Atomic Razor

The Inhumanoids

The Inhumanoids, I have probably said before, was the name of a science fiction tv series I wanted to create in a John Menzies notebook in Junior 4, in 1976-7. like most of my grand plans came to naught. Still, a pretty nifty name. OF course, someone else came up with it: "Inhumanoids is the title of an animated series and the name of a Hasbro toy property that were both released in 1986." I was there a decade earlier when I was 8. There's also this book from 2017. Great minds or not.

Anyhoo, this has nothing to do with what I actually wanted to talk about, which is that I was watching CNBC this morning and the estimable Carl Quintanilla (I assume you all follow him on Bluesky), where his takes are definitely spicier than on the broadcast channel) was taking about the technoeconomic tensions between China and the US. He mentioned China's lead in robots including "humanoids". Yes, he used the word humanoids on CNBC. It's probably not the first time and it certainly won't be the last, but it something that we have reached an epoch at which one can casually, in a mainstream medium, refer to China as having a lead in humanoid robots. As we have said, from CES and other sources, it's pretty clear than humanoid robots are going to be a thing. The clunk Boston Dynamics robots are long since a thing of the past. Of course, they have to be able to do useful stuff and it's probably going to be a while before we have one in our home has a personal butler, but I look forward to my first sighting of one in the wild.   

The Urgency of Interpretability

Dario Amodei has an interesting article on the urgency of interpretability. Of course, interpretability is nothing new. It's a classic classic AI problem. You can't just use a neural net blackbox to make mortgage decisions because you have to be able to justify them. This is one of the AI ethics that the kinds of researchers who get to get funded by the Turing Institute work on. I seem to recall it being discussed at workshop I went to at Imperial over a decade ago. In those days, you probably couldn't have build a neural net-based system even if you had wanted to, so you probably had handwired logic. Amodei is talking though about we can get interpret the actions of frontier models. Well, a big advantage over human minds is that we can look inside. Of course, what we see is mostly large matrices. But people, well, some people, a few people, have spent a lot of effort looking into this question,. There's certainly things we can do. But as, AI 27 suggested recently, this might only get us so far.   

Of course, interpretability and alignment aren't necessarily the safe. You could have an aligned, but non-interpretable A(G|S)I, one that, like the Eschaton, moved in mysterious ways. Equally, you could have non-aligned, but interpretable A(G|S)I, one that was perfectly open about not caring much, if at all, for humans. But here's the thing. There was a point, I guess, in the mid-2000s, when Yudkowsy realised that (a) he wasn't a good enough programmer to program the seed AGI; (b) even if were, there would be no way to ensure that it would stay aligned. He realised that his grift was better serviced running a website or writing sophomoric essays or Harry Potter fanfic. MIRI did do some work on 

No publications since 2021, which given the current situation is noteworthy (there's a couple of November 2024 preprints). Their homepage declares

The AI industry is racing toward a precipice.

The default consequence of the creation of artificial superintelligence (ASI) is human extinction.

Our survival depends on delaying the creation of ASI, as soon as we can, for as long as necessary.

The general feeling is that the doomsters and the accelerationists (Altman, certainly, Amodeii, presumably) lost. I am not sure where that leaves the grift. I guess you can buy ESY's books on Amazon. But I wonder what his plan is. Does he actually believe this stuff? I mean I do. Gemini 2.5 is a better physicist than I was at 27. It's also a better virologist. That's not a combo we have had before. I except an extinction-level event in the next years, whether deliberate or accident, state or non-state, viral, bacterial, fungal, grey goo, nanotech, mirror life.

Here's a picture of ESY with Altman, but I don't know when this was taken.

