A Five-Point Plan to Regulate AI

The future of AI has been growing more uncertain, and with the Jensen Huang fiasco this week following such a strong data center backlash in the midterms, it is as uncertain as it has been, at least since the subject started to really take off. I will not go deep into the business of Nvidia in this post. It’s its own rabbit hole that I don’t want to make the topic of this piece. All I will say is that I hope to have the power to change the world someday, and if/when I do, I hope to prove that power does not have to corrupt one’s morals and values. Hold me accountable to that if I ever get there. But as the AI bubble comes to a new point of question, I wanted to share some of my thoughts on what I think makes sense with respect to AI policy. This was a question that was posed to me by The Change Campaign’s board at our last two meetings, so I have been working to articulate a position, and this feels like a good moment to share my work thus far.

In full transparency, I use AI. Not to write these posts, in part I started writing these posts as an exercise to generate my own thoughts without AI, but I use it because running a one employee organization trying to solve the problems of the world has been hugely aided by Claude. It is an incredible tool, especially as I have gotten better at working with it, so I don’t want to see AI just go away. However, there needs to be serious regulations on AI’s development. The lack thereof has been concerning for a long time, as I along with most people have been expressing for a long time now. The Nvidia news this week only escalates those concerns. 

I am not an AI expert by any means, so I do not have all the answers, but over the past several months I have been studying AI policy to try and develop common sense policies that can begin to regulate AI and address our concerns. I have developed an initial five point policy proposal:

1. AI data centers can only be built in communities in which a majority of the population agrees to the development of a data center in their area. 

2. AI data centers must use closed loop cooling systems, develop renewable energy infrastructure to power the data center, and modernize the grid it operates on.

3. In times of an energy crisis, AI data centers must go offline before any residential customers are forced to go without power.

4. AI developers are accountable for the development of reasonable safeguards to protect users from harms caused by AI and will be held liable when reasonable safeguards fail. 

5. Incidences of circular financing will be scrutinized for round-tripping and companies taking out debt for AI build out will be taxed on their borrowing. 

Based on what I have learned about AI, these policies make sense to me to protect some of the greatest risks that I believe exist with AI. This is the first time I am sharing them, so let me know what I am missing, what resonates, and these will continue to develop. I have not studied many AI policy proposals as much as I’ve studied concerns regarding AI, so there are likely popular policies that I am missing here, but these would provide protection for our society. I am a bit nervous to be honest sharing these, this is outside my normal policy arena, but sharing provides a good opportunity for feedback and to learn more about what I don’t yet know from there.

Before I get further into the reasoning behind the five points I focused on, I will note that there are sector specific concerns about AI that my proposal does not address. There are serious conversations about AI in schools, for example, that I have many thoughts on, but I am more focused, at least in this piece, on the general AI industry. There is also a major gap because I just don’t have enough information yet on it, but there is serious work to address the concerns that arise from the Nvidia news. This includes national security and oversight of at least the Commerce Department, and potentially other areas that were negligent in their actions.

The five points are in three general buckets: addressing AI infrastructure build out, addressing widespread harms caused by AI, and addressing the AI bubble that has created massive risks in the US economy. On the AI build out, the ways in which AI data centers were introduced to the American public was terrible. There was no care for public relations, it was a moment where very wealthy companies took the position that they had so much money that they could not be stopped. There was minimal if any effort to address the concerns or needs of the communities the data centers were being built in (or proposed to be built in). We saw town halls where constituents attended to unanimously voice their concerns and opposition to the construction of a data center, and local government officials deciding otherwise. I do not know what deals were made behind closed doors, but that is not what representative government looks like. 

There are serious concerns around AI data center build out: land use, water use, and energy use, which is what the first three points aim to address. There are potential benefits from an AI data center that a community may opt in for. For example, in Loudoun County, close to where I live in DC, data centers (note: most of these are not AI data centers) are expected to pay well over $1 billion in local taxes for the county. Additionally, data centers can create new jobs for residents. While a lot of the jobs are around construction and will not last forever, some jobs to maintain operations are likely to persist long into the future. But those wishing to construct data centers must get buy-in from the community they are going into. This is one of the issues that has the greatest consensus in America today and in recent history. People do not like AI data centers, so getting buy-in from the communities hyperscalers ultimately build in is critical, not just for the community’s sake, but it is also what is best to build trust in the AI industry.

