Atomic Razor

Twenty for the 20s Revisited at Mid-Decade

Incredibly, we are already exactly halfway through the 2020s today (actually slightly more than half because we have had two of the three leap years). Back in the Before Times, five tears ago exactly, on 31 December 2019, I put forward Twenty for the 20s. At mid-decades, it seems appropriate to review my prognostications. 

Twenty for the 20s

Back in 2014-5, after I had been to Arvon, Simon Ings suggested a write  fix-up, which was a very good idea, and I did a series of posts of the world in 2034, 2054... which, I think still stand-up pretty well. I am currently, when things like Christmas, I know, don't get in the way doing in the Institute for the Future Coursera course on futures thinking. Their focus is typically on the ten-year horizon, so here are my Twenty for the 20s:

1. Level 10 Black Swan. Which would be an extinction event. It's quite hard to wipe out the whole human race, but not impossible. No Personal Nanofactories by 2029, so I think this won't happen.

PNFs are probably slightly more likely now than I thought in 2019. The probability of a Level 10 Black Swan - gray goo, engineered pandemic, unFriendly AGI - is definitely higher than in 2019, but still, I think unlikely.

2. Level 9 Black Swan. The collapse of the current technological civilisation. Again, quite hard to do, but again certainly not impossible. Full-scale thermonuclear exchange? Natural or artificial pandemic? Climate breakdown-induced breakdown of global food supply chains? I don't think this will happen, but, of course, that is in large part because I don't want it to.

Well, we had a pandemic. Given we are in the gravest geopolitical situation since the 1930s, the probability of a catastrophic event or chain of events seems much higher than in 2019.

3. Level 8 Black Swan. That WWI-level event that we have felt lurking our there. Given the current nature if the catastrophe, this is quite likely. You simply can't have the kinds of people in charge who we have in charge in certain countries without something catastrophically awful happening. I think there is a real chance of one of these. History might well be about to return to Terra and change all our lives forever.

COVID-19 and the Russia-Ukraine War probably count as a Level 8 Black Swan. Given what is going to happen next year, I think it unlikely that we get through the rest of the decade without at least one more event of this magnitude. Could be something technology-related.

4. Heterodox economics. There have been hints towards this including with the Labour Party, but Corbyn was never the man to sell cybereconomics and Beerism. Might Cummings be? Not sure yet about Modern Monetary Theory, but there are enormous possibilities outside the neoliberal consensus. 

Neoliberalism  has run its course, but I don't see the weak and divided left having any will or capacity to put anything innovative into action. What comes after neoliberalism, in the post-WTO, post-Washing Consensus era is mostly going to look like neoliberalism, just with higher tariffs, some onshoring, less regulation, the end ot the rule of law and massive systematic endemic corruption. My guess is that the US economy will look a lot more like the Chinese economy by the end of the decade, but on a day-to-day basis , CNBC will still feel much as does today or in 2001.

5. Neonationalism. We will see how that shows in the UK over the next four years. Can only hope that we can get our acts together to avoid the worst consequences of this. But I am not optimistic. There is that Level 8 Black Swan lying in ambush.

Right on the money on that one. the UK at least has a Labour government through 2029, but I don't see many countries in Europe or, indeed, the world, surviving the wave of Putinisation/Orbanisation we are going to see in the second half of the 2020s.

6. Scottish Independence. Both more compelling and harder to achieve than in 2014. That hard border along the Cheviots is really going to be an issue. I think they missed the Scots chance, which is both bad and good.

No-one could have foreseen what would happen to Eck and Sturgeon. Labour's inevitable failures over the next few years are lively to see something of a revival for the independence movement, but I don't know how this plays out with Orbanisation. Certainly, Scotland isn't to be independent by 2030.

7. Globalism. The African decade? We have already seen breakouts by companies like Samsung and Huawei. But what about a Nigerian or Brazilian company?

Nothing there yet as far as I know. It would be good to see something by the end of the decade.

8. A(G)I. We have had an AI Summer. Everyone is expecting an AI Winter. Clearly, Deep (Reinforcement) Learning is not the whole answer. Nevertheless, how many of us expected Go to be solved and solved in the way it was? My hope is that if we let a thousand philosophy bloom, there is a chance that some PhD student in China can come with an idea that a postdoc in Canada can leverage and suddenly open up a whole new front. Plus, we are getting the fourth edition of Russell and Norvig early in 2020. I suspect some time before the end of the decade we will have systems that meet the definition of domain-specific AGIs idiot savants/dog/cat/great ape-level AGI. And remember, humans didn't even evolve language for something like the first three-quarters of existence, which is incredibly freaky, and suggests you can glom language on top of non-linguistic GI.   

Well, we certainly didn't have the AI Winter. It's been a continuous AII and now AGI Summer. Proto-AGI systems exist and with o3 we have a system vastly more powerful than the one we knew about only two weeks ago. DeepSeek seems a huge deal, perhaps even a Sputnik-level one. NVIDIA is the second largest company in the world in terms of value on the basis of AI. We can vastly cut training time and Noam Brown, who "solved" poker and Diplomacy recently opined  “It turned out that having a bot think for just 20 seconds in a hand of poker got the same boosting performance as scaling up the model by 100,000x and training it for 100,000 times longer." Demis Hassabis got the Nobel for AlphaFold. Solve AGI and then solve everything else. Dario Amodei thinks we will have a data centre of Nobel Prize-winning scientists by 2027 that will give us 50-100 years of progress in biology and medicine (and in physics and chemistry too) in 5-10 years. This is literally the only thing that is keeping me going right now. 

