Typically, there are three big concerns that we talk about:
- The worry that terrorists will use AI to create doomsday viruses
- Worries about job displacement, human obsolescence, and economic dislocation
- The worry that superintelligent AI is a new dominant species that will disempower and possibly destroy humanity
But if the industry really does become dominated by a few giant companies, we have a fourth big thing to worry about — extreme inequality. If AI’s economic benefits are highly concentrated, we could end up with a comparatively small number of people controlling most of the purchasing power in our economy. In the extreme scenario, this could lead to a small number of people holding all the power in the world.
Something I had never considered. I use only ChatGPT. Does anyone have any experience using Claude vs ChatGPT vs Meta? Any qualitative difference?
Short video here: https://x.com/ianbremmer/status/2047745979479146516?s=20
Full video here: Ian Bremmer: Who Is Actually Running the World?
Ian Bremmer: Geopolitical Fragmentation and the Rise of “Techno-Polarity”
I. Executive Summary
The core thesis posits that the post-Cold War global order, characterized by American hegemony and integrated trade, has reached its termination point, yielding to a G-Zero world. This environment is defined by a vacuum of global leadership where neither the United States nor any other sovereign entity possesses the will or capacity to drive a collective agenda. The traditional “Pax Americana” has fragmented into three distinct, non-overlapping sectors: a security-focused global order dominated by the U.S. and its allies; an economic order characterized by deep U.S.-China interdependence and emerging protectionism; and a burgeoning digital order where technology companies are evolving into sovereign actors.
Ian Bremmer argues that in Western democracies, the state has lost its monopoly on governance to algorithm-driven platforms that intermediatize human interaction and social stability. Unlike China, where the state maintains a stranglehold on data and AI development, Western technology firms now exercise “techno-polarity,” functioning as geopolitical actors that determine national security outcomes and social cohesion without electoral accountability. This shift is accelerated by a “political revolution” within the United States, where systemic distrust has transformed domestic politics into a grievance-based struggle for institutional control, effectively rendering the U.S. a “divided state” rather than a unified global leader.
The “Strategic Thinking” framework presented emphasizes two critical phases: the objective assessment of the “terrain” (the factual environment) before reaction, and the deliberate allocation of “time” as the primary scarce resource. Bremmer concludes that the survival of global stability depends on whether fragmented powers can establish “guardrails” for transformative technologies—specifically AI and biotechnology—before a systemic “crisis” necessitates forced cooperation.
II. Insight Bullets
- The G-Zero Reality: We have transitioned from G7/G20 leadership to a state where no country or group of countries has the influence to direct the global agenda, leading to increased volatility.
- Techno-Polarity: In the West, technology companies have become sovereign actors that write their own regulations via the algorithms they deploy in real-time, often superseding state influence.
- US Internal Revolution: The election of Donald Trump and the rise of figures like Bernie Sanders are symptoms of a systemic belief among the populace that democratic institutions are “sclerotic” and broken.
- Grievance-Based Politics: US elections have shifted from policy debates to grievance-based contests where political adversaries are viewed as existential enemies, eroding checks and balances.
- Social Media as a Border Guard: The US government is increasingly mirroring authoritarian states by requiring 5-year social media histories for visa applicants, a move verified by Customs and Border Protection (2025).
- Reverse Technology Transfer: European nations are now conditioning Chinese investment on technology transfers to Europe (specifically in EV and battery tech), reversing the historical flow of intellectual property.
- Algorithm Intermediation: Human beings are becoming “hybrid entities” whose social, commercial, and political interactions are entirely filtered through proprietary AI models.
- Information Tribalism: The fracturing of shared information ecosystems prevents “Step 1” strategic thinking (understanding the terrain), as different demographics cannot agree on basic facts.
- The “U-Curve” of Openness: Bremmer’s original “J-Curve” theory (openness leads to stability) has shifted to a “U-Curve,” where authoritarian states (China) use technology to consolidate power more effectively than democracies.
- Australian Digital Guardrails: Australia has implemented a mandatory minimum age of 16 for social media accounts to protect adolescent mental health, as documented by the eSafety Commissioner (2025).