Because the thing is that, as hinted at by AI 27, you are probably going to have to have a trust network. So, you trust an AI that can interpret an AGI that can interpret an ASI. But it's like trying to explain the thinking of a Fields Medal-winning mathematician to the person on the Clapham omnibus. Maybe, the ASI is so good at interpreting verself that ve can make verself understood. But it has always seemed clear to me and I think Geoff Hinton and, it seems, ESY that no dynamic cybernetic system driven buy complex constantly evolving metagoals (and metametagoals) can ever be truly trusted. As with a person, you have just got to hope they don't unexpected go off the rails. Most people don't most of the time, but that's because humans are powerfully influenced by other humans and societal norms. That won't apply in the same to weakly god-like entities. You just can't know for sure how something is going to react in all circumstances and forever.   

Perhaps we need that Butlerian Jihad after all

(Only kidding!)

Rational Agent

Interesting article by Jasmine Sun on her Substack (I know, I know), an interview with "an AI researcher, founder, and community-builder". Two things stand out.

Firstly,

What does “AGI” mean to you?

I use the classic definition of AGI, instead of moving the goalposts. The “G” in AGI is about generality. There are two types of AI: narrow and general. We used to train models for specific tasks. For every task, you’d build a huge, specific data set and train a model just for that task. That changed with the alpha version of Claude I was using in the summer of 2022. It could do any task you asked it to do. It was fucking crazy. It didn't care whether it had specific training data. It was general while prior models were narrow.

When you say “moving the goalposts,” what do you mean?

There are ideological and corporate reasons why it’s important for some people that we never achieve AGI. The reality is that AGI is a coordination tool, rather than a specific technology or moment. If you look at the history of the AGI concept, it’s the organizing principle behind the more ambitious—you might say religious—side of the field.

Which is interesting because I have sometimes suggested that what we have is a kind of proto-AGI because generative AI has such a wide range of capabilities. As Dario Amodei points out in a recent essay, FMs are, effectively, "grown". That's why they are robust in contrast to traditional software systems, which are brittle. Bit, of course, we don't tend to think of FMs as robust because we concentrate on what they can't do (robustly). From the perspective of 2002, 2022 Claude would have seemed like AGI. but we don't think of it as AGI because it can't do some of the things we would reasonably expect AGI to do. Which is why I think he's wrong that just claiming AGI hasn't arrived is about keeping the hype cycle going. Claiming AGI has arrived would boost the hype cycle, but also inevitably lead to a lot of flak. WE aren't there yet, but we might be in a couple of years. Gemini 2.5 Pro is very powerful and it's only going to get more powerful.

Secondly, 

            Also, I talk to this group of clones of myself all the time on Telegram. I have 14 years of journals, about 30,000 pages, and I put all that writing into a system that ideates with me.

Do you have clones of mentors or friends, like an imaginary board of advisors?

I never got into that. But I cloned an ex-girlfriend of mine once. Her clone was really, really her.

Leaving aside the ethics of cloning his ex-girlfriend (indeed), how is this done? Is this a system he created himself or is this something off the shelf? I can't find anything very obvious, but, of course, you could do finetuning on top of an existing FM with RAG or whatever based on your own data. In my case, that would be my blog. So, I guess this is something that might be relatively easy to build. Not I could that with my Jetson Orin Nano Super Developer Kit. It's an intriguing possibility. 

Divergent Trajectories: A Comparative Analysis of Artificial Intelligence and Drexlerian Molecular Nanotechnology

I asked Gemini Deep Research with 2.5 Pro why A(G)I has advanced so much more successfully over the last 30 years compared to True Drexlerian Molecular Nanotechnology. This was inspired by the sad death of Damien Broderick recently and finding that the revised edition of The Spike was just $2.99 on Amazon US. I read the original version back in February 1999, where I first came across Eliezer S. Yudkowsky. The second edition contains significant few stuff from the late 1990s  and very early 2000s, so worth a punt. However, the reference to an article in the August 1995 edition of Wired might well be in the original. "Wired asked several nanotech experts to estimate when we will reap the rewards of their research."