9. Self-driving cars. A hard problem, but think where we were in 2009. Does not require full AGI. I hope to own one by 2029.

I saw my first one in the wild in Phoenix in March. Might require AGI! I doubt I will own one by 2029, but certainly hoping we have them in Seattle within the next year or two. 

10. Electric cars. These are already a thing in that I see at least one pretty every day just driving around Preston. I even saw an electric scooter outside a kebab shop the other day. Should be fairly ubiquitous by 2029.

There were pretty common in Central London in June 2023 because of the congestion charge. New York really needs to follow suit, but there are plenty of them in Seattle and I have been in several.

11. Hydrogen/methane economy. If we want to completely decarbonise the global economy by 2045, we are going to need a lot of local hydrogen and methane (for instance, from biodigesters for use in ships) production, so I hope to see the infrastructure for this becoming common.

I saw a hydrogen doubledecker in London last summer! There's definitely stuff going on this space, but sadly it's likely to slow over the next few years for geopolitical reasons. We shall see.

12. Fusion. There are quite a few fusion startups around. Can one make the breakthrough? This is the thing we need to solve the energy crisis fully and also to open up the Solar System.

Supposedly, there are companies that are contracted to deliver fusion reactors and this is an area that will be helped by AGI.

13. Quantum computing. This won't break encryption because we already know of types of encryption that QC can't break. It will certainly be good for simulation of quantum systems and quite possibly for various database-related applications and for brute forcing certain problems. Whether we will have desktop quantum processors by 2029, even in the lab, is another matter.

Definite progress is being made with quantum computing and the recent Google announcement is apparently a big deal, but I would like QCs to be able to do something useful, not just demonstrate a principle.

14. True Drexlerian Molecular Nanotechnology. We haven't seen as much progress in this area as we expected in the 1990s. We have seen an AI Summer and we are overdue an NMT one. It is not as though it has suddenly become a bad idea. Might well be helped by quantum computers, which will be good for simulation of chemical systems.

This is the kind of thing your data centre of Nobel Prize-winning scientists might be able to do fairly rapidly, so who knows?

15. Return to the Moon. We should get this. And a moonbase. Yeh! But will it be Moonbase Britannica with the bally Union Jack on it?

We should get this, though the moonbase might not amount to much by 2029 and there isn't going to be a Moonbase Britannica There might be by 2040.

16. Heavy Launch Vehicles. If we get the Starship and the New Armstrong , things could get quite exciting quite quickly. Orbital tourism? asteroid mining? Mars base? A golden age of uncrewed missions to the outer planets?

Should get the first orbiting of Starship in 2025 as well as the New Glenn. A Chinese HLV by 2029? No idea what the New Armstrong timetable might be. Reddit postings from a couple of months ago suggest there;'s no actual work in progress on it.

17. Life extension. Would be good to have solid progress here. I do think it is going to be an definite theme of the 2020s (certainly for me personally!), A Methuselah Mouse would be good.

No Methuselah Mouse yet, but the Methuselah Foundation still exists and wants to make 90 the New 50 by 2030. There's definitely been steady progress on a number of fronts in the first half of the decade including in areas like cellular rejuvenation, so who knows?  If Amodei is right we could see spectacular progress by the early 2030s, so figures crossed. Yes, I am finally - definitely! - going to have to focus on this area in 2025. (I know, I know.)

18. New Physics. I fear we will be arguing about string theory and its lack of any empirical backing. Would be fun if the X17 particle is real, but I'd wager it isn't. What would be great would be to have a AGI come up with a Theory of Everything. (I do expect at the least that an AGI will have achieved a first in physics from Oxford by 2029.) What about Dark Matter? Or Dark Energy? Or Inflation? Physics feels as though it has run into the sand right now. Might we be in a Physics Summer by 2029?

I certainly think an AGI will have got a first in physics from Oxford by 2029, probably 2025. AGI definitely seems the most promising route to any new physics.

19. SETI. We are certainly going to know a lot more about extrasolar planets? Might we have found life on Mars? What about those disappearing planets? Fast radio bursts? Perhaps we might observe something for which the most plausible explanation is a large-scale engineering work in the universe. Or even get the Message.

It seems very hard to believe in a space opera universe teeming with intelligent life. Perhaps we might detect a biosignature in the atmosphere of an exoplanet?

20. Technological Singularity. Not, I think, by the end of 2029. But I do think we will start getting some crazy stuff after 2025. I think it might feel closer by then. But I am still expecting the Singularity at 03:14:07 UTC on 19 January 2038.

We are definitely going to get crazy stuff after 2025. The chance of an early Singularity has definitely gone up in the last five years, but 2038 still seems a reasonable date to work towards. I might even still be alive.

Well, the Razor was here on 31 December 2009 and I can only hope we are still here on 31 December 2029. We will see.

The Razor was here on 31 December 2024 at least!

Back in 2014-5, after I had been to Arvon, Simon Ings suggested a write  fix-up, which was a very good idea, and I did a series of posts of the world in 2034, 2054... which, I think still stand-up pretty well. I am currently, when things like Christmas, I know, don't get in the way doing in the Institute for the Future Coursera course on futures thinking. Their focus is typically on the ten-year horizon, so here are my Twenty for the 20s:

1. Level 10 Black Swan. Which would be an extinction event. It's quite hard to wipe out the whole human race, but not impossible. No Personal Nanofactories by 2029, so I think this won't happen.