Two ways, in addition to economic benefit, I believe AI data centers can get buy-in is by utilizing technology that does not drain/pollute a community’s water for cooling purposes, and to improve the energy efficiency and grid of the community they are building in. Water scarcity is becoming a bigger issue as the climate crisis and overpopulation issues continue. We should not need to argue that fresh water should first and foremost be used to sustain and nurture living creatures. If anyone wishes to argue that AI is sentient, which is a whole other conversation, then let me clarify by saying that it should meet the needs of natural intelligence first. Fortunately, there is infrastructure capacity where both of these are possible. Closed water cooling systems do not constantly drain nor pollute water supplies. We should take advantage of that innovation.

On the energy concerns around AI data centers, it should be a no-brainer that residents should not pay for an increased electricity bill because the demand from a data center, especially one where the residents had no say, is driving up the cost of energy. The largest consumers of energy on a grid should pay the most for the grid’s energy. Everyday Americans footing the bill for mega-billionaires at the forefront of these buildouts is such a gross indecency. Those who are building these data centers have the money to not just pay for their own energy usage, but should improve the energy source in the communities they are entering. This will build good will and help make necessary improvements to the US energy grid. I was in China not long ago and saw their incredible energy infrastructure. The long-term winner of AI will not be based on who has the best model, but will depend on who has the best energy infrastructure to carry out AI. It is in the best interest of these companies to improve the energy infrastructure that they rely on to fuel the continued growth of this industry. And, on a similar note, to rely on a non-renewable resource for one of the most energy-demanding industries is so short sighted that I do not have words to express my disappointment in the lack of coherent strategy. In addition to improving the grid in the places where these data centers are built, these companies should also build renewable energy sources to improve their own sustainability as well as for the community.

Beyond just the data center build out, we need regulation protecting against the harms AI is causing. A concrete example is the chatbots who assist people in carrying out violent actions against themselves or others, as we have seen examples of real and alleged suicides and mass shootings that were assisted by AI. AI companies saying that their models cannot be held responsible for this are wrong. There must be safeguards in place to reasonably protect people, especially young people, using these models, and accountability in place when reasonable safeguards fail. The descriptions above could be negligence and/or wrongful death. People are held to that standard, and those building these models should be held to that standard as well. If we cannot protect against reasonable harm, then these models are not ready to go to market. 

Finally, the irresponsible financing that has fueled the AI bubble must be regulated. The amount of leverage these companies have placed on the AI build out threatens to lead to a recession. It has not all been strategic investment, but rapid growth for no greater strategy than to beat the other players, which has created a precarious situation. The money required to build not just data centers but also train the models is so large that if there is not continued exponential growth, which we are already seeing slow, then there is a real likelihood that these companies will never truly be profitable. While given AI’s utility it is unlikely that the entire industry will be unprofitable forever, there is so much debt on these companies that it is very likely that many of these companies will eventually go bankrupt. We must protect against how this impacts everyday Americans and our society at large. While circular financing itself is not illegal, if there is illegal activity to inflate valuations through circular financing, that must be prosecuted and stopped. Because of the tax advantages of leveraging debt, to moderate such an over-levered market, taxing debt beyond a certain amount would presumably help reduce the amount of stress the AI build out is putting on private markets. This strategy has potential to grow into a reasonable way to tax extreme wealth that has created loopholes to get out of paying its fair share back into society, but I see a beginning with the way debt is inflating the AI bubble. 

These five points are not all that needs to be done, but I think they address some of the big needs that have led to concerns and threats with the current way that AI has built out. I am still at the beginning of trying to learn about this, so curious to know what you agree with, what you simply disagree with, and feel I am missing or should add. This is a different kind of post for me, a bit further out there, much less within the realm of what I know best, but it’s important to put out proposals and learn from reactions. So let me know what you think, and thanks, as always, for reading!