Hundreds of bills about AI have been introduced to Congress. Almost all die in committee, and usually they only address one aspect of how AI could impact civilization: data centers, children's wellbeing, transparency, etc.
From my experience watching how the Epstein Files topic played out (more below), I think it may be prudent to assume that we will only have one meaningful shot at getting something substantive and well-thought-out about AI passed in the short-term. Public attention and political will are fickle things. Even if they endure to a certain level of strength and persistence (as with the Epstein Files topic), it seems that getting subsequent legislation passed on a subject in which there is strong opposition can still be a herculean effort. For AI, I do not think we should waste the opportunity while public attention and political will are mounting.
I've attempted to draft legislation that intends to address the full-spectrum of AI-related challenges we'll face: near-term and long-term, domestic and international, mundane and existential, immediate and ongoing. The structure is to legislate into existence a slate of interim technical working groups (which turn into permanent government entities outside of Congress) mandated to produce time-bound analysis/recommendations across 19 AI domains, and then to force Congress to actually act on that input through procedures it already uses called Hammer provisions.
After forcing Congress to act across these 19 domains, the technical working groups turn into a permanent AI Council and international diplomatic body for ongoing work: maintaining evidentiary records, oversight, and preparing subsequent legislation to be ready as needed. I'm confident the draft is incomplete, but I believe the overall strategy is sound. I'm curious whether others are pursuing something similar, and whether it makes sense to combine efforts and ensure it is a complete, durable approach. I was precise in enumerating both the roles and expertise that are mandatory to serve in the technical working groups, as well as mandating the level of quality of their inquiry by identifying the specific topics which would need to be contemplated. Otherwise, I would imagine we end up with suboptimal performance. AI risks cut many ways and we'd be damned if the centralization of AI capabilities results in an authoritarian state.
Right now, it seems like the biggest public rallying cry is around a pause. And I've heard that if legislation is successfully passed to induce a pause, then the next thing that should be asked for is more time. Currently, I think this may be a mistake. I think that a pause will likewise induce a waning of the political will and public attention needed to get things done. We should strike while nervous systems are hot. When people are primed to demand substantive action, we should have a complete package ready, if possible, about the most optimal course. In my mind, that means creating the infrastructure so that we can continuously update such a course.
This likely means removing the obligation of figuring out what to do about AI from the shoulders of members of Congress who are now interfacing with different AI groups, authors, activists, and academics, have hardly any technical expertise themselves, and are therefore outsourcing to leg staff or trusted contacts as a heuristic for good sense-making; essentially relying on the heuristic of character and credentials, instead of being able to evaluate content themselves - if they are even operating in that good faith in an election year. I am not saying a pause should not happen. I am not saying I have the answers about what to do. What I am saying is that we could be prepared with the set of questions that would need to be answered, and legislate the process for how to answer those questions in the interim, while being able to update the answers as we go along without relying on Congress internally, and forcing Congress to act on those answers.
If folks come together and figure out how to meaningfully and strategically combine legislative text, an AI governance bill could have all of the expertise, resources, direction, and power it needs to self-update and run on auto-pilot for the time being. Like most people here probably already believe, I don't think we should waste energy on a static solution, even if that solution is to buy more time to develop something more flexible later. My approach is Title II of The MAD Act (my omnibus bill aimed at targeting a confluence of issues related to power asymmetry in this country), which is nicknamed "Demand A Plan for AI." Some of you may have glanced at it, but because it is over 250 pages of everything in the kitchen sink, I think it may be worth summarizing abstractly, in case folks wanted to copy parts of the high-level strategy and draft something fresh, rather than amend the existing bill text, if folks think this is a valid strategy at all. Another summary is available here (may be slightly out of date).
I am already meeting with members of Congress across the country about this title, so even if the strategy itself isn't seen as worthwhile, I'd like to update what I'm advocating for immediately.
So What Happened With The Epstein Files and Why Is It a Good Case Study?