PNFs are probably slightly more likely now than I thought in 2019. The probability of a Level 10 Black Swan - gray goo, engineered pandemic, unFriendly AGI - is definitely higher than in 2019, but still, I think unlikely.

2. Level 9 Black Swan. The collapse of the current technological civilisation. Again, quite hard to do, but again certainly not impossible. Full-scale thermonuclear exchange? Natural or artificial pandemic? Climate breakdown-induced breakdown of global food supply chains? I don't think this will happen, but, of course, that is in large part because I don't want it to.

Well, we had a pandemic. Given we are in the gravest geopolitical situation since the 1930s, the probability of a catastrophic event or chain of events seems much higher than in 2019.

3. Level 8 Black Swan. That WWI-level event that we have felt lurking our there. Given the current nature if the catastrophe, this is quite likely. You simply can't have the kinds of people in charge who we have in charge in certain countries without something catastrophically awful happening. I think there is a real chance of one of these. History might well be about to return to Terra and change all our lives forever.

COVID-19 and the Russia-Ukraine War probably count as a Level 8 Black Swan. Given what is going to happen next year, I think it unlikely that we get through the rest of the decade without at least one more event of this magnitude. Could be something technology-related.

4. Heterodox economics. There have been hints towards this including with the Labour Party, but Corbyn was never the man to sell cybereconomics and Beerism. Might Cummings be? Not sure yet about Modern Monetary Theory, but there are enormous possibilities outside the neoliberal consensus. 

Neoliberalism  has run its course, but I don't see the weak and divided left having any will or capacity to put anything innovative into action. What comes after neoliberalism, in the post-WTO, post-Washing Consensus era is mostly going to look like neoliberalism, just with higher tariffs, some onshoring, less regulation, the end ot the rule of law and massive systematic endemic corruption. My guess is that the US economy will look a lot more like the Chinese economy by the end of the decade, but on a day-to-day basis , CNBC will still feel much as does today or in 2001.

5. Neonationalism. We will see how that shows in the UK over the next four years. Can only hope that we can get our acts together to avoid the worst consequences of this. But I am not optimistic. There is that Level 8 Black Swan lying in ambush.

Right on the money on that one. the UK at least has a Labour government through 2029, but I don't see many countries in Europe or, indeed, the world, surviving the wave of Putinisation/Orbanisation we are going to see in the second half of the 2020s.

6. Scottish Independence. Both more compelling and harder to achieve than in 2014. That hard border along the Cheviots is really going to be an issue. I think they missed the Scots chance, which is both bad and good.

No-one could have foreseen what would happen to Eck and Sturgeon. Labour's inevitable failures over the next few years are lively to see something of a revival for the independence movement, but I don't know how this plays out with Orbanisation. Certainly, Scotland isn't to be independent by 2030.

7. Globalism. The African decade? We have already seen breakouts by companies like Samsung and Huawei. But what about a Nigerian or Brazilian company?

Nothing there yet as far as I know. It would be good to see something by the end of the decade.

8. A(G)I. We have had an AI Summer. Everyone is expecting an AI Winter. Clearly, Deep (Reinforcement) Learning is not the whole answer. Nevertheless, how many of us expected Go to be solved and solved in the way it was? My hope is that if we let a thousand philosophy bloom, there is a chance that some PhD student in China can come with an idea that a postdoc in Canada can leverage and suddenly open up a whole new front. Plus, we are getting the fourth edition of Russell and Norvig early in 2020. I suspect some time before the end of the decade we will have systems that meet the definition of domain-specific AGIs idiot savants/dog/cat/great ape-level AGI. And remember, humans didn't even evolve language for something like the first three-quarters of existence, which is incredibly freaky, and suggests you can glom language on top of non-linguistic GI.   

Well, we certainly didn't have the AI Winter. It's been a continuous AII and now AGI Summer. Proto-AGI systems exist and with o3 we have a system vastly more powerful than the one we knew about only two weeks ago. DeepSeek seems a huge deal, perhaps even a Sputnik-level one. NVIDIA is the second largest company in the world in terms of value on the basis of AI. We can vastly cut training time and Noam Brown, who "solved" poker and Diplomacy recently opined  “It turned out that having a bot think for just 20 seconds in a hand of poker got the same boosting performance as scaling up the model by 100,000x and training it for 100,000 times longer." Demis Hassabis got the Nobel for AlphaFold. Solve AGI and then solve everything else. Dario Amodei thinks we will have a data centre of Nobel Prize-winning scientists by 2027 that will give us 50-100 years of progress in biology and medicine (and in physics and chemistry too) in 5-10 years. This is literally the only thing that is keeping me going right now. 

9. Self-driving cars. A hard problem, but think where we were in 2009. Does not require full AGI. I hope to own one by 2029.

I saw my first one in the wild in Phoenix in March. Might require AGI! I doubt I will own one by 2029, but certainly hoping we have them in Seattle within the next year or two. 

10. Electric cars. These are already a thing in that I see at least one pretty every day just driving around Preston. I even saw an electric scooter outside a kebab shop the other day. Should be fairly ubiquitous by 2029.

There were pretty common in Central London in June 2023 because of the congestion charge. New York really needs to follow suit, but there are plenty of them in Seattle and I have been in several.