Respondent	Molecular Assembler	Nanocomputer	Cell Repair	Commercial Product	Nanotech Law
Robert R. Birge	2005	2040	2030	2002	1998
Donald W. Brenner	2025	2040	2035	2000	2036
K. Eric Drexler	2015	2017	2018	2015	2015
J. Storrs Hall	2010	2010	2050	2005	1995
Richard E. Smalley	2000	2100	2010	2000	2000
Bottom Line	2011	2041	2029	2004	2009

I had always thought of Smalley as a critic of MNT, but perhaps I was wrong. He died in the 2000s sadly, but the others are still alive. Obviously, Donald W. Brenner was the one to got closest. He's still active, although not seemingly in research directly related to MNT, but I don't think he's going to be finally unveiling his molecular assembler this year. Someone might; we still have 8 months to go after all! There are obviously lots of commercial products that at some level are nanotechnology, although not really TDMNT, and I suspect that's what Brenner was thinking of in saying 2000. I suspect we will have nanocomputers by 2041. A lot of depends on whether we get the Amodei data centre/country of (super)geniuses and 50-100 years of bio/phystech progress in 5-10 years. Cell repair by 2029 seems optimistic by 2029, but Brenner suggested 2035, which could be on the money still. A molecular assembler by 2030? I am not going to bet on it, but then we didn't really expect to have AGI by 2025, yet we kind of do. 

The Gemini report included this useful table of comparative analysis of AI and Drexlerian MNT development.

Factor	Artificial Intelligence (AI)	Drexlerian Molecular Nanotechnology (MNT)
Core Domain	Information Processing	Matter Manipulation
Fundamental Challenge	Algorithm/Logic/Learning	Physical Fabrication/Control
Development Path	Incremental, Paradigm Shifts	Foundational Breakthroughs Needed
Key Enablers	Algorithms (ML/DL), Data, Computation (Moore's Law, GPUs)	Precise Positional Control, Mechanosynthesis (Theoretical/Nascent)
Hardware Dependence	Leveraged existing/advancing hardware	Required invention of core hardware (assemblers)
Simulation/Testing	Highly effective, rapid feedback	Useful but limited, physical realization is key
Tangible Early Outputs	Yes (games, chatbots, expert systems)	Limited (STM images, theoretical designs)
Funding Cycles	Boom/Bust (Winters) but Resilient	Initial interest, then marginalization/niche funding
Key Bottlenecks Overcome	Combinatorial explosion (via heuristics/ML), knowledge acquisition (via ML), computational limits (via hardware)	Feasibility critiques (fat/sticky fingers), thermal noise, control complexity (largely unresolved for universal assemblers)

Back in 1995, Drexler said he "believes that nanocomputers could finally provide the horsepower needed for artificial intelligence." That seemed to have turned out to be the case (I know, I know, but Gemini 2.5 Pro is better than me at 57 at physics,; it's also better than me at 27 at physics and people really should try the latest frontier models; o3 might not know how many "r"s there are in "strawberry" or be able to do simple algebra, but Gemini does; I suspect o3 also does, but if ask it an "easy" question it doesn't get answered by the full model; one thing that's clear is that different FMs have different capabilities and they are advancing rapidly). The chart very neatly lays out how we got to AGI before TDNMT. In summary, self-supervised training on large datasets of transformer architecture neural nets with backprop and gradient descent plus RL(HF) and various other tweaks like mixture of experts. There's still a lot of low hanging fruit and opportunities to combine different design philosophies, which is why I expect progress to continue and that we might get to something very like AGI by 2027 or so. Of course then, in the ideal world, we can use AGI to solve all the other problems. And one of the first and most fundamental problems to solve is TDMNT. Recall that Drexler's original 1981 nanotech paper was on protein engineering. Consider AlphaFold and the existence proof of the biological cell. Might we finally achieve the dream of the nanomachine that is formed from various proteins that fold into the engineered shapesof machine's components? It's going to be a crazy ride. Stay tuned.  

All That Is Solid Melts into Air II

Asked Gemini Deep Research with 2.5 Pro for a list of the top 10 largest global companies by market capitalization in April 1975 wondering if would do better than Gemini 2.5 Pro on its own. It certainly gave me a nicely formatted table and these companies are substantially the same as the previous list. Royal Dutch Shell and British Petroleum make the list, so we get five of the sisters (weren't there seven?). 