- Resource Randomness: Patriotism and religion are viewed as “random” accidents of birth; strategic thinkers must separate personal identity from these variables to assess global shifts objectively.
- Time as the Final Equalizer: Regardless of wealth or power, time remains the only immutable constraint; strategic success is determined by how one allocates time relative to long-term trends.
- The Disrespect Variable: Russia’s geopolitical actions are driven less by economic gain and more by a “sense of lost empire” and perceived disrespect from the West post-1991.
- AI Arms Control Gap: Unlike the Cold War’s nuclear “hotline,” there currently exists no global framework or “hotline” for AI or bioweapon containment between the US and China.
- Geopolitical Buddhism: Effective strategy requires a clinical detachment from one’s own ideas; holding onto an idea as “part of yourself” prevents the necessary pivot when the environment changes.
IV. Actionable Protocol (Prioritized)
High Confidence Tier (Level A/B Policy Evidence)
- Digital Age-Gating: Implement strict digital boundaries for minors. The Australian “Online Safety Amendment” serves as a verified template for reducing algorithmic exposure in the under-16 demographic to mitigate mental health decline UNICEF (2025).
- Sovereign Data Auditing: Entrepreneurs and institutions must treat social media history as a permanent public record. Mandatory disclosure for international transit is now a formalized US entry protocol CBP (2025).
Experimental Tier (Strategic Framework)
- Terrain-First Assessment: Adopt Bremmer’s “Step 1” protocol: Suspend all “good/bad” judgments until the terrain is mapped using divergent data sources to counteract algorithmic bias.
- Localized Resilience: In a G-Zero world, rely on local “city-state” dynamics (e.g., the “New York effect”) rather than national stability, as urban centers often maintain higher social cohesion and economic integration than the states that house them.
Red Flag Zone (Safety Data Absent)
- Unregulated AI Proliferation: Claims that “AI will eventually regulate itself” lack empirical safety data. The current absence of US-China AI arms control suggests a high-risk window for “black swan” bioweapon or economic marketplace events.
- Institutional “Memory” Bias: Relying on the stability of 20th-century institutions (UN, IMF) as a guarantee for future stability is dangerous; these entities are currently in a “cycle of decay” and lack the enforcement mechanisms for 21st-century techno-threats.
You have to wonder… how long before curious high school biology students start developing novel viruses while playing around after school in the lab …
Source: https://x.com/heygurisingh/status/2047736841437348326?s=20
and in another news story …
“That night in the scientist’s home office, the chatbot explained how to modify an infamous pathogen in a lab so that it would resist known treatments. Worse, the bot described in vivid detail how to release the superbug, identifying a security lapse in a large public transit system … The bot outlined a plan to maximize casualties and minimize the chances of being caught.” NYT (Gift Article): A.I. Bots Told Scientists How to Make Biological Weapons.
2 Likes
With the help of AIs, they could maximize the viral danger. It is a matter of optimization, lethality must be high but not too much, whereas infectivity must be high, the incubation time should be long enough, with host tissue tropism, mutation rate, immune evasion being other factors to optimize.
It turns out that SARSCOV2 was close to this creepy optimization, had the IFR (infection fatality rate) been higher, it could have meant total social disruption.
1 Like
The pushback we are seeing in San Francisco:
I define AI populism as a worldview in which AI is viewed not only as a normal technology but as an elite political project to be resisted. It regards AI as a thing manufactured by out-of-touch billionaires and pushed onto an unwilling public to achieve sinister aims like “capitalist efficiency” (layoffs) and “population management” (surveillance). AI populists don’t really care whether ChatGPT is personally useful, or if Waymos eke out some safety gains: AI’s utility as a tool is immaterial relative to the unwelcome societal change it represents.
Among the public, AI populism shows up as individual attempts to block AI encroachment; for example, data center NIMBYism, AI witchhunts among creatives, and in the extreme, assassination attempts like what happened to Altman this week.
I don’t know what exactly motivated Altman’s assailants, of course, just as I don’t know what specific thing radicalized Luigi Mangione or Tyler Robinson. But the 20-year-old Molotov-thrower had joined a Pause AI Discord and penned a Substack post on existential risk, writing that AI executives are “sociopaths/psychopaths” and “gambling with your future and the lives of your children… These people are almost nothing like you.” We know less about the second set of attackers, except that they are also young: 23 and 25.
