really in favor of brakes. Very cool story.)
To state the obvious problem, this is extremely dangerous. AI companies all being on one side of the political aisle, with the amount of data and user influence they have, is not a democracy; it’s a conclusion.
So how do we solve for this?
We separate the software (AI models) from the hardware (data centers).
This is not unprecedented – in the 1930s, the Glass-Steagall Act created a clean separation between commercial banking and investment banking, in order to stabilize the banking industry at the time.
Similarly, we can say, you either operate a ton of GPUs, or you are an AI R&D lab, but you can’t do both.
The data centers, with large scale economies, might even become regulated like utilities, ensuring that power does not get centralized in a few individuals or entities.
The AI R&D labs are the more laissez-faire side, and they can use any data center. We would expect many competing models, as this side does not require owning the capital.
This model also creates a natural control system. If the AI model is malicious, the power to shut it down belongs to a different entity from the one who wrote the malicious AI.
And it allows for oversight by the data center as to how the AI model is using and handling a person’s data (without seeing user data), making sure Fourth Amendment rights are protected (idea inspired from Dan Carlin).
There is a ton that would still need to be ironed out. This is just a late free-form writeup after some shocking election results.
The Great Risk
If the data centers are heavily regulated, that indirectly puts the government in the authority role to tell AI models what they can say or not say.
The structure must be set up in such a way that this does not occur. Just like power utility companies are forbidden from discriminatory pricing, the data center utilities must be strictly adherent to the First Amendment, not to some safety oversight committee.
Rules and regulations around AI safety should be enforced at the AI model level, not at the bare metal of the GPUs they run on.
The Great Opportunity
This is how we sidestep at least one large landmine as we navigate through the singularity: the tendency toward centralization of power within AI conglomerates.
And hopefully also create room for political compromise, as the Democrats can limit total data center capacity to any one model, and have oversight, while the Republicans can still have free and competitive markets on the AI R&D side with an even playing field.
There is much more that can be architected here, but it’s not my intent to be prescriptive. This is a skeleton for others to build on.