Table 1: Estimated Top 10 Largest Global Companies by Market Capitalization (Circa April 1975)

 

Estimated Rank	Company	Primary Industry	Headquarters Country	Estimated Market Cap (USD)	Supporting Evidence Notes
1	IBM	Technology / Computers	USA	N/A	Top 3 US Market Cap YE 1975 6, Top 10 Revenue 4, Nifty Fifty Leader 15
2	AT&T	Telecommunications	USA	N/A	Top 3 US Market Cap YE 1975 6, Dominant Utility 9
3	Exxon (Standard Oil of NJ)	Oil & Gas	USA	N/A	Top 3 US Market Cap YE 1975 6, #1 Revenue 4, Major Oil Player 8
4	General Motors	Automotive / Industrial	USA	N/A	#2 Revenue 4, Historical Giant 9, Likely Top 10 Market Cap
5	Royal Dutch Shell	Oil & Gas	UK / Netherlands	N/A	Major Global Oil Company 17, Leading Non-US Contender
6	General Electric	Industrial Conglomerate	USA	N/A	Top 10 Revenue 4, Historical Giant 6, Likely Top 10 Market Cap
7	Texaco	Oil & Gas	USA	N/A	Top 5 Revenue 4, Major Oil Player 8, Likely Top 10 Market Cap
8	Mobil	Oil & Gas	USA	N/A	Top 5 Revenue 4, Major Oil Player 8, Likely Top 10 Market Cap
9	British Petroleum (BP)	Oil & Gas	UK	N/A	Major Global Oil Company 16, Noted Strong Stock Performance 16, Likely Top 10
10	Eastman Kodak	Chemicals / Photography	USA	N/A	Historically Significant 9, High Profits 4, Plausible Top 10 Contender over lower-margin Industrials

The oil companies all still exist though Mobil merged with Exxon and Texaco is now Chevron. There's company called AT&T, but just a Baby Bell rebranded. The actually interesting bit of AT&T that we might care about and helped invent the future is, perhaps a little bizarrely (or not), owned by Nokia (which is also Alcatel, Lucent and Siemens telecoms). General Motors still produces not very good cars, but not so much in Europe if at all. No-one's saying what's good for GM is good for America and vice versa any more. In 1975, Rex Malik, who I must have come across on The Computer Programme, published And tomorrow ... the world?: Inside IBM, which I have a copy of and must read when I get it here because I suspect it's a fascinating account of IBM at its peak. In a way, AT&T invented more of the future that IBM did, we all hacve devices running Unix, but who is running OS/360 (or, indeed, z/OS; yes some people, but I have never come across any IBM hardware in the wild, except for someone who had an IBM workstation at DERA Holton Heath that ran AIX - we did briefly use a 5250 timesheet application when I first went to Ericsson)?

General Electric was dissolved into several parts in 2024, but several of the parts are still substantial and use the GE brand. But it a good example of a company that was absolutely driven into the ground by neoliberal ideology. I wonder what the return on a share of GE over the last 50 years has been. At times, probably good. Of course, in the long run, it's only cash that matters. 

Of course, the real loser here is Eastman Kodak. About 4000 work for the residual company. How the mighty are fallen. Kodak did do other stuff. I seem to recall an O'Reilly book recommending Kodak CD-ROMs. But it's difficult to pull off such profound pivots. The chemical photography business went from a mainstay of technological society to little more than a hobbyist niche in just a decade or so. They were mugged by history. Got to hope that the rest of us aren't.