What seems likely is that the anti-elite and nihilistic attitudes that have dominated US political culture in the last few years are transmuting into anger at AI billionaires. Young people are particularly incensed. Gen Z already grew up in a world that they felt was shrinking, where grift and shitcoins and sports gambling looked like the only paths up. Now, they’re being told AI is the reason they can’t get a job—and potentially never will. Just as the United Healthcare CEO seemed like a justified target to many disillusioned and radicalized young people, so will AI executives be to many more.
Obstructionism, cancel culture,
terrorismdirect action: this is what politics looks like when faith in democratic institutions has collapsed.
2 Likes
Read the full article here (gift link):
Silicon Valley Is Bracing for a Permanent Underclass
Whether you talk with engineers, venture capitalists, founders or managers, or with doomers, accelerationists, lefties or libertarians, the so-called San Francisco consensus on the impact of A.I. for workers is bleak. Many are convinced that advanced A.I. will soon surpass human capabilities. This would produce tremendous growth and scientific achievement, but it would also displace millions of jobs as fewer humans are needed to make the economy run. The technology will depress economic mobility and exacerbate inequality, while ferrying power and wealth to the A.I. companies and the existing owners of capital.
2 Likes
Meanwhile, in China:
Article: Chinese Courts Rule Companies Cannot Fire Workers Simply to Replace Them With AI - Caixin Global
2 Likes
There has been some more news on AI models producing novel solutions to old problems that people were not able to solve. Previous examples of this boiled down to models finding solutions that already appeared in the literature that nobody noticed. But more recent ones are cases where models produced truly novel solutions to problems people had worked on for several years. e.g.:
We introduce a new method for bounding Erdős sums of primitive sets, suggested from output of GPT-5.4 Pro, based on Markov chains with von Mangoldt weights.
(I added emphasis to the word “new”.)
If it happens in math, it can happen in other disciplines, as well. In some ways math is harder than other fields, and in some ways it’s easier. In math you tend to have long chains of logic where an error at any point in the chain completely invalidates the argument. (More nuance: proofs often do have errors, but they can also often be repaired.) My impression of biomedical fields, say, is that there is a higher tolerance for errors in logic than there is in math – the chains of logic tend to be much shorter and less rigid in biomedicine; though, you tend to use more ad hoc facts and what seem like deus ex machinas in movies (Why does such and so medical outcome occur? Well, it’s because of the XYZ gene, case closed.) and you rarely have a sense of finality with arguments proved. However, AI models are getting better at handling all this ambiguity and things.
Another interesting sign is this tweet from a U.C. Berkeley professor:
https://x.com/tonylfeng/status/2050576299123679299#m
While I agree in principle, in practice I think AI raises tough questions about what we even mean by “good science” in the context of mathematics. There’s an infinite number of true mathematical statements, many of which we can but do not bother to prove because we consider them “routine” and therefore uninteresting. What counts as “routine” is subjective, but I think it approximately means “doable using (only) well-known existing techniques”. If AI becomes consistently stronger than humans at certain mathematical skills (for example, “combining already-existing techniques in a new way”), then certain types of previously non-routine problems will become routine in a higher level sense: doable by the well-known technique of querying an AI chatbot. At that point, are those problems still interesting?
He’s starting to worry that AI will be able to solve harder and harder problems – even using new methods, or new combinations of old ones – which could cause people to question whether what they are doing is worthwhile anymore. A response to this tweet was given by the mathematician Ken Ono, who wrote:
https://x.com/KenOno691/status/2050719804924064056#m
One interesting line from that long tweet:
If AI had been around then, it likely would have solved it, because AI isn’t constrained by our human silos.
He seems to be implying that models are getting good enough to actually solve research-level problems from back in 1996. Imagine what it means for physics, chemistry, biomedicine, etc.
3 Likes
Insilico’s “LabClaw” = First Real Glimpse of a Self-Driving Drug Discovery Lab (2026)
Insilico Medicine just announced LabClaw, an AI system that coordinates an automated wet lab (“LifeStar2”) to run drug discovery in a semi-autonomous loop.
