In Intelligence Explosion analysis draft: introduction, Luke Muehlhauser and Anna Salamon wrote

Svante Arrhenius' (1896) models of climate change lacked modern climate theory and data but, by making reasonable extrapolations from what was known of physics, still managed to predict (within 2°C) how much warming would result from a doubling of CO2 in the atmosphere (Crawford 1997).

As a part of the project "Can we know what to do about AI?", I've summarized my initial impressions of Arrhenius's predictions and the impact that they might have had. The object level material is all draw from Wikipedia, and I have not vetted it.

  • Arrhenius's chemistry was sound: the equation for how the Earth's temperature varies as a function of concentration of carbon dioxide is the same equation used today.
  • For the most part, Arrhenius didn't model how increased carbon concentrations would impact other factors that influence the Earth's temperature. I don't know if this is because he wasn't aware of these, because he thought that they were sufficiently small to ignore, or because he didn't try to.
  • Knut Ångström criticized Arrhenius's claim on scientific grounds, giving a different model which predicted no climate change from increased carbon concentrations. My surface impression is that Arrhenius was a much more accomplished scientist than Knut Ångström was. To the extent that this is true, I think that Ångström's view should be heavily discounted, but I haven't investigated further.
  • While Arrhenius recognized that the use of fossil fuels could increase atmospheric concentrations, he underestimated how fast carbon emissions would increase (by a huge margin) because he didn't recognize how widespread fossil fuel use would become.
  • People later thought that Arrhenius's prediction that atmospheric carbon would increase was wrong, because they thought that oceans would serve as great carbon sinks. It would be interesting to look into whether they had good reasons for thinking this at the time.
  • Arrhenius predicted that global warming would have positive humanitarian impacts on balance, global warming now appears to have negative humanitarian impacts on balance. 

Taking this all together, based on my surface impressions, I think that this case study gives evidence against attempting to predict the far future being useful:

  • To the extent that Arrhenius was right, he was largely ignored. 
  • Arrhenius could have been wrong (the countervailing theories could have been right), but this warrants further investigation. 
New Comment
23 comments, sorted by Click to highlight new comments since:

It would be interesting to look into whether they had good reasons for thinking this at the time.

Henry's Law? Correct me if I'm wrong, but the oceans hold something like 98% of the CO2 in the biosphere, so it's easy to predict that in equilibrium they'll absorb something like 98% of the fossil carbon we add to the biosphere. It might not have been as easy to predict the transient behavior; IIRC the oceans are only absorbing CO2 at roughly a third of the rate at which we're now emitting it, just because the mixing processes are so slow.

Thanks.

It seems to me there is a difference between doing science (in particular, figuring out the ways in which the physical world works) and making far-future predictions.

When Newton described the law of gravity he did not make far-future predictions about objects falling. He just discovered a law of nature. In the same way, Arrhenius was trying to figure out how the link between atmospheric CO2 and global climate works. It's just normal science and as such doesn't offer evidence for or against far-future predictions.

Oh, and by the way, whether global warming has positive or negative net impact critically depends on its magnitude. Minor global warming (up to about 2 degrees C, I believe) is commonly considered beneficial.

When Newton described the law of gravity he did not make far-future predictions about objects falling. He just discovered a law of nature. In the same way, Arrhenius was trying to figure out how the link between atmospheric CO2 and global climate works. It's just normal science and as such doesn't offer evidence for or against far-future predictions.

There might be, e.g., economic laws that can be thought of as "science" that have relevance to predicting things related to artificial intelligence.

Oh, and by the way, whether global warming has positive or negative net impact critically depends on its magnitude. Minor global warming (up to about 2 degrees C, I believe) is commonly considered beneficial.

Can you give a reference?

I don't think economics is relevant -- Arrhenius was doing "standard" hard science and not predicting what human societies might or might not do. The laws of economics are quite different from laws of nature.

For positive net impact see e.g. the Stern Review. The main factors are increased agricultural productivity (because of CO2) as well as the reduction in winter heating and winter-related deaths.

I don't think economics is relevant -- Arrhenius was doing "standard" hard science and not predicting what human societies might or might not do. The laws of economics are quite different from laws of nature.

Even so, Arrhenius's successful prediction still constitutes a weak argument for it being possible to predict the future.

I don't think anyone is contesting that it's possible to predict the future.

The real issue here is making good far-future forecasts concerning things (or ideas, patterns, arrangements, etc.) that do not exist yet -- and here I don't think the Arrhenius example provides even a weak argument.