11. Hydrogen/methane economy. If we want to completely decarbonise the global economy by 2045, we are going to need a lot of local hydrogen and methane (for instance, from biodigesters for use in ships) production, so I hope to see the infrastructure for this becoming common.

I saw a hydrogen doubledecker in London last summer! There's definitely stuff going on this space, but sadly it's likely to slow over the next few years for geopolitical reasons. We shall see.

12. Fusion. There are quite a few fusion startups around. Can one make the breakthrough? This is the thing we need to solve the energy crisis fully and also to open up the Solar System.

Supposedly, there are companies that are contracted to deliver fusion reactors and this is an area that will be helped by AGI.

13. Quantum computing. This won't break encryption because we already know of types of encryption that QC can't break. It will certainly be good for simulation of quantum systems and quite possibly for various database-related applications and for brute forcing certain problems. Whether we will have desktop quantum processors by 2029, even in the lab, is another matter.

Definite progress is being made with quantum computing and the recent Google announcement is apparently a big deal, but I would like QCs to be able to do something useful, not just demonstrate a principle.

14. True Drexlerian Molecular Nanotechnology. We haven't seen as much progress in this area as we expected in the 1990s. We have seen an AI Summer and we are overdue an NMT one. It is not as though it has suddenly become a bad idea. Might well be helped by quantum computers, which will be good for simulation of chemical systems.

This is the kind of thing your data centre of Nobel Prize-winning scientists might be able to do fairly rapidly, so who knows?

15. Return to the Moon. We should get this. And a moonbase. Yeh! But will it be Moonbase Britannica with the bally Union Jack on it?

We should get this, though the moonbase might not amount to much by 2029 and there isn't going to be a Moonbase Britannica There might be by 2040.

16. Heavy Launch Vehicles. If we get the Starship and the New Armstrong , things could get quite exciting quite quickly. Orbital tourism? asteroid mining? Mars base? A golden age of uncrewed missions to the outer planets?

Should get the first orbiting of Starship in 2025 as well as the New Glenn. A Chinese HLV by 2029? No idea what the New Armstrong timetable might be. Reddit postings from a couple of months ago suggest there;'s no actual work in progress on it.

17. Life extension. Would be good to have solid progress here. I do think it is going to be an definite theme of the 2020s (certainly for me personally!), A Methuselah Mouse would be good.

No Methuselah Mouse yet, but the Methuselah Foundation still exists and wants to make 90 the New 50 by 2030. There's definitely been steady progress on a number of fronts in the first half of the decade including in areas like cellular rejuvenation, so who knows?  If Amodei is right we could see spectacular progress by the early 2030s, so figures crossed. Yes, I am finally - definitely! - going to have to focus on this area in 2025. (I know, I know.)

18. New Physics. I fear we will be arguing about string theory and its lack of any empirical backing. Would be fun if the X17 particle is real, but I'd wager it isn't. What would be great would be to have a AGI come up with a Theory of Everything. (I do expect at the least that an AGI will have achieved a first in physics from Oxford by 2029.) What about Dark Matter? Or Dark Energy? Or Inflation? Physics feels as though it has run into the sand right now. Might we be in a Physics Summer by 2029?

I certainly think an AGI will have got a first in physics from Oxford by 2029, probably 2025. AGI definitely seems the most promising route to any new physics.

19. SETI. We are certainly going to know a lot more about extrasolar planets? Might we have found life on Mars? What about those disappearing planets? Fast radio bursts? Perhaps we might observe something for which the most plausible explanation is a large-scale engineering work in the universe. Or even get the Message.

It seems very hard to believe in a space opera universe teeming with intelligent life. Perhaps we might detect a biosignature in the atmosphere of an exoplanet?

20. Technological Singularity. Not, I think, by the end of 2029. But I do think we will start getting some crazy stuff after 2025. I think it might feel closer by then. But I am still expecting the Singularity at 03:14:07 UTC on 19 January 2038.

We are definitely going to get crazy stuff after 2025. The chance of an early Singularity has definitely gone up in the last five years, but 2038 still seems a reasonable date to work towards. I might even still be alive.

Well, the Razor was here on 31 December 2009 and I can only hope we are still here on 31 December 2029. We will see.

The Razor was here on 31 December 2024 at least!

The World in 2034 Revisited

 

Back in October 2014, as I have mentioned probably several times over the years, I went to the Arvon sf course at Totleigh Baron tutored by Simon Ings and Liz Jensen. Simon told me to get The Mundanomicon finished by Christmas and then write a series of short stories that I could fix-up into a novel. Well, by this Christmas, The Mundanomicon was finished, well, certainly a lot nearer to be finished that it ever has been.  The idea for the fix-up is still a good one. Five 20,000-word novellas set every twenty years from 2034 to 2115 (sic). Could go for six and make it 101 years. Anyway, we are now halfway to 2034, so let's see how my predictions for 2034 are stacking up ten years out. 

• twenty years isn't a very long time. The worlds of 2014 and 2034 are still recognisably the same world. 2034 is 2014 turned up 11. Many people are able to lead lives that are not very different to the ones they lead in 2014. I will be 66

Yes, 2024 isn't that different to 2014 or 2004 in a lot of ways. 2034 might not be very different, but the next few years are going to be a wild ride. 2034 might be quite different to 2034, politically and technologically.

• the UK has left the EU. Scotland has left the UK. Catalonia and the Basque Country are independent. Belgium has split up. Europe is the Sick Man of the World

This was pre the Brexit referendum, so I got that right. Scottish politics didn't evolve quite like I imagined and I don't know if he will see any separatism in Europe. We will, I think, see the death of democracy in Europe with most countries thoroughly Orbanise/Putinised. We have had a massive shooting war. Future ones are possible.