What it actually does:
- AI designs experiments and candidate molecules
- Robotic wet lab executes them
- Data is analyzed automatically
- AI decides the next round of experiments
- Humans supervise, not micromanage
1 Like
I assume we will achieve recursive- self-improving AI, and of rhat leads to Airtifical SuperIntelligence (ASI). The natural consequences of that will be profound - but - operational rollout will always be slower than scientific discovery. Bureacracy, regulation, employment laws, unions, human managers will all slow things down in terms of rollout.
But with ASI controlled labs physics, chemistry and biology will all progress at phenomenal speeds. Suggesting a super abundant world will arrive before mass unemployment.
The permanent underclass may eventually be all of us, and we may all be on a universal high income…
1 Like
You are right that businesses, especially larger ones, typically are very slow to adopt new tech. What I am hearing here in SF is that these slower companies will be killed by faster moving AI focused startups …
But this is interesting…
3 Likes
yes, that has to be right. The only question is the speed…
Large companies tend to have erected barriers to entry (compliance amd regulation in banking etc) very small companies are often the last to be disrupted because the superprofits are elsewhere (small local cleaning firms may be the last to be replaced by tesla optimuses). And the public sector will be the slowest of all to adopt AI.
So mass AI- induced unemployment will take time. And even a delay of a few years will be enough time for super intelligence to make some truly wondrous scientific discoveries in energy, material science and fundamental physics. So my essential point is that the world into which the overwhelming masses become unemployed - could look very different to the one we know today. If it is truly “abundant” by then - unemployment won’t be as bad as people fear. Assuming we get the politics of redistribution correct…
This sounds like a huge breakthrough in drug discovery using AI:
Paper:
3 Likes
Brilliant - it’s open source so anyone can use it to generate and dry test hypotheses. AI agents let loose with tools like this could make incredible progress.
1 Like
And, unfortunately, the casual reader who might opt to just skim the paper might come away thinking “ho-hum, another neural-net-improves-medicine paper…” That’s what I would have thought, except that I actually listened to the video first while driving. As the guy says in the first few minutes:
This might genuinely be one of the biggest breakthroughs in medicine and drug discovery we’ve seen in decades. A new paper just dropped and it’s about this new AI model called MAML. And this can completely transform biosciences and medicine. It could potentially come up with a ton of new cures for diseases like cancer. And this can make drug discovery way faster and cheaper and more accurate. This could also lead to personalized medicine and beyond.
I suppose that should make one skeptical (even after hearing the dozen or so unbelievable things they’ve used it to do that he mentions). I’d rather not ask GPT-5.5-thinking “tear it apart”, and just hold on to the excitement and hope it’s real and people pick it up and develop it further (and with that kind of prompting, you wouldn’t get the truth anyways)…
1 Like
The impact of AI models like this will be absolutely enormous - but its worth analyzing what it solves immediately, what it solves iteratively and what’s left …
what it does:
The bottleneck in drug discovery has never been ideas — it’s been the cost of testing them. A wet lab binding assay costs £500–5,000 and takes weeks. MAMMAL-class models run in seconds for pennies. That collapses the funnel: instead of testing 50 candidates experimentally, you screen 50,000 computationally and send only the top 20 to the lab. The hit rate per experiment goes up dramatically. That’s not incremental — it restructures the entire economics of early-stage discovery.
It also immediately opens up the prospect of precision / tumour specific cancer treatments.
In terms of longevity - it means that for all our current targets (mtor, ampk, sirt etc) we can quickly identify “clean” molecules with minimal side effects/off target toxicity. So it can quickly identify a better rapamycin for example.
The compounding effect is what’s Im most excited by. Each generation of model gets trained on data that includes results from the previous generation’s predictions. As validated hits accumulate, the models get better, which produces better hits, which feed back in. And all so very quickly - the next 5 years will be huge.
But it wont do much for the rest of the pipeline: pathways and metabolic systems still need to be teased out, and it doesn’t avoid wet testing … we get to candidate drugs at light speed - and then we go back into the slow lane for animal testing… and even slower for human testing
On this last point …The UK equivalent of the FDA is targeting with shortening the timelines - and is having huge success .
1 Like