[jamest (OP)]

Geoff asks, 'Are there any ways now, or anticipated, in which AI will cause or encourage actual reductions in greenhouse gas emissions and lessening of environmental degradation? Or is AI all negative on the environmental front?'

[paul cotter]

Personally I believe it's overall effect will be negative, based on the prodigious power consumption of the associated data centres. AI operates by scraping the web for information and can be useful in getting a quick answer with the proviso that the answer may be right/wrong- that info is already in existence and can be accessed by conventional methods anyway. Overall I see a definite negative with some possible very minor positives.

[alancalverd]

Like almost any tool, you can use it constructively or destructively. Thanks to advances in facial recognition it now takes me longer to file my company accounts via positive identification, and all company directors are regarded as guilty until proven innocent of financial crimes. Yesterday I had to criticise a research proposal that would have involved putting photographs of people's eyes on the internet because iris recognition now compromises patient confidentiality. My car still misreads speed limit signs and tells me to slow down (to the annoyance of following traffic) when there is no need to do so.

So AFAIK, the answer seems to be that AI is a time- and energy-waster at best, and dangerous at worst.

[evan_au]

More efficient AI processors
I saw an estimate that AI capabilities is doubling about every 7 months.
- Fortunately, AI power consumption does not seem to be doubling every 7 months
- Part of this is due to efforts by NVIDIA et al to increase the efficiency of new AI processors (other brands are available)
- I understand that one of the early cars with "self-driving" capabilities consumed about 1kW of power - much more than the roughly 25W of a human driver's brain. So efficiency improvements are possible!

New Chemical Processes?
There has been progress in using AI to predict how proteins will fold, and interact with other chemicals (this won a Nobel Prize in 2024).
There is a possibility that in future, custom proteins could be used to:
- Break down chemical pollutants
- Generate electrical power from photosynthesis
- Capture and store CO2

Local Weather Prediction?
There have been claims that AI software on a desktop PC can produce weather predictions for a particular locality much faster than a supercomputer modeling the entire planet using Navier-Stokes equations.
- Of course, a significant fraction of that computer power is still required to integrate sensor measurements to define the weather now; this is needed for input to the PC AI model.

Controlled Nuclear Fusion?
Containing a hot plasma has been a quite intractable problem.
- Modelling plasma using magneto-hydrodynamic software is processor-intensive, and can't readily be done in real time.
- Perhaps AI may be able to assist with plasma management? (definitely speculative!)

These could be useful for improving the environment.

[paul cotter]

Capturing and storing CO2 is already possible but it is an inherently energy intensive process. I don't see how proteins could improve the energetics involved. Some proteins make excellent catalysts but a catalyst just speeds up/enables a reaction, it does not change the energetics.

[alancalverd]

Trees capture CO2 brilliantly, using solar energy only, and no transistors, so don't need cooling.

If you need energy, you can cut down a tree: dry wood has about half the energy density of natural gas and ten times that of a battery, so an ideal road fuel!