Quick Hits
It’s been a quiet week. Only a few, smallish items were not covered in-depth elsewhere. Do read about the xz saga, a fascinating example of both the pitfalls and resilience of the open-source software ecosystem.
The Apple Vision Pro has added “spatial personas,” allowing its 3D models of the wearer to escape the Zoom-style box they previously were confined to in video chats. Now, two Vision Pro users can interact with each other much more seamlessly. It’s easier to see than it is to read about. If you squint, you can see an intriguing vision for the future.
Multiple open-source versions of Devin, the autonomous software engineer created by Cognition AI, are now competitive with the original on benchmarks. Devin, and its open-source variants, are primarily based on “scaffolding,” software and prompt engineering placed around foundation models like GPT-4. While few actors in the open source community have the compute to train a frontier foundation model, they can be fast followers with this sort of scaffolding, which requires ingenuity, rather than compute, to build. This also means that better foundation models can be plugged directly into this scaffolding with minimal work, meaning that sudden jumps in the capability of agents are both possible and probable.
Main Article
One of the sentiments I hear expressed most commonly, from posts on X to the halls of Congress, is “where are we headed with technology?” or “what are we accelerating towards?” As a recently Politico piece wonder aloud, “Can anyone control AI?” A closely related sentiment I hear often is “AGI may fundamentally transform the world, but no one asked us about whether we want that transformation to occur.”
Some believe that a direction of technological development can be imposed from the top down. Lina Khan, Chair of the Federal Trade Commission, remarked in a recent interview with Jon Stewart:
There’s no inevitable outcome here. We are the decisionmakers, and so we need to use the policy tools and levers that we have to make sure that these technologies are proceeding on a trajectory that benefit[s] Americans and we’re not subjected to all the risks and harms.
When electricity began to make its way into the home, the walls of many houses were dark red or green. In part, this was a design choice made to cover up the soot and smoke stains caused by the kerosene lamps that were most Americans’ source of interior light prior to electricity. Electricity obviated such concerns, meaning that walls could be lighter colors. White paint in interior design is a consequence of electricity. Who could have foreseen that?
Perhaps more to the point, electrification of industry enabled large factories to place electrified equipment anywhere, rather than having to place the most power hungry equipment as close as possible to the steam engines that powered factories before electricity. This, combined with better interior lighting, enabled the assembly line. This meant that manufacturing jobs became less skill-intensive, profoundly changing the dynamics of the labor market for decades. Who could have foreseen that? Paging Dr. Khan!
A few months ago, I was a panelist at a political theory conference in Budapest. One of my co-panelists, Ken McIntyre, presented a paper on voluntary and involuntary association. Conservatives, as a general rule, tend to like voluntary association: Lion’s Clubs, churches, etc. Ken’s point, though, was that there is an awful lot of involuntary association in the human experience. Children have involuntary associations with their families and their schools, for instance whether they like them or not. At the deepest level, all of us are in a kind of involuntary mutual association with reality: none of us asked to be born.
I often return to Ken’s talk when I think about how to respond to questions like “what are we accelerating towards?” or “how come no one asked whether we want the transformations AGI will bring?” or “what technological trajectory do we [policymakers] want to decide to take?” If humanity were a singular mind, these kinds of questions would be legible. But of course, the world is not governed by a singular mind, and for that reason “the world” is not governed at all in any meaningful sense. There is no “we” who “decides” questions such as these, and there is no polling place to cast one’s vote. History does not unfold by show of hands.
Many of us realize this, yet still fall into the trap of speaking as though the world has a helmsman. To some extent this is because of the English language itself, which mandates a subject, and hence a decisive actor. It is easy to fall into cause-and-effect traps unless one is cautious, and such caution in colloquial settings often veers into tediousness (did COVID “cause” lockdowns? What, exactly, did cause the lockdowns?).
Despite the tricks our language may play on our minds, we should not shy from the reality: humanity has no helmsman and no destination to which we are headed. Some see this as an immutable fact of the world; others see it as a tragedy to be corrected, often with urgency. As Michael Oakeshott wrote:
Surveying the scene, some people are provoked by the absence of order and coherence which appears to them to be its dominant feature; its wastefulness, its frustration, its dissipation of human energy, its lack not merely of a premeditated destination but even of any discernable direction of movement. It provides an excitement similar to that of a stock-car race; but it has none of the satisfaction of a well-conducted business enterprise. Such people are apt to exaggerate the current disorder; the absence of a plan is so conspicuous that the small adjustments, and even the more massive arrangements, which restrain the chaos seem to them nugatory; they have no feeling for the warmth of untidiness but only for its inconvenience. But what is significant is not the limitations of their powers of observation, but the turn of their thoughts. They feel that there ought to be something that ought to be done to convert this so-called chaos into order, for this is no way for rational human beings to be spending their lives. Like Apollo when he saw Daphne with her hair hung carelessly about her neck, they sigh and say to themselves: “What if it were properly arranged.” Moreover, they tell us that they have seen in a dream the glorious, collisionless manner of living proper to all mankind, and this dream they understand as their warrant for seeking to remove the diversities and occasions of conflict which distinguish our current manner of living. Of course, their dreams are not all exactly alike; but they have this in a common: each is a vision of a condition of human circumstance from which the occasion of conflict has been removed, a vision of human activity coordinated and set going in a single direction and of every resource being used to the full. And such people appropriately understand the office of government to be the imposition upon its subjects of the condition of human circumstances of their dream. To govern is to turn a private dream into a public and compulsory manner of living. Thus, politics becomes an encounter of dreams and the activity in which government is held to this understanding of its office and provided with the appropriate instruments.