As I understand it, it mainly has to do with the speed. Too fast, and plants (and some animals) can't keep up with their temperature zones.

I think that this case study should cause one to update away from predicting the far future being useful

This should be a pretty tiny update, though, since it's only one case study, and it's a mere surface impression.

Changed to

I think that this case study gives evidence against attempting to predict the far future being useful

I think that this case study gives evidence against attempting to predict the far future being useful

Isn't it evidence for a more narrow conclusion? Specifically, that attempting to precisely predict the far future isn't going to work well? Arrhenius was correct that climate change was going to happen, just wrong about several particulars. Or was he right for so many wrong reasons that it's not worth noting?

Here by useful I meant "prescribing actions that turn out to have social value." I agree that Arrhenius had the right general idea, and that this is noteworthy.

"People later thought that Arrhenius's prediction that atmospheric carbon would increase was wrong, because they thought that oceans would serve as great carbon sinks. It would be interesting to look into whether they had good reasons for thinking this at the time."

The oceans contain about 50× as much CO2 as the atmosphere, and they're estimated to be removing CO2 from the atmosphere about 1/4 as fast as mankind is adding it. (The terrestrial biosphere is also estimated to be removing CO2 from the atmosphere, through "greening," about 1/4 as fast as mankind is adding it, so if anthropogenic CO2 emissions fell by more than half then atmospheric CO2 levels would be falling, rather than rising.)

So, the principle was correct, it's just that those "negative feedbacks" are not keeping up with the high rate at which mankind is emitting CO2.

"Arrhenius predicted that global warming would have positive humanitarian impacts on balance, global warming now appears to have negative humanitarian impacts on balance. "

No, Arrhenius was right even about that.

The best evidence is that manmade climate change is benign, and CO2 emissions are very beneficial, for both mankind and natural ecosystems. I’m one of over 30,000 American scientists who’ve signed a “petition” attesting to that fact.

CO2 is “plant food.” Are you old enough to remember when terrible famines were often in the news, in places like Bangladesh? Throughout history, famine was one of the great scourges of humanity, the Third Horseman of the Apocalypse... until now.

Thankfully, famines are becoming rare, and one of the reasons is rising CO2 levels, which have increased worldwide agricultural productivity by about 20%.

http://tinyurl.com/1920sciamCO2

http://co2science.org/data/plant_growth/plantgrowth.php

If we lacked that additional productivity, mankind could approximately make up the difference by putting all of the world's rainforests under cultivation.

Those benefits are well-measured, by thousands of agricultural studies. The supposed major harms are all just hypothetical.

Rising CO2 levels not only make plants healthier and more productive, they also make them more water-efficient and drought-resistant. One of the consequences is that the Earth is "greening," especially in some arid regions, like the Sahel (southern Sahara), as reported in National Geographic:
https://www.sealevel.info/Owen2009_Sahara_Desert_Greening-atGeo30639457.html

EXCERPT:
“Images taken between 1982 and 2002 revealed extensive regreening throughout the Sahel, according to a new study in the journal Biogeosciences. / The study suggests huge increases in vegetation in areas including central Chad and western Sudan. ...
“’Before, there was not a single scorpion, not a single blade of grass,’ he said. “’Now you have people grazing their camels in areas which may not have been used for hundreds or even thousands of years. You see birds, ostriches, gazelles coming back, even sorts of amphibians coming back... The trend has continued for more than 20 years. It is indisputable.’”‍‍‍‍‍‍

We are currently enjoying what has traditionally been called a "climate optimum." It is generally acknowledged that the Earth has warmed, on average, about 1°C (±0.4°C) since the "pre-industrial" Little Ice Age, and it is also generally acknowledged that the warming has been beneficial. Yet activists claim that warming more than another 0.5 °C would be harmful, rather than beneficial. That claim is not based on scientific evidence, and is not credible.

The direct impacts of global warming are obviously minor. Except at high northern latitudes, where "Arctic amplification" makes the brutal winters there a bit milder, we’re on track for at most about one degree of additional warming by 2100 (and even that assumes no large-scale transition to nuclear energy). At temperate latitudes, that’s like moving only about 70 miles south, or planting only about one week later.

So folks promoting solar and wind boondoggles hype other supposed harms, like sea-level rise, or extreme weather, or polar bears' peril. But those problems aren’t actually happening.