• levels of discontent with the global economic and political systems remain high, but nationalism proves an ineffective approach in the face of neoliberal globalisation and no-one has yet come up with any alternatives that have gained much traction

Levels of discontent are high. I was wrong that about nationalism and neoliberal globalisation.

• Peak Oil causes a burst of inflation from the mid-2020s onwards and an energy crisis

It's odd now we worried about Peak OIl, although, sadly, Tr*mp II is going to be a deadly last hurrah for Big Oil.

• climate change leads to disruption of the water and food chains in some regions

I think a given by 2034.

• overall economic growth remains at about the average historical level, but is unevenly disrupted

Probably. Hard to know if the Tr*mp II will tank the US/global economies or a war or second pandemic cause a depression.

• some countries are trying to do to the Indians what the Indians are trying to do the Chinese. Rapid prototypers/3D printers/localised manufacturing are starting to disrupt global supply chains

We are certainly seeing onshoring, so global supply chains will be disrupted and will look very different by 2034, I suspect. Will we see the emergence of African or Latin American countries as major players in the world economy?

• a Coalition of the Willing is engaged in an endless war in Pakistan after the contents of a dirty bomb exploded in San Diego are traced to rogue elements of ISI

"Real men go to Tehran."

• electric self-driving cars are a thing

Tr*mp II means that electric cars will less ubiquitous than they ought to be, but I saw my first self-driving car in the wild in Phoenix this year and electric cars were common in London last year, so electric self-driving cars will certainly be a thing in 2034,

• Moore's law has continued, more or less and many people live in high bandwidth, data-intensive circumstances accessing augmented reality through contacts or glasses

A reasonable assumption.

• quantum computers are a thing, so are cat/dog/chimp-level AGIs, idiot savant personal assistants and Maschinenmensch-style robots

We arguably have something like cat/dog/chimp-level AGIs and idiot savant personal assistants in 2024 (I know, I know) and the Maschinenmensch-style robots are getting tasty.

• a 1968-level flu pandemic occurs. Life expectancy increases by the normal 0.2 years/years. The Queen is 106 and still the Queen

Only a 1968-level pandemic! How wrong I was there! Well, the Queen made it to 96. I thought she wouldn't die. We will probably have a king in 2034, but may be a different one to the one we have right now. We will probably get some nifty drugs over the next decade. 

• there have been several major incidents of cyberterrorism

A Cyber Pearl Harbor is a real possibility and I guess the Israeli pager attack was a kind of cyberterrorism.

• a number of arcologies have been constructed

Isn't Thiel into seasteading? Bannon, natch, was a Biosphere 2 spokesperson, so who knows?

• it looks as though small-scale fusion reactors are going to be possible. Breakthroughs in molecular nanotechnology promise and carbonoid materials promise the capability of growing cheap, efficient solar panels and strong, light smart materials

A very good chance of this by 2034. Fusion is likely to be a thing soon and AGI and quantum computers will help with drugs and materials technology.

• vanity projects funded by billionaires have put people on the Moon and Mars.

Given were we are now with Starship and M*sk, this is very likely.

Back in October 2014, as I have mentioned probably several times over the years, I went to the Arvon sf course at Totleigh Baron tutored by Simon Ings and Liz Jensen. Simon told me to get The Mundanomicon finished by Christmas and then write a series of short stories that I could fix-up into a novel. Well, by this Christmas, The Mundanomicon was finished, well, certainly a lot nearer to be finished that it ever has been.  The idea for the fix-up is still a good one. Five 20,000-word novellas set every twenty years from 2034 to 2115 (sic). Could go for six and make it 101 years. Anyway, we are now halfway to 2034, so let's see how my predictions for 2034 are stacking up ten years out. 

• twenty years isn't a very long time. The worlds of 2014 and 2034 are still recognisably the same world. 2034 is 2014 turned up 11. Many people are able to lead lives that are not very different to the ones they lead in 2014. I will be 66

Yes, 2024 isn't that different to 2014 or 2004 in a lot of ways. 2034 might not be very different, but the next few years are going to be a wild ride. 2034 might be quite different to 2034, politically and technologically.

• the UK has left the EU. Scotland has left the UK. Catalonia and the Basque Country are independent. Belgium has split up. Europe is the Sick Man of the World

This was pre the Brexit referendum, so I got that right. Scottish politics didn't evolve quite like I imagined and I don't know if he will see any separatism in Europe. We will, I think, see the death of democracy in Europe with most countries thoroughly Orbanise/Putinised. We have had a massive shooting war. Future ones are possible.

• levels of discontent with the global economic and political systems remain high, but nationalism proves an ineffective approach in the face of neoliberal globalisation and no-one has yet come up with any alternatives that have gained much traction

Levels of discontent are high. I was wrong that about nationalism and neoliberal globalisation.

• Peak Oil causes a burst of inflation from the mid-2020s onwards and an energy crisis

It's odd now we worried about Peak OIl, although, sadly, Tr*mp II is going to be a deadly last hurrah for Big Oil.

• climate change leads to disruption of the water and food chains in some regions

I think a given by 2034.

• overall economic growth remains at about the average historical level, but is unevenly disrupted

Probably. Hard to know if the Tr*mp II will tank the US/global economies or a war or second pandemic cause a depression.