Of course, the first problem is that everyone’s private dreams differ from one another. The second problem is that private dreams of a world order tend not to work out well when they’re put into practice. This is where the classical liberal notion of the state as an umpire emerges: because of the fundamental irreconcilability of our respective dreams and ambitions, no single agenda should be allowed to dominate. Minority interests must be protected, and ideas should remain in competition in a robust marketplace. The role of the state, in the classical liberal vision, is to enforce basic rules on the process of competition in a predictable manner.
AGI has existed in the minds of futurists, technologists, and scientists for decades as the canonical transformative innovation—the “last invention,” as some have said, or at least the last purely manmade invention. Thus it is also the source of many private dreams. Because AGI is developing alongside, and fundamentally intertwined with, advancements in fields such as materials science, space exploration, and bioengineering, the possibilities for the future are vast. If DNA and RNA become as manipulable as Python and JavaScript, we are in for strange times indeed—quite a bit of “untidiness,” as Oakeshott might say.
I believe that, no matter how intelligent AI becomes, one-world governance will remain neither feasible nor desirable. Indeed, as human activity becomes more diverse, and as our ambitions rise along with our capabilities, I believe the value of the state as a neutral umpire will increase, not decrease. Whether our political leaders (or indeed, the electorate) agree with me is another matter altogether.
Regardless, as the world becomes progressively more complex, the visions of any one person or group of people will come to seem ever more jejune and myopic. The United States Department of Justice, for example, has spent millions of taxpayer dollars and years of scarce staff time on a lawsuit against Apple to ensure that alternative near field communication-based payment infrastructures are permitted on the iPhone, among a few other equally trivial things. This is apparently an important part of what the Department of Justice, over multiple presidential administrations, believes is necessary for a dynamic yet balanced economy in the 21st century.
As machines become more capable, one of the increasingly uncomfortable questions that will be asked of every individual and institution is: “what value, precisely, do you bring to the table?” If “deciding our technological destiny” is off the table, what value is it that an ideal government would provide to help manage some of the chaos that will surely come our way soon?
One can think of many things. What if I could cryptographically validate my identity, or any major aspect of my identity, without revealing anything except for precisely the information I need to convey? It could help ensure that all the world’s open source software is regularly screened for bugs, with fixes submitted to the project developer. It could use technology to create new ways to pilot legislation, testing it in the real world to gauge its effects. It could expand that infrastructure to state and local governments, giving birth to a new, rigorously evidence-based school of policymaking. It could build robust technical standards and interoperability protocols for generalist AI agents. It could leverage America’s unparalleled technological innovation and thick capital markets to expand the capabilities of its military, rather than designing over-engineered contraptions from the top-down and hoping that a private sector firm will be able to build it on anything like the originally envisioned budget and timetable. It could do all of these things at once, and more.
What is the commonality behind these ideas? They expand capabilities, either of all Americans, of the government itself, or both. They are additions to our epistemic and technological toolkits. They create value, rather than redistribute or diminish it.
It does not create value to pass a law which says “it is illegal for your technology to do bad things, only good things are legal, and if you do bad things we will punish you in various ways.” That was already obvious to everyone, and yet by making the statement in such simple terms, it only serves to confuse (Who, exactly, is responsible for harms? How do we tell? The laws thus far are silent on such matters.) They create chaos in a period that is already chaotic.
The broader point, though, is that it is possible to imagine a productive role for a robust and energetic government as society grapples with AI. It’s just that this role is quite different from the role policymakers are accustomed to playing.
Happily, some of the ideas I have described are in the works. We need much more work of this kind, because if it is not led by the government, someone else will do it. And the legitimacy crisis our government currently faces will only worsen.
I suspect the coming decade (or more) will be among the most dynamic in the living memories of most people alive today. Dynamism is another way of saying “fat trimming.” Do our policymakers wish to be part of the fat?
We do not know where all this dynamism will take us. “We” do not decide such things. That does not mean one should give up hope for making a positive change, however. Because there is no destination and no way to set one, individual efforts can meaningfully impact our direction. In this sense, “you” are more powerful than the “we” we all imagine, because “we” do not exist as a decision-making body in any coherent sense, while “you” do. You can build tools, develop ideas, and persuade others to do the same. You can create value, and thanks to technology, your ability to do so is increasing by the month. Each of us can do that hard work, and together, “we” can bring about a better future.