CO2 has been rising steadily for 2/3 of a century, yet sea-level rise has not detectably accelerated:

https://sealevel.info/1612340_Honolulu_Wismar_Stockholm_vs_CO2_annot3.png

https://sealevel.info/MSL_global_thumbnails5.html

Hurricanes & typhoons are not worsening:

https://sealevel.info/frequency_12months_2018-09-30_with_trendlines.png

https://sealevel.info/global_running_ace_2018-06.png

Strong tornadoes have actually declined:

https://sealevel.info/EF3_to_EF5_tornadoes_historical_woodtv_dot_com_annot1_1006x759.png

And the polar bears are fine:

https://polarbearscience.files.wordpress.com/2015/05/crockford-unofficial-polar-bear-numbers-to-2015-sept-1-final1.jpg

Here's what Arrhenius wrote:

https://sealevel.info/Svante_Arrhenius_p63_excerpt_touting_the_good_news_of_global_warming3.png

Notice that he even predicted "polar amplification."

The best evidence indicates that he was exactly right.

Arrhenius predicted that global warming would have positive humanitarian impacts, for reasons that are retrospectively wrong, and instead global warming appears to have negative humanitarian impacts.

This seems like an oversimplification. It would seem more accurate that Arrhenius thought on balance it would have a net positive humanitarian impact and current consensus is that it is net negative.

Thanks, I fixed this.

Do we know whether Arrhenius was right at the time that it was a net positive? 1896 was quite a long time ago, as far as populations and economies go.

No idea. I have no idea if anyone has looked into this in that much detail. My guess would be that he was wrong, since agriculture would have been even more insensitive to change in many parts of the world (although less so in other areas since monocropping wasn't as common).

He was writing soon after the end (in retrospect) of the Little Ice Age, which I think is generally agreed (then and now) to have had negative effects. So a predicted return to warmer temperatures would be a good thing - it's not called the Medieval Climate Optimum for nothing!

The difference between the Optimum and the LIA was less than 1C, so saying warming up another 1C would be even better was extrapolation. But Arrhenius probably didn't have historical climate data of that precision, and it may have been the reasonable prediction to make.

The LIA and the Optimum weren't global - some places cooled down, others heated up. (As is true of most climate change). But again, Arrhenius wouldn't have known that. The last 800 years of history up to his time, on either side of the North Atlantic, were of gradual cooling down with deleterious effects.

Yes, the Little Ice Age was part of what I was thinking of, although I was unsure it was really relevant - the Little Ice Age should've ended at least 50 years before Arrhenius seems to have done his relevant work.

[-]Dentin-30

One must be very careful here: Arrhenius is almost certainly using a different utility function for 'positive' than current consensus uses today. As I understand it, fewer people die yearly of heat than freezing and warming opens more land than it closes off, but in today's environment more concern is given to things like "OMG GLOBAL WARMING" than actual data.

[edit]I was not referring to temperature or the temperature calculation model. I was referring to the net positive or negative effect on society. Sorry for the confusion.[/edit]

Are you trying to make sure we don't inadvertently discard [hypothesis: unusual utility function]? Well I'll say the same thing about [hypothesis: Arrhenius simply had a bad model even though his utility function was not terribly different from ours] From the wiki ( http://en.wikipedia.org/wiki/Svante_Arrhenius ) "In his calculation Arrhenius included the feedback from changes in water vapor as well as latitudinal effects, but he omitted clouds, convection of heat upward in the atmosphere, and other essential factors. His work is currently seen less as an accurate prediction of global warming than as the first demonstration that it should be taken as a serious possibility."

-sounds like he used a simple, fast model, rather than a detailed one that only a cray-4something supercomputer could run in less than a year. All he has to do is neglect, say, storm damage for his model to feed the wrong results to his final utility function even if he manages to predict the correct temperature.

[-]Dentin-10

No, I'm not saying anything at all about temperature or the model; I was talking about the social effects, eg 'positive effect on society'.

Positive and negative in this day and age is dominated by public opinion and is very different than what it was back then. His view back then could have been as simple as "fewer people will freeze to death and there will be more arable land and better crops". Ours view today marginalizes those effects and seems almost entirely based on the idea that change of any sort is negative.

Oh, I’msure he gave different weights to different things in his utility function than say…well pretty much anyone other human…but there are plenty of models that show a disaster for any “typical” human utility function. The ones showing disaster: venus and disaster: new ice age…are not exactly rare, though I’m not exactly sure how seriously to take them myself.
"Positive and negative in this day and age is dominated by public opinion"

Relying on Public Opinion is a cheap and dirty variant of Auffman's agreement theorem; it gives plenty of bad results, but it's a million times easier to use, and is still slightly better-maybe-than pure random-decisions....er, maybe.

Either that or we just differ in terms of what we're labeling utility function versus part of the model?