• some countries are trying to do to the Indians what the Indians are trying to do the Chinese. Rapid prototypers/3D printers/localised manufacturing are starting to disrupt global supply chains

We are certainly seeing onshoring, so global supply chains will be disrupted and will look very different by 2034, I suspect. Will we see the emergence of African or Latin American countries as major players in the world economy?

• a Coalition of the Willing is engaged in an endless war in Pakistan after the contents of a dirty bomb exploded in San Diego are traced to rogue elements of ISI

"Real men go to Tehran."

• electric self-driving cars are a thing

Tr*mp II means that electric cars will less ubiquitous than they ought to be, but I saw my first self-driving car in the wild in Phoenix this year and electric cars were common in London last year, so electric self-driving cars will certainly be a thing in 2034,

• Moore's law has continued, more or less and many people live in high bandwidth, data-intensive circumstances accessing augmented reality through contacts or glasses

A reasonable assumption.

• quantum computers are a thing, so are cat/dog/chimp-level AGIs, idiot savant personal assistants and Maschinenmensch-style robots

We arguably have something like cat/dog/chimp-level AGIs and idiot savant personal assistants in 2024 (I know, I know) and the Maschinenmensch-style robots are getting tasty.

• a 1968-level flu pandemic occurs. Life expectancy increases by the normal 0.2 years/years. The Queen is 106 and still the Queen

Only a 1968-level pandemic! How wrong I was there! Well, the Queen made it to 96. I thought she wouldn't die. We will probably have a king in 2034, but may be a different one to the one we have right now. We will probably get some nifty drugs over the next decade. 

• there have been several major incidents of cyberterrorism

A Cyber Pearl Harbor is a real possibility and I guess the Israeli pager attack was a kind of cyberterrorism.

• a number of arcologies have been constructed

Isn't Thiel into seasteading? Bannon, natch, was a Biosphere 2 spokesperson, so who knows?

• it looks as though small-scale fusion reactors are going to be possible. Breakthroughs in molecular nanotechnology promise and carbonoid materials promise the capability of growing cheap, efficient solar panels and strong, light smart materials

A very good chance of this by 2034. Fusion is likely to be a thing soon and AGI and quantum computers will help with drugs and materials technology.

• vanity projects funded by billionaires have put people on the Moon and Mars.

Given were we are now with Starship and M*sk, this is very likely.

"The most affordable generative AI supercomputer"

I have bought one of these bad boys, which we discussed a while back: https://www.sparkfun.com/products/22098?nvid=em-945-13766-0005-000. Well, UI say bought, actually I have back-ordered one. I have no idea when or even if it will arrive. I also have no idea if I will be able to get it to work. For instance, it has lots of ports, perhaps it's possible to connect it to a keyboard and monitor. There's unboxing video linked to on the Spark Fun page, so at least I know it has a power cable and there's lots of documentation for it. It's primarily, as we discussed designed to be used for embedded applications, but I want to use it for experimenting with foundational models and seeing if I can do something interesting with those 67 TOPS. It will be interesting to see if I can get it to do anything, but I should learn a lot trying.

"The Place I'd Like to Live"

Many sf writers write a lot of short stories. There's even a traditional career path for an sf writer. Start with short stories for fiction fanzines, move on to better fiction fanzines, then semi-pro magazines, take an extended leave of absence to attend Clarion, get published in pro magazines, then anthologies. Get picked up for some best of year collections, perhaps get nominated or even win a prize, get an agent, write two novels on spec,rewrite novels to agent's direction, rewrite novels to publisher's direction, get a two-book deal, get a longer deal, give up your day job and move to a big house in a national park and enjoy the well-earned fruits of your success and the admiration of your peers and fans.

Banksie wrote very few short stories. He had one collection, The State of the Art most of which is taken up by the eponymous novella, "The State of the Art". Two later stories published by the Birmingham SF group were excised sections from novels. Only one of the stories in The State of the Art was published in an sf magazine (Interzone). There's a couple of Culture stories and it's possible that some of the stories were drawer stories; we know "The State of the Art" was as well as several of Banksie's early sf novels. I suspect all of the rest were commissioned. An interesting question is why he didn't write short stories in the 1970s. Perhaps he did. We know he was assiduously practicing his craft from the mid-1970s. There wasn't much of a market for sf stories in the UK in the mid-1970s. There was the New Worlds paperback series as well as New Writings in SF. We know Banksie and Ken McLeod read New Worlds because they were heavily influenced by M. John Harrison's critical writings there. Getting published in NW or NWiSF might have been about being part of the gang, but Banskie and Ken could have hung out at the One Tun and tried to inculcate themselves with Moorcock. Perhaps they were too busy writing or drinking. The 1970s were very different in terms of the discoverability of things, but there were three secondhand sf bookshops in London the as well as new book ones like  Dark They Were and Golden Eyed. We know Ken was hanging out at the libertarian bookshop and, of course, this was the heyday of Compendium Books. It was easy enough to find weirdness in 1970s London, so we can assume that Banksie and Ken knew about fandom and US sf magazines. 

As Stephen Gallagher once opined, it's better to practice writing short stories because if you write 20 5,000-word stories you can see that you are getting better whereas one 100,000-word novel is likely to be a artistic failure and you have to start again from scratch. It took me 15 years to write The Mundanomicon (116k words) and it's really supposed to be the first part of a trilogy. But Banksie were much focussed and talented than me. Perhaps the novel was Banksie natural length. The curio in his case is that having not nowhere initially with his sf novels, he broke through with a mainstream on and then was able to get his sf drawer ones published. So, don't throw away those old MSs!