 

All That Is Solid Melts into Air

The ten (actually eleven) largest companies in the world as of April 2025:

• Apple (AAPL): ~ $3.0 - $3.4 Trillion (Technology - USA)  
   
• Microsoft (MSFT): ~ $2.9 - $3.2 Trillion (Technology - USA)
• Nvidia (NVDA): ~ $2.7 - $3.2 Trillion (Technology/Semiconductors - USA)
• Amazon (AMZN): ~ $1.9 - $2.4 Trillion (E-commerce/Cloud Computing - USA)
• Alphabet (GOOG/GOOGL): ~ $1.9 - $2.3 Trillion (Technology - USA)
• Saudi Aramco (2222.SR): ~ $1.7 - $1.8 Trillion (Oil & Gas - Saudi Arabia)
• Meta Platforms (META): ~ $1.4 - $1.8 Trillion (Social Media/Technology - USA)
• Berkshire Hathaway (BRK.A/BRK.B): ~ $1.0 - $1.1 Trillion (Investment/Finance - USA)
• Tesla (TSLA): ~ $0.8 - $1.2 Trillion (Automotive/Clean Energy - USA)
• Broadcom (AVGO) / TSMC (TSM): These two companies often trade places for the 9th/10th/11th spots.
  • Broadcom: ~ $0.8 - $1.1 Trillion (Semiconductors - USA)
  • TSMC: ~ $0.8 - $1.0 Trillion (Semiconductors - Taiwan)

So, that's one oil company, one holding company and nine tech companies (I guess Amazon at some level is a retailer). If we look at the year the companies were founded

1. Apple: 1976
2. Microsoft: 1975  
3. Nvidia: 1993  
4. Amazon: 1994
5. Alphabet (Google): Google was founded in 1998. The parent holding company, Alphabet Inc., was formed in 2015.  
6. Saudi Aramco: 1933 (Originally California-Arabian Standard Oil Company)
7. Meta Platforms (Facebook): 2004 (Founded as Facebook)
8. Berkshire Hathaway: 1839 (Originally as Valley Falls Company, a textile manufacturer. Warren Buffett took control in 1965, transforming it into a holding company).  
9. Tesla: 2003  
10. Broadcom: 1991 (Founding of Broadcom Corporation). Note: The current entity, Broadcom Inc., was formed when Avago Technologies (with roots tracing back to a Hewlett-Packard division founded in 1961) acquired Broadcom Corporation in 2016 and adopted its name.
11. TSMC (Taiwan Semiconductor Manufacturing Company): 1987  

So, one in 1839, but really 1965, one in 1933, two in the 1970s, one in the 1980s, four in the 1990s, two in the 2000s. So, only three of these companies existed 50 years ago and one of those no-one had heard of.

Gemini 2.5 Pro doesn't give me a definitive list for April 1975, but suggests

Considering these reference points and the economic climate of 1975, the largest companies by market value likely included a mix of these names. While a precise, verified ranking for April 1975 isn't available from the search results, the top 10 probably consisted of companies such as:

1. IBM
2. AT&T
3. Exxon
4. General Motors
5. General Electric
6. Eastman Kodak
7. Texaco
8. Mobil
9. Sears Roebuck
10. Procter & Gamble or DuPont

Other major companies like Gulf Oil, Chevron (Standard Oil of California), Amoco (Standard Oil of Indiana), Shell, Xerox, and 3M were also very large and could potentially have been in the top 10 depending on market fluctuations during that specific period.

So, none of the top ten companies from April 1975 are in the top ten today. We see a preponderance of oil companies. I suppose General Electric did make computers and, of course, Unix and C, for instance, came out of of AT&T, although none of these companies are really tech companies as we think of them today with the possible exception of IBM. (A related issue is what do we mean by tech companies. The Weasel might argue that in some ways the internal cultures of these companies might not be so different from modern ones in some ways. Then again, no-one was sending other people emails, to first approximation.) The global economy is approximately 5.5 to 6 times larger in 2025 than 1975 and we can see that much of that extra value is linked to computers or at least that computer companies have captured much of that value. Of course, this fits with our prediction for a technological singularity (of some kind) at some point in the next few decades: "All That Is Solid Melts into Air".

