A not-yet-named epistemic story
Need stuff on page…
Government hires the wrong group…
The group has a plan…
“Um… boss?”
“Yes, what is it?”
“Strange thing happened. So, the results of the grant to an “Accuracy and Validity Study” have started to come in. It was apparently part of Executive Order 14002; the weird one amending Reagan’s cost-effectiveness order.”
“Ah, one of those orders, I haven’t had time to go through them all.”
“Well, in the details it grants $20 Million to some small independent think-tank. They didn’t trust the regular officials for this sort of work. They requested it be given to an outside party.”
“So the government just wasted $20 Million?”
“Probably. The interesting thing is that there seems to have been some sort of mistake. The think-tank was going to be chosen by an in-house committee, but things got hectic and it seems like they rushed the process and made an error. I’m pretty sure that when the order was written it was intended that the contract go to the Bipartisan Research Center or the US Decision Institute.”
“Those are both fairly conservative groups, right?”
“Yes. But the contract instead went to some group none of us have heard of called the US Decision Center.”
“So the government just wasted $20 Million.”
“We have a meeting with them next week to discuss preliminary findings.”
A few days later…
“Thanks for inviting me here. As you might imagine, our team has been quite busy since we got the contract.”
“Yea, sure, how are things going with that?”
“Well, to be fully honest, we started with 2 of us, so we had to grow a bit to get through the contract.”
“And this was… something about accuracy in the government? You did investigation to improve our cost effectiveness estimates or something.”
“Well, the task we were given, more specifically, was to optimize the accuracy of estimates placed by government bodies.”
“Sure, so do you have a few changes to recommend?”
“Well, yes, but it’s a bit complicated. I suggest taking a seat.”
(Seat taken)
“I should probably make you aware that UDC is rather novel in that all of our work is backed by a Legal Trust Guarantee. Our colleagues at the Center for Experimental Epistemics and the New Incentives Institute deeply review our work on a stochastic basis, and if they determine that any of it is not correct or has been considerably suboptimal, then all of our funds will be rerouted to an third party entity aimed only at suing us and ruining our personal lives. It’s an agreement we’ve made a while back, it helps make sure that our clients can trust us.”
“Please just get to the point.”
“So the rules of our US contract specify that we must detail a standard whereby Government claims must be verified in order to be used to justify laws. It’s rather a natural extension of cost-effectiveness requirements. We’ve been experimenting with what standards would look like. The leading candidate is that we take binary claims and evaluate them based on their expected probability of being true. In order for a claim to be considered worthwhile for official approval, it needs to to be considered 85% likely, or around 2.5 bits of accuracy.”
“Ok, so we just remove a few claims here and there?”
“Well… we’ve been reviewing official records, and have found that only 35% of official empirical estimates make this mark.”
“So we remove a bunch of claims?”
“Well, those are the empirical statements. Statements like, in 2040 GDP growth will change by over 3% per year. There are also normative and metaphysical statements. Here’s a simple example. I take it you are familiar with the phrase, ”We hold these truths to be self-evident, that all men are created equal, that they are endowed by their Creator with certain unalienable Rights, that among these are Life, Liberty and the pursuit of Happiness.”
“Yes, of course”
“Well, believing truths to be self-evident requires very precise philosophical positions that are difficult to find conclusive evidence for, and finding accepted evidence for these claims otherwise appears to be quite the challenge.”
“You’re questioning The Declaration of Independence?”
“A fair amount of legal doctrine is justified under that document, so we’d really want it to hold up for proceeding justification.”
“This sounds… insane. But at the end of the day, I’m sure we can just remove a bunch of claims and justify things in other ways. Maybe we just need to formally not make 90% of the sorts of claims we are used to?”
“Well, for empirical claims, there’s an average of 1.5 bits of information. For normative and metaphysical claims we’ve measured an average of 3 zeros.”
“Like, 1000 bits? That’s good, right?”
“No, like 0.0001 bits of information. It’s quite low madam.”
“So we just remove these sorts of claims?”
“Well, that’s the problem. Most of the laws don’t quite work without them. It’s easy to say that a Sales Tax modification will increase revenue by a fixed percentage. But it requires a belief in a normative statement in order to claim that increasing revenue is a good thing.”
“But obviously increasing revenue is clearly a good thing.”
“Can you produce a vast set of clear and definitive evidence proving so?”
“Well, I mean intuitively it’s obvious.”
“It might seem intuitive to you, but we can’t accept that as sufficient evidence. According to the calculators this would be… maybe 3 microbits, plus or minus 4 microbits. You were raised in a culture that heavily encouraged said intuitions. The more local culture advocates for something, the less credence we can apply. We’ve also talked to teams of Anthropologists, and apparently intuitions between cultures around this topic are highly sporadic. We’ve initiated plans to pay small villages to adopt randomly generated cultures in order to measure the intuitions of children raised under said cultures, but that will take a while.”
“Our country was founded on philosophical principles, you can’t question the basic propositions of this country.”
“Contractually speaking, those foundations are some of the first things we need to question. The bits simply don’t add up in our calculations. We formally agreed to have standards for correctness, and we, simply put, don’t have sufficient evidence when using reasonable standards. If you wanted ethical or metaphysical questions to be accepted at far lower epistemic standards, there should have been a few clauses in the contract specifying those details.”
“Okay, so… so… just what does this mean going forward? What impact will this have for our legislation?”
“We weren’t really tasked to answer that question, but we’ve been analyzing the last 500 pieces of legislation and on basic review, zero of them would have met basic accuracy criteria. If dramatic measures were taken to reduce the number of claims in them and make them more modest, then, I believe… zero would have met the basic accuracy criteria.”