The Epstein Files topic was a heated political maelstrom for a short while. Uniquely bi-partisan at some points as well: MAGA was promised justice and the Left thought this was a perfect opportunity to expose the President. Transnational political figures operating with impunity trafficking children is one of those things that can outrage the public and scare Congress into acting. And Congress did act. Unfortunately, The Epstein Files Transparency Act (EFTA) was riddled with loopholes that were predictably exploited and gaps which largely remain unaddressed (see my talk on the three gaps of the Act as I spoke about them in D.C.). Congress recently attempted to close some of these loopholes with the EFTA II. Massie and Ro are attempting another discharge petition, and as of this writing, the petition is one signature short of the 218 needed, but Johnson's decision to cancel two of the House's three remaining September workweeks narrows the window to get the signatures before recess. I think Congress may be pressing on because when polled, the public still cares about this issue, but the EFTA II still has major gaps. Immense political energy is being spent on an incomplete solution to close some loopholes, but still misses gaping issues - like asking other agencies besides the DOJ for records. In case folks don't know - I have a background in scoping government files from my work at the Internet Archive, so this is an informed opinion. From this work, Brewster Kahle (founder of the Internet Archive) has endorsed my candidacy, as have 3 Epstein Survivors. The EFTA could have been more thought-through the first time, and instead has obligated Congress to continuously push for impartial updates to address the issue. I am not sure we should assume the same strategy is best for AI.
So What Does the MAD Act Cover?
While the mechanism for how the MAD Act (Title II) works I hope is sufficiently summarized on this page (though some things may be out of date), I think it's useful to list the 19 domains for a cursory look at whether this is sufficiently comprehensive in its approach. Again - it is meant to cover topics more politically salient (even if pedestrian) as well as wicked technical topics meant to be addressed from multiple stakeholder positions:
IP, training data & creator compensation
Electoral integrity & AI threats to democratic processes
Domestic AI-generated electoral disinformation & synthetic media
AI security, critical infrastructure & incident response
Agentic AI systems
Autonomous weapons, IHL & arms control
If anyone knows of legislation which they think is better to introduce at this time, thinks any of these arguments are invalid, or on the contrary - think this is a valid strategy and want to coordinate to update the bill and advocate for it together, please reach out: Rep@JamieJoyce.com
Hundreds of bills about AI have been introduced to Congress. Almost all die in committee, and usually they only address one aspect of how AI could impact civilization: data centers, children's wellbeing, transparency, etc.
From my experience watching how the Epstein Files topic played out (more below), I think it may be prudent to assume that we will only have one meaningful shot at getting something substantive and well-thought-out about AI passed in the short-term. Public attention and political will are fickle things. Even if they endure to a certain level of strength and persistence (as with the Epstein Files topic), it seems that getting subsequent legislation passed on a subject in which there is strong opposition can still be a herculean effort. For AI, I do not think we should waste the opportunity while public attention and political will are mounting.
I've attempted to draft legislation that intends to address the full-spectrum of AI-related challenges we'll face: near-term and long-term, domestic and international, mundane and existential, immediate and ongoing. The structure is to legislate into existence a slate of interim technical working groups (which turn into permanent government entities outside of Congress) mandated to produce time-bound analysis/recommendations across 19 AI domains, and then to force Congress to actually act on that input through procedures it already uses called Hammer provisions.
After forcing Congress to act across these 19 domains, the technical working groups turn into a permanent AI Council and international diplomatic body for ongoing work: maintaining evidentiary records, oversight, and preparing subsequent legislation to be ready as needed. I'm confident the draft is incomplete, but I believe the overall strategy is sound. I'm curious whether others are pursuing something similar, and whether it makes sense to combine efforts and ensure it is a complete, durable approach. I was precise in enumerating both the roles and expertise that are mandatory to serve in the technical working groups, as well as mandating the level of quality of their inquiry by identifying the specific topics which would need to be contemplated. Otherwise, I would imagine we end up with suboptimal performance. AI risks cut many ways and we'd be damned if the centralization of AI capabilities results in an authoritarian state.