As for my there are no short stories after the 1980s, he was well enough established and probably rich enough that he didn't need the money or egoboo from a commissioned piece. David suggests Civilization as a reason too (I am guessing Banksie said this in a talk some time) and that seems deeply plausible as we know Banksie liked games (Empire! might be like a bit like the boardgame Civilisation, which I had and is presumably still in Preston) and, as I can attest from 1992, nothing has ever been able to eat time more efficiently than Civilization. So, if it weren't for Sid Meier, we might have had more Banksie short stories!

"My Secular Heaven"

I got a copy of Banksie's The Culture The Drawings a while back and I was looking at this morning in my study when I couldn't sleep. It's a reproduction of the background material that he produced for the Culture series, probably mostly, if not entirely in the 1970s. Of course, what one is overwhelmingly reminded of, is one's own RPG notebooks, although Banksie's are vastly more elaborate and original than anything I did. It also helps that he was better at drawing than me. Now, this feels like source material for an RPG, which makes it surprising that there has never been a Culture RPG. Surely Banksie was asked about it. I know that he didn't want the material in the sketchbooks released until after the series was finished, but there was plenty of material in the novels that could adapted for the game. And here's the thing, we know that Banksie was interested in games.

The Steep Approach to Garbadale features a family whose fortune is predicated on a boardgame, Empire!, that presumably occupies a Monopoly-like cultural niche, but is presumably more like Civilisation and Diplomacy and has actual deep and interesting strategy. In an email to the Weasel I opined "The boardgame idea seems very "us": imagine a cross between "Monopoly" and "Risk", but good." In a later missif, the Weasel stated "The game is, sensibly, not described very much allowing one to fill in the blanks. It is described as allowing you to fight and trade and preach and bluff, and having cards and pieces, also that for the Far East it was increased to allowing 8 players. I am guessing Risk meets Civilisation meets Diplomacy." There is also 1D chess in Walking on Glass and, of course, Azad in The Player of Games. So, Banksie's interest in games is amply attested to. We also know he was a self-confessed war nerd or at least the self-insert in Dead Air is, and we can imagine the impact that Six Days War had on an impressionable 14-year old, the son of a naval officer. It seems inconceivable then Banksie didn't have Airfix models and was unaware of wargaming. Surely, he would have come across RPGs in London in the late 1970s. What would he have made of Traveller in 1977? It might have been too small-scale for him, but I would still have thought there was much of interest to me there and certainly by the time the later supplements are being published.

It's hard to imagine no-one ever came up to him at a con and asked if they could have the RPG rights. We know, for instance, that Pratchett was a RPG player and I understand when he visited the sf group at university he talked for hours about them at the after-party and there was a Pratchett GURPS supplement as well as the very cent Discworld RPG.GURPS would have been perfect for the Culture because it's just adding a supplement rather designing a game from scratch. Did Banksie say no? Did someone obtain the rights, but never finish the game? There's never been anything to stop someone doing their own DIY Culture campaign and there's some discussion about this online. T5 might allow the tech levels to go high enough, but Mindjammer and Sufficiently Advanced are also suggested as suitable systems. that are strongly flavoured by the Culture. I had never heard of SA, although it definitely looks like my kind of thing. We might get a Culture RPG in the future, but the intriguing question is why haven't we had one already? Why not Banksie not want one? Perhaps one day that question will be answered.

“And incidentally – a happy Christmas, to all of you at home”

Normal service will be resumed tomorrow. I suspect that "Joy to the World" will be more a lot fun than "The Feast of Steven", although I wish it were not lost. We've also got the new Wallace and Gromit to look forward to. Did I say I went to the same school as Nick Park?

Source: https://archivetvmusings.blog/2014/12/10/doctor-who-the-feast-of-steven/

The Future is Another Country

They do things differently there: A(G)I, climate breakdown, the geopolitical crisis, mRNA, new diseases, "populism", reusable heavy lift vehicles... It's going to be a wild ride. Of course, the Tofflers were talking about future shock in 19770. It was the first book I read in 2001. Very Brunner. If I recall correctly The Shockwave Rider is heavily derived from it.Has the future always looked exponential. It certainly did to Henry Adams who, as he have discussed, discovered both "Moore's Law" and the Singularity (phase change in history). Of course, Adams was writing in the late C19th/early 20th; the world looked a lot by then than it does now. (I will discuss Adams more when I eventually get round to writing the "Electrical Sciences" article.) Virginia Woolf said "On or about December 1910 human nature changed." Of course, by humans, she probably just meant the Bloomsbury Set, who were the only people who counted. But perhaps we might "On or about December 2024 human nature changed," what with o3 and Tr*mp II and M*sk. It might well change a lot more times by January 2038.

