[egnehots]

well it's human nature to struggle with collective action when the risks are unclear, vague and not shared. stakeholders are juggling immediate, tangible concerns, like climate change, economic stability, and political issues, making it tough to justify moving AI up the priority list.

[ben_w]

Extra fun: AI can make a big (+ and -) difference to climate change[0], mess with the economy[1], and get used as a tool to sow political chaos[2].

But sure, humans are necessarily very myopic, it's necessary that we ignore 98% of the issues in the world or we wouldn't be able to even function.

[0] High power use, can help roll-out renewables and storage

[1] What happens when those humanoid robots we see demos of, get good enough to replace all the staff in the factories where they get made? And the rest of their supply chain?

[2] Imagine if the pizzagate conspiracy theorists had had access to an un-censored sound-and-video GenAI tool

[janalsncm]

I would like to learn more about what AI can do specifically to solve the climate crisis.

My guess is that a lot of the actual lift would come from industrial automation to create cheaper green products. I guess that is “AI” in some sense.

But if we are building solar panels, the R&D budget should be put towards streamlining the build process. Figure out how to commoditize solar panels so that oil is too expensive.

Building huge “foundation” models like I see huge AI labs doing is a bit like building better visualizations of an impending asteroid impact. It’s not really what we need right now.

[ben_w]

> I would like to learn more about what AI can do specifically to solve the climate crisis.

Robots that install PV. I think there's even an YCombinator startup doing exactly that? Needs a higher degree of AI to function outdoors than in a nice fixed factory setting.

(Especially if they also drive the trucks containing the PV, but who knows how long we'll have to wait for that AI…)

> Figure out how to commoditize solar panels so that oil is too expensive.

Good news: it already is :D

> Building huge “foundation” models like I see huge AI labs doing is a bit like building better visualizations of an impending asteroid impact. It’s not really what we need right now.

Mm. These things give increasing levels of generalisability: the biggest weakness of previous models, was that you could train them to learn one thing very well, and they'd suck at anything else.

In one sense that's still true of the new models, it's just that they're being trained to build a world model from most written text and most pictures and most videos, so they have a very broad range of things they're OK at — much less likely to be confused by a rattlesnake they mistake for a powerline, for example.

[southernplaces7]

>imagine if the pizzagate conspiracy theorists had had access to an un-censored sound-and-video GenAI tool

They didn't but a huge number of other conspiracy theorists still running their own ideas do have access to all that with today's AI, and we don't see a vast watershed of billions of people being brainwashed into believing complete nonsense to any degree greater than has already been the case for a long, long time before AI came along.

People do have a certain level of discernment, even when absolutely bombarded with propaganda and fakery. Usually, it seems to take, finally, coercion to make them simply swallow too much of something obviously absurd. This too was the case before AI and, simultaneously now, widespread access to information sources that let you verify the veracity of nearly anything you like in minutes as long as it's not grossly complex to untangle.

Even the Nazis of the 1930s and the bolsheviks earlier, despite all their mass efforts at convincing through propaganda and misinformation (applied to people with less ability than today to find contrary sources of information) ultimately didn't convince as many as they'd have liked voluntarily. They had to coerce them into just never openly disagreeing.

I don't think we're in danger of AI by itself doing anything major to suddenly make billions of people behave much differently in their beliefs from how they already have for centuries at least.

[ben_w]

> They didn't but a huge number of other conspiracy theorists still running their own ideas do have access to all that with today's AI

Current video tools are easily distinguishable from reality if you pay attention. Sliding feet, distorted geometry, occasionally even lacking object permanence.

They're improving rapidly and I have no reason to expect this is the best they'll ever be.

Even still-image tools often still generate things with fundamental errors that can be noticed, and despite this they are already being used because "you can fool all of the people some of the time, and some of the people all of the time"; even the current tech still moves the needle on that.

Of course, the (preexisting) mere existence of the possibility is itself a convenient source of deniability for anything that you don't want to believe — I wonder how many people refuse to believe that Trump really stored boxes of classified documents in a bathroom in Mar-a-Lago despite the photo?

> Even the Nazis of the 1930s and the bolsheviks earlier, despite all their mass efforts at convincing through propaganda and misinformation (applied to people with less ability than today to find contrary sources of information) ultimately didn't convince as many as they'd have liked voluntarily. They had to coerce them into just never openly disagreeing.

And the former won power in the first place in a democracy.

Likewise Rwanda, the violence followed from the propaganda.

Conspiracy theories don't need everyone to believe in them to cause problems — that's why I gave the Pizzagate example where (IIRC) the biggest harm was someone firing a gun in the restaurant demanding to see the basement it didn't have.

[southernplaces7]

>And the former won power in the first place in a democracy.

No actually, despite all their massive propaganda drives, the Nazis never won through elections. They just could't get enough votes to do so and instead used backroom maneuvering with other established politicians to gain the chancellorship appointment (not election) for Hitler so that he could use that to manipulate existing laws into forming a one party dictatorship.

The relevance of this is there for my earlier point: Even in such a case, the real danger was government and its legal powers to suppress, coerce and repress, not so much an organization's ability to spew out bad information and propaganda.

[dragonwriter]

> No actually, despite all their massive propaganda drives, the Nazis never won through elections.

a Nazi-DNVP coalition did, in fact, secure a majority in the snap elections called immediately after the one in which the Hitler-led Nazi-DNVP minority coalition came to power, and it wasn't until after they did the Hitler established totalitarian control.

(Of course, winning the largest share of seats, and forming a minority coalition because no group of other parties can form a majority coalition and you are able to secure the support of the elected head of state is also winning through elections, even if it isn't a commanding mandate. Systems of choosing a government by elections that can have ambiguous results usually have a set of methods of resolving them, and securing the position in the election from which to emergy victorious from the ambiguous-results-resolution system is still winning through elections.)