Right now, it seems like the biggest public rallying cry is around a pause. And I've heard that if legislation is successfully passed to induce a pause, then the next thing that should be asked for is more time. Currently, I think this may be a mistake. I think that a pause will likewise induce a waning of the political will and public attention needed to get things done. We should strike while nervous systems are hot. When people are primed to demand substantive action, we should have a complete package ready, if possible, about the most optimal course. In my mind, that means creating the infrastructure so that we can continuously update such a course.
This likely means removing the obligation of figuring out what to do about AI from the shoulders of members of Congress who are now interfacing with different AI groups, authors, activists, and academics, have hardly any technical expertise themselves, and are therefore outsourcing to leg staff or trusted contacts as a heuristic for good sense-making; essentially relying on the heuristic of character and credentials, instead of being able to evaluate content themselves - if they are even operating in that good faith in an election year. I am not saying a pause should not happen. I am not saying I have the answers about what to do. What I am saying is that we could be prepared with the set of questions that would need to be answered, and legislate the process for how to answer those questions in the interim, while being able to update the answers as we go along without relying on Congress internally, and forcing Congress to act on those answers.
If folks come together and figure out how to meaningfully and strategically combine legislative text, an AI governance bill could have all of the expertise, resources, direction, and power it needs to self-update and run on auto-pilot for the time being. Like most people here probably already believe, I don't think we should waste energy on a static solution, even if that solution is to buy more time to develop something more flexible later. My approach is Title II of The MAD Act (my omnibus bill aimed at targeting a confluence of issues related to power asymmetry in this country), which is nicknamed "Demand A Plan for AI." Some of you may have glanced at it, but because it is over 250 pages of everything in the kitchen sink, I think it may be worth summarizing abstractly, in case folks wanted to copy parts of the high-level strategy and draft something fresh, rather than amend the existing bill text, if folks think this is a valid strategy at all. Another summary is available here (may be slightly out of date).
I am already meeting with members of Congress across the country about this title, so even if the strategy itself isn't seen as worthwhile, I'd like to update what I'm advocating for immediately.
So What Happened With The Epstein Files and Why Is It a Good Case Study?
The Epstein Files topic was a heated political maelstrom for a short while. Uniquely bi-partisan at some points as well: MAGA was promised justice and the Left thought this was a perfect opportunity to expose the President. Transnational political figures operating with impunity trafficking children is one of those things that can outrage the public and scare Congress into acting. And Congress did act. Unfortunately, The Epstein Files Transparency Act (EFTA) was riddled with loopholes that were predictably exploited and gaps which largely remain unaddressed (see my talk on the three gaps of the Act as I spoke about them in D.C.). Congress recently attempted to close some of these loopholes with the EFTA II. Massie and Ro are attempting another discharge petition, and as of this writing, the petition is one signature short of the 218 needed, but Johnson's decision to cancel two of the House's three remaining September workweeks narrows the window to get the signatures before recess. I think Congress may be pressing on because when polled, the public still cares about this issue, but the EFTA II still has major gaps. Immense political energy is being spent on an incomplete solution to close some loopholes, but still misses gaping issues - like asking other agencies besides the DOJ for records. In case folks don't know - I have a background in scoping government files from my work at the Internet Archive, so this is an informed opinion. From this work, Brewster Kahle (founder of the Internet Archive) has endorsed my candidacy, as have 3 Epstein Survivors. The EFTA could have been more thought-through the first time, and instead has obligated Congress to continuously push for impartial updates to address the issue. I am not sure we should assume the same strategy is best for AI.
So What Does the MAD Act Cover?
While the mechanism for how the MAD Act (Title II) works I hope is sufficiently summarized on this page (though some things may be out of date), I think it's useful to list the 19 domains for a cursory look at whether this is sufficiently comprehensive in its approach. Again - it is meant to cover topics more politically salient (even if pedestrian) as well as wicked technical topics meant to be addressed from multiple stakeholder positions:
If anyone knows of legislation which they think is better to introduce at this time, thinks any of these arguments are invalid, or on the contrary - think this is a valid strategy and want to coordinate to update the bill and advocate for it together, please reach out: Rep@JamieJoyce.com
Thanks for your time.