Eleventh Wrangler

In Littlewood's Miscellany, which I read when I was at QMW, J.E. Littlewood says that for the, I think, 1904 Cambridge maths tripos, to the examiners' book for which he had access, if you scored 100% on the bookwork and 20% on the riders, you would have ranked as 11th wrangler and beaten John Maynard Keynes, who was 12th. (Keynes sat the tripos in 1904, so it would make sense for the Littlewood to have used the 1904 book, but perhaps it was some other year and he was just making the point that that level of performance would put you ahead of Keynes). The point being that bookwork is hard (there's a lot of algebraic manipulation and doubtless geometric manipulation and probably arithmetic) and thus having a system (a reasoning model) that can do the bookwork would be a significant advance. And, indeed, the notion that you can just input a PDF of a TeXed exam question and get a system to answer it correctly would have seemed absurd a decade ago Now, of course, we have had systems like Mathematica for decades, but they are solving specific equations we have inputted, not solving ab initio mathematics or physics exam questions formatted for humans. So, it seems to me that a system that could get a first on the 1904 or 2025 Cambridge maths tripos would be doing something interesting. "Oh, it's just scrapped it off the internet!" Why then have LLMs up to now been so bad at maths then?Feeding next year's exams into a reasoning model and seeing how it does would be a interesting experiment at the very least. And, as I have said, that would apply equally to PPE or Literae Humaniores.

 

Chollet says of o3 on blog:

 

Of course, such generality comes at a steep cost, and wouldn't quite be economical yet: you could pay a human to solve ARC-AGI tasks for roughly $5 per task (we know, we did that), while consuming mere cents in energy. Meanwhile o3 requires $17-20 per task in the low-compute mode. 

 

And high-compute mode is probably x172 that. How many person-minutes you get for $5 will vary a lot depending on the person. Let's say $150/hour, $5 is then 2 minutes of time. A 3-hour exam would then be 90 tasks. At x172, that's 344 minutes of time, nearly 6 hours or twice as long as the exam. So, an interesting question is how much compute would o3 take to do an Oxbridge maths/physics papers? As Dave Clements said exams are not a good model of the world and that a better proxy would be developing a good Python 4 language. I think LLMs could do that or at least assist in doing much more rapidly than a team of people could do it, but I suspect that the first reasoning model is awarded a Oxford first, unofficially or officially, probably this summer, the goalposts, as ever with A(G)I, will be moved again.  

Above the Senior Wrangler

o3 appears to be good at maths: 

o3 scored 96.7 percent on the 2024 American Invitational Mathematics Exam, missing just one question. The model also reached 87.7 percent on GPQA Diamond, which contains graduate-level biology, physics, and chemistry questions. On the Frontier Math benchmark by EpochAI, o3 solved 25.2 percent of problems, while no other model has exceeded 2 percent.

The American Invitational Mathematics Exam is an exam for US high schoolers, which you get to do by scoring well in other maths exams. I have no idea how hard it is. Does it compare to Maths S-Level or 1980s Oxbridge entrance exams? It might be different. GPQA is a graduate-level Google-proof Q&A benchmark.

We present GPQA, a challenging dataset of 448 multiple-choice questions written by domain experts in biology, physics, and chemistry. We ensure that the questions are high-quality and extremely difficult: experts who have or are pursuing PhDs in the corresponding domains reach 65% accuracy (74% when discounting clear mistakes the experts identified in retrospect), while highly skilled non-expert validators only reach 34% accuracy, despite spending on average over 30 minutes with unrestricted access to the web (i.e., the questions are "Google-proof"). The questions are also difficult for state-of-the-art AI systems, with our strongest GPT-4 based baseline achieving 39% accuracy.

So, 87.7% is much better than GPT-4 and better than most PhD students. Of course, people like Marcus are claiming that they are cheating by scraping stuff off the internet. Perhaps, but would that explain such a big leap in performance? You might claim that it can answer questions that are like ones in its training set, but surely that's a case of well, yes, so? But what I want to know is how well do these systems do against a classic proxy for human intelligence, Oxbridge maths and physics exams. Would o3 get a first in mods and finals? There are lots of Oxbridge past papers, but I don't think many are online at least not with solutions. Somewhere, probably in Preston, I have got all of the Oxford physics finals papers from 1973, when the format changed, to 1988 (I wonder if I kept my 1989 papers). How would o3 do on them? Given the nature of the questions (each one involves solving a problem as well as bookwork), it would be fascinating to know. I suspect I might get the chance to ask a reasoning model these questions some time in 2025. Will we see the first reasoning model to be placed above the senior wrangler next year?

 

The New New Thing

It's a big deal that François Cholletthinks that o3 is a big deal. Chollet invented Keras and literally wrote the book on deep learning. He has been a noted sceptic about foundational models because they weren't very good at all at some tasks that people are good at. Now we have reached an epoch at which

OpenAI's new o3 system - trained on the ARC-AGI-1 Public Training set - has scored a breakthrough 75.7% on the Semi-Private Evaluation set at our stated public leaderboard $10k compute limit. A high-compute (172x) o3 configuration scored 87.5%.

There are some tasks that o3 still is very good at and ARC-AGI-2 tests are ready to be rolled out and ARC-AGI-3 is under development. It will be interesting to see how quickly other reasoning models catch up. However

Techmeme

‪@techmeme.com‬

Sources: OpenAI's GPT-5, codenamed Orion, is behind schedule and faces technical hurdles, including high computing costs and limited high-quality training data (Deepa Seetharaman/Wall Street Journal)

Of course, it could be a case of that it doesn't really matter and that the energy will shift from LLMs to reasoning models that don't require as much data to train on and can iterate much quickly. Whatever way, it looks like the next few months are going to be a crazy ride and it's entirely possible that we will have something like AGI by 2026.

 

o3 uses uses thousand of dollars of compute per task. But what is as task? If a task is the equivalent of a day's or week's work then thousand of dollars is the equivalent of a what you pay a human expert. So, the potential for this to have a huge impact economically can already be seen.