I found the discussion of AI's energy usage to be frustrating. Of course, a single query's energy demand may be negligible, but on the whole it is indisputable that centers are using up a larger and larger portion of our finite energy supply. This is causing energy costs to rise in places where data centers are putting more strain on the grid. Tech companies have no market incentive not to continue to use large amounts of energy, a public good, which is the definition of a negative market externality. The problem needs to be addressed. Mat would, it seems, have the public foot the bill to expand energy supply (and increase fossil fuel emissions) so that tech companies can build more data centers?
You raise a few separate issues. Consumer AI use is negligible as a driver of energy demand compared to training and enterprise stuff, and there is more legitimacy to the poor data center policy that has to do with states competing to attract business and poor infrastructural capacity to handle it, which is a government oversight. AI companies are diversifying their data center builds overseas due to an inability for the US to be competitive - as you said their incentive is to pursue cheap reliable energy, so poorly prepared states are offering incentives to keep that custom local before they can handle it.
The extent to which those costs are being socialized, which is a real problem, is a state failure they are now trying to address. So tldr the US needs to find ways to be competitive in this new economy, and are doing so chaotically. AI development is not going to stop in any case. There is also the sticky question of what often economically dormant areas in flatlands would do otherwise - there is a reason many of those areas are trying to court economic activity, and positive and negative examples.
The fossil fuel issue is where I think my Koomey recommendation is instructive as to the relative total energy burden of data centers and emissions, and often ridiculous projections of future demand. I think it is legitimate for local data centers and potential negative externalities of those builds to be scrutinized, but my feeling is we need more energy and better planning, as the trade off of losing that business to elsewhere will also affect people.
AI accelerationists seem to want to skip to a "Star Trek" utopian tech culture where everyone spends time programming their personal holodeck experiences, without the grueling political work and beyond moon-landing level advancement in political technology that will be required to make that reality benefit the 99%. Tech advances historically don't often lead to positive political advances - in fact quite the opposite, particularly in the short term. Luddism from a correct historical perspective should not be derided as anti-technology or analogue nostalgia. Rather, it is about proactively neutralizing the novel weapons of class warfare being used against labor. I rarely hear AI optimists acknowledge that the ethics of using AI are entirely contextual to the political framework.
I dont see how advancing and experimenting in the AI space is an ethical endeavor for artists and media creators unless they have a actionable strategy for using AI to directly effect positive political change. "Break some eggs to make an omelet" is mostly what I am hearing - ignoring the historical lessons of the industrial revolution, social media and other cataclysmic shifts in which untempered tech acceleration resulted in massive human strife, intractable political destabilizations, and dangerous consolidations of wealth and power. At the current course, I think it will be generations before generative AI use in the media and arts will have a net benefit to human wellness, and the faster we push the tech, the longer the timeline could be. Nevermind that like forever chemicals and social media, we have very little research into how widespread AI use might harm human development and psychology. Being exposed to generated falsehoods and artificial realities on an hourly basis for years could have devastating consequences for the human psyche. I doubt it can be healthy to permanently wander in the uncanny valley. Star Trek perhaps did not spend nearly enough time exploring the predictable problems of holodeck addiction and psychosis that AI chatbots are already giving us a glimpse of.
I found the discussion of AI's energy usage to be frustrating. Of course, a single query's energy demand may be negligible, but on the whole it is indisputable that centers are using up a larger and larger portion of our finite energy supply. This is causing energy costs to rise in places where data centers are putting more strain on the grid. Tech companies have no market incentive not to continue to use large amounts of energy, a public good, which is the definition of a negative market externality. The problem needs to be addressed. Mat would, it seems, have the public foot the bill to expand energy supply (and increase fossil fuel emissions) so that tech companies can build more data centers?
You raise a few separate issues. Consumer AI use is negligible as a driver of energy demand compared to training and enterprise stuff, and there is more legitimacy to the poor data center policy that has to do with states competing to attract business and poor infrastructural capacity to handle it, which is a government oversight. AI companies are diversifying their data center builds overseas due to an inability for the US to be competitive - as you said their incentive is to pursue cheap reliable energy, so poorly prepared states are offering incentives to keep that custom local before they can handle it.
The extent to which those costs are being socialized, which is a real problem, is a state failure they are now trying to address. So tldr the US needs to find ways to be competitive in this new economy, and are doing so chaotically. AI development is not going to stop in any case. There is also the sticky question of what often economically dormant areas in flatlands would do otherwise - there is a reason many of those areas are trying to court economic activity, and positive and negative examples.
The fossil fuel issue is where I think my Koomey recommendation is instructive as to the relative total energy burden of data centers and emissions, and often ridiculous projections of future demand. I think it is legitimate for local data centers and potential negative externalities of those builds to be scrutinized, but my feeling is we need more energy and better planning, as the trade off of losing that business to elsewhere will also affect people.
AI accelerationists seem to want to skip to a "Star Trek" utopian tech culture where everyone spends time programming their personal holodeck experiences, without the grueling political work and beyond moon-landing level advancement in political technology that will be required to make that reality benefit the 99%. Tech advances historically don't often lead to positive political advances - in fact quite the opposite, particularly in the short term. Luddism from a correct historical perspective should not be derided as anti-technology or analogue nostalgia. Rather, it is about proactively neutralizing the novel weapons of class warfare being used against labor. I rarely hear AI optimists acknowledge that the ethics of using AI are entirely contextual to the political framework.
I dont see how advancing and experimenting in the AI space is an ethical endeavor for artists and media creators unless they have a actionable strategy for using AI to directly effect positive political change. "Break some eggs to make an omelet" is mostly what I am hearing - ignoring the historical lessons of the industrial revolution, social media and other cataclysmic shifts in which untempered tech acceleration resulted in massive human strife, intractable political destabilizations, and dangerous consolidations of wealth and power. At the current course, I think it will be generations before generative AI use in the media and arts will have a net benefit to human wellness, and the faster we push the tech, the longer the timeline could be. Nevermind that like forever chemicals and social media, we have very little research into how widespread AI use might harm human development and psychology. Being exposed to generated falsehoods and artificial realities on an hourly basis for years could have devastating consequences for the human psyche. I doubt it can be healthy to permanently wander in the uncanny valley. Star Trek perhaps did not spend nearly enough time exploring the predictable problems of holodeck addiction and psychosis that AI chatbots are already giving us a glimpse of.
thanks for having me CUJO
always a pleasure
great talk 🐱 my link to join is adrienmeunier777@gmail.com