"

Too Cheap to Meter

Just imagine that in the 1960s the world had built ten (or a hundred) times as many nuclear power stations than it did, so that the use of coal or gas or oil for power generation had been completely removed by, say, 1980. You could use hydro or tidal or geothermal or solar thermal, all technologies that existed at the time. Of course, this would have been difficult. Fossil fuels were cheap and efficient, but it's not as though we didn't build lots of nuclear power stations. Of course, with lots of 1960s- and 1970s-era nuclear power stations, we are going to get lots of nuclear disasters and well as the usual radiation releases, probably at least one full meltdown, but there are ways round that and it might still be preferable to climate breakdown. Nuclear proliferation is going to be a big issue in this world as everyone has nuclear power stations, so might not be a utopia.

You have to get rid of the use of gas for heating, cooking, but, again, that could be OK. We had an electric cooker. You can use hydrogen in internal combustion engines and fuel cells already existed, so you can go down that route too. Glass microbeads might be the safest approach, but I suspect we would also see FeTi, compressed gas, liquid hydrogen at 20K, slush hydrogen, lots of possibilities. Planes, trains and ships could all use hydrogen. So, by 1990 or, we could have completely decarbonised the economy. Of course, it would have been expensive and there are going to be lots of issues, though there would be incentives to make things safer, but this doesn't require any new technology just will. Of course , that was never going to be there and here we are. 

Ten Ages of the World

I am 57 next week. I can't remember much before 1975, so my time window is about 50 years.

1975: our houses in Preston and Seattle were there. We, of course, lived in the one in Preston. Beechwood Avenue and 45th Avenue SW looked, in many ways, much as they do today. We lived in advanced technological society under capitalism (you could, I suppose, argue about whether it was "late"; according to the term was coined in 1928 and was used widely in Germany from the 1960s and especially from 1972, also by Frederic Jameson a great deal, so we could say the capitalism of 1975 was late even if there are significant differences from the hypermodernity of 2025) . But if you told someone in 1975 that in 2025 the richest man in the world would have made his fortune from electric cars and reusable rockets and he would be the shadow president and doing all the other things, 2025 would seem a pretty science fictional place. More about this in a later post. 

1525: Preston existed and you could go to the pub,. The broad out line of the some of the main streets would be the same as today. We know a lot about C16th Europe and in some ways it's not so far from the current day West. We are already in the Early Modern Period after all.People talked something like modern English, although it would probably take a while to adjust, but you could probably navigate the society without that much trouble. Seattle doesn't exist, but the Coastal Salish people are here and we can deduce something of their lives from archaeological and ethnographic evidence. They won't yet be in contact with Europeans, but things are about to start changing even before the direct encounter with the spread of disease, trade goods and new technologies as well as population displacements. The outline of the continents is pretty much what we have today. 

~3000 BCE: there are people in Preston, but what life is like in pre-Bronze Age (so neolithic) Britain is pretty uncertain. The Bronze Age is going elsewhere, but won't reach Preston for the best part of a millennium. There are people living in the Puget Sound area and we can deduce something of their lives from archaeological and ethnographic evidence. The really interesting question is how different were they culturally, technologically, linguistically from the later inhabitants. We know in the case of Preston, very different, but how would that map to Seattle? What language did they speak in Lancashire. Some kind of proto-Celtic or was it more like Basque? There are definitely marshy areas that affect the outlines of some coastlines, so some maps would look slightly strange at high enough resolution.

~48000 BCE: depending on whether or not we are in a ice age or not, the maps will more or less different, but the continents are in their present position. There are probably no people in Puget Sound or Preston, but there are certainly hunter-gatherers trucking around Africa, Asia, Europe and Oceania and they can talk. How much they are like modern hunter-gatherers is debatable and it might take a while to learn their languages as they are probably nothing like any modern ones. Lots of charismatic megafauna around that we don't have today.

~500,000 BP. Probably not modern humans as early as this and probably no language, but definitely fire and tools. Depending on the sea level, lots of interestingly different crinkly bits around the coasts. Rivers and lakes would be different, but much of the EArth wouldn't look that different from space. 

~5,000,000 BP: the continents will be in slightly different places, so things will probably look a bit different from space. Plenty of hominids around as well as lots of familiar and less familiar animal and plant species. Probably a bit Uncanny Valley, but maybe not a bad place to live.

~50,000,000 BP: the continents haven't banged into each other yet, although the broad outlines of the continental masses will be recognizable. We are very much in the Age of the Mammals (no non-avian dinosaurs), but not many familiar species. Early primates though are around and we have trees and grasses and flowering plants. Would be a fascinatingly different place to visit because it would be both so alien and so familiar simultaneously.

~500,000,000 BP: the continents look nothing like today. The Cambrian Explosion has occurred, so the oceans are teeming with life and most of the modern phyla exist, but there very little on land yet, so this is an almost unrecognisably alien world with almost nothing we expect from the Earth present. 

~5,000,000,000 BP: the Sun and the Solar System and the Earth have not been to form yet and won't for another 400 million years. Yet to an observer on a planet in our position in the Milky Way, the night sky would probably look familiar enough. According to the FoAK, "8.8 billion years (5 Gya, z = 0.5): Acceleration: dark-energy dominated era begins, following the matter-dominated era during which cosmic expansion was slowing down", which might be significant, but would be hard to detect at the time. Of course, as I have suggested, this transition does suggest that someone did some large-scale engineering with the universe at this time that affected the fundamental structure of the cosmos causing the acceleration to start.

~50,000,000,000 BP: the universe doesn't exist. Which is weird. Why is the universe so young? Especially given that the future of the universe is so long? The Big Rip might be 22 billion years in the future, but probably won't happen in which case the future might last to we get to

years when a Boltzmann brain could appear in the vacuum via a spontaneous entropy decrease or years when quantum tunnelling in any isolated patch of the universe could generate new inflationary events, resulting in new Big Bangs giving birth to new universes. (yes!). Doesn't that suggest, like the cosmic acceleration, that there's something weird going on? Of course, we can use anthropic reasoning to argue that it's not a surprise that we live at this epoch. The modal person will just before the end. But you can see why Hoyle preferred the steady universe for philosophical reasons or why I think there must be an ensemble of classes of universes. Also, I think there's a novel in the discovery of a star that's 50 billion years old. How can it possibly exist in our universe?

It has been and will be a hell of a ride. How knows what we might get to observe in next 50 years as we approach the Epoch of the Data Center of the Country of Supergeniuses?

Prioritising 50 Projects V

41. design sustainable garden or permaculture experiment.

We need, at some point, to do over the back garden and I want a Beer-Pask cybernetic pond. But this is going to happen until 2026 at the earliest.

42. write a play or screenplay.

Suffers from the problem of all my writing projects - time.

43. poetry course.

Plenty of those at Hugo House. But probably not until 2026 at the earliest.

44. learn a new programming language (Go, Julia, Lisp, Rust).

In the epoch of Cursor though, is that still worthwhile?

45. create an annotated bibliography or research compendium.

This would be a very good idea.

46. experiment with biohacking or cognitive enhancement.

I wonder what sort of things I meant or that could be done. Probably not to 2026.

47. Carping On ("an Old Rectory production") podcast.

But who would want to listen to us?

48. develop an interactive fiction story or text-based game.

Suffers from the problem of all my writing projects - time. I have to make it.

49. create a personal knowledge management system to catalogue and link ideas, projects and notes across my various interests.

Kind of the same as 45, but yes.

50. design and 3D print my own miniatures for wargames.

Would I really want to design them? Although I suppose "design" covers a lot of ground. It's possible, I guess, if I get a 3D printer, but that's not happening for a while.

A(G|S)I Comes at A(G|S)I Time

The Weasel is fond of saying that steam engines come at steam engine time. So, we need a combination of technology (metallurgy having advanced to the point that you can make a boiler that won't burst under pressure), a killer app (pumping water out of mines) and a set of technoeconomic innovations that make it easier to bring together clients, entrepreneurs, inventors, investors and workers. A fourth requirement would be the set of inventions related directly to how the steam engine works.

The thing is that people have been using cannons in Europe since the C14th, so was it really advances in metallurgy that made steam engines practical at that epoch? People have been mining tin in Cornwall for thousands of years, so was there really a sudden need for pumping water out with steam engines? There were certainly a lot of social and economic innovations in the late C17th/early !8th (central banks, stock markets, insurance, limited liability companies, etc, etc), but it is not as though people weren't doing large, medium and small scale commercial and state enterprises for hundreds or thousands of years before that.A technology has intricate as printing became widespread from the mid-C15th, shipyards were churning out ships, people were toiling away in mines and building water and windmills as well as elaborate clocks, so it is not obvious that steam engine time, sat the least couldn't have been at least a century or two earlier.

As I said on Bluesky yesterday, AI-1956, the Dartmouth summer workshop programme, symbolic AI, GOFAI, the Royal Route to Human-Level Intelligence, turned out to be a bust. But that wasn't obvious at the time. The term AI was coined to avoid having to use the C-word, cybernetics for the workshop, because then they would have to have invited Norbert Wiener. But it's hardly the case that there were lots of other approaches at the time - neural nets, Ashby's Design for a Brain, Rosenblatt's Perceptron - that were potentially promising. 1950s computers were incredibly feeble compared to modern ones, but it didn't seem like to people at the time and you only need ~8 kB to start doing really interesting and useful things. It wasn't obvious in 1956 that intelligence couldn't be achieved through the combination of a few algorithms and heuristics. After all, no-one had in a position to try to build a brain up until then.       

So, Lighthill was right! There was no point doing AI research in the 1970s because you just didn't have the scale - compute and data - to do anything useful. Of course, Lighthill wouldn't have thought of it like that, and it might be that a lot of what was going on was academic knife grinding, but, of course, to first approximation, it was true. For decades, AI was 20 or 50 years off because for decades, no-one had the scale to do anything useful It's still possible that AGI is 20 or 50 years off. That seems unlikely in an age where we have robust quasi-AGI already (by robust, I mean that you can ask Gemini 2.5 Pro anything in natural language and it will usually have a pretty good go at answering your question or doing the requested task, even if it doesn't always get it 100% right, that's better than 95% of co-workers in my bitter professional experienc; this capability is probably the most extraordinary thing about FMs - we complain about hallucinations or incomplete tasks, rightly so, but the shift from the brittleness of traditional software engineering is still astonishing), but there could be many gotchas yet and, of course, we probably want hybrid systems that incorporate reinforcement learning and symbolic AI too as well as improved capacity to generalise from small samples and possibly not train with backprop and gradient descent because the brain doesn't. Let a thousand philosophies bloom.

Noam Brown has achieved many extraordinary things - solved poker, systems that play chess like humans, a system for playing Diplomacy - and said at the recent NVIDIA conference that we could have got here a couple of decades earlier. AGI time could have been in the late 2000s rather than the late 2020s! I want to dig deeper his thinking, but it seems to be that more thinking time would help with reasoning models. FMs respond typically very quickly, but what if they thought about a problem for hours instead? But obviously you could go back to the 1970s, as the Weasel suggested to me, to ICL in 1973, and design a programme that would get us to AGI sooner. We will talk about this more later, but you would build neural nets trained with backprop and gradient descent. This could have been done in 1973. You could use an ICL Distributed Array Processor,do lots of OCR applications, create a big training corpus, do self-supervised learning, get to advanced architectures like transformers, mixture of experts, thinking time. But you probably also need a couple of breakthroughs, such as more efficient learning. Get the internet earlier and maybe you can have enough data two decades earlier, even if you raw compute isn't much or any faster. Obviously, specialised processors, like GPUs and neuromorphic chips. It doesn't seem to unreasonable that we could have had systems a bit like modern FMs in 2005. But then hindsight is 20/20.Steam engines in 1580? AGI in 2007? Combine the two. We could have the Singularity in about 1809. There's a novel in that.