“But all of this is just recommendation, I assume? This… it can’t be binding… How do we get around this? Maybe we could turn your ideas into recommendations, and have a different committee figure out the next steps.”
“Well, I’m afraid that the rules of Order 14002 specifically state that the standards generated from Order 14002 must be used if ever removing or modifying the rules of Order 14002. So you could try to overturn it, but you’ll need a mightily impressive defense of your reasoning. According to our estimates… there is an exceedingly low chance of your administration succeeding.”
“I’m confused. Please re-explain all of that, tell it to me like I’m 10.”
“Okay, I will summarize. For years, the United States Government has been feigning dramatic overconfidence to ram bills past the legislature and the American public. Not just bills, really, but all actions in general. Basically all of it is based on a series of wildly speculative accusations, not that much different from arguing based on beliefs in the daily emotional states of the god Hephaestus. In their defense, most other groups were much worse.
A few years ago, the previous administration signed the order to prevent Congress from passing laws without relatively high standards for justification. What they didn’t realize is that Congress was previously using incredibly low standards of justification. Not just Congress but the entire American and Global system; on inspection it’s all quite terrible. I assume that the administration expected the gatekeepers to be some political pushover, but, for some reason, they chose us.
We, a small think-tank made up of two Ancient Cynicism scholars, have extensive legal obligation to report the absolute truth as we see it. The truth as we see it, will feed up to the to-be-formed Council of Accuracy and Validity, starting next year. At that point, unless things change around a whole lot, and we really mean a whole lot, we expect that the amount of new legislation that is possible to be passed is exactly zero.”
“So you’re saying we’re fucked.”
“On what kind of epistemic standard would you like me to answer that question with?”
“So we can’t pass anything? Our government will stop making new bills?”
“Probably, for 5 years.”
“And what happens in 5 years?”
“Well, that’s when clause A43 initiates. Clause A43 states that after a 5 years leeway period, previously passed laws will be considered for similar standards, and a formal process will be undergone to undo bills that don’t pass it. Executive Order 14002 is excepted of course.”
“I’m going to need to talk this over with a few other people at the White House.”
“Understood.”
“Is there anything else?”
“Well, I did want to run some… proposals past you. I think we may be able to find some solutions, but they will be expensive and take a while.”
“Let’s set up another meeting then.”
A few days later
“Thanks for agreeing to this meeting, let’s get started.”
“So, following up with our last meeting, the situation now is a bit of a mess. Starting in a few months, new legislation will need to pass some fairly strict standards of evidence to be passed. This is a challenge because the previous and current standards are, in comparison, barely existent. This will impact new legislation soon, and almost all existing legislation in 5 years.”
“Yes, yes… so, what’s the ask? Are you requesting another $20 Million?”
“I’m afraid you may not appreciate the scale we are dealing with.
Let me… try to explain. Are you familiar with the new field of intellectual information economics?”
“Intellectual information economics?”
“It’s quite new. Intellectual information economics is really a natural extension of Information Economics, but applied to economic decisions around making progress on issues surrounding intellectual development.”
“That sounds ridiculous, intellectuals are all over the place, you can’t just make it into a science.”
“If you sacrifice some precision you can come up with estimations that are predictive enough to be useful. It’s not very certain, but it’s good enough for high-level decision making.
“I’m skeptical.”
“I’ll explain how it works. The first challenge is to measure the scope of the task at hand. We’ve made a preliminary map of a sample of the space of premises necessary to verify the current legislation and used it to estimate the total.”
(Pulls up network diagram of such premises.)
“As you can see, we’ve ordered things into claims that are relatively easy to be verified on the right, and this cluster of metaphysics on the bottom left. The colors of the dots indicate the magnitude of the certainty we have on them, and the size indicates the relative importance in the network.”
“There’s a lot of white dots here.”
“Yes, there’s a lot of work to do. We can use this network to estimate the Expected Total Uncorrelated Information Potential. The total seems to be something like 18 gigabits plus or minus 1.4 orders of magnitude. Of course, this isn’t actually possible to obtain with current limitations of time and space. There are many questions we doubt we can ever know the answers to, in the next few billion years. But we think that if we can make it to just… 150 megabits, we have a shot at hitting a threshold necessary to pass potential, well-written laws.
Now, in order to get to 150 megabits of designated marginal certainty, we would need to expend a considerable number of effort. What we need to do here is to make considerable advances in philosophy and world knowledge. ”
“So we hire the best Academics? We’re the United States, we have access to the very best.”
“It’s not going to be enough. we’ve been evaluating current government efforts and possible government efforts. If we hypothetically put together the total efforts of the CIA, NSA, Intelligence committees, research committees, DARPA, IARPA, RAND foundation, and all of our other think tanks, for a total of 10 years, we think we could get to.. ” (checks papers)… “120 kilobits, plus or minus 2.3 orders of magnitude. If we expanded capabilities to include absolutely all of Silicon Valley and the top Intellectuals from Europe and South America, we could get to around 215 kilobits.”
“I thought you said you have a solution.”
“We may. You see, the above calculations assume that fundamental theory, tools, culture, coordination, and research techniques stay constant. On inspection it’s possible these could get improve dramatically. It’s not a sure bet by any means, but at least there’s some way for the math to work out. It’s a possibility.”
“Go on…”
“To be honest, when we began inspecting things we were expecting a more efficient market, so to speak. It’s not exactly like things in this area are efficiently run right now. Contemporary scientific practices don’t seem at all optimized for useful scientific progress. Philosophy departments generally don’t seem to be trying to solve the sorts of questions we need solved. Intelligence divisions use still methods from the 70s, it’s like they haven’t noticed Tetlock. We can also change hiring policies for Government roles, currently we are losing many of the best people to industry and abroad.
So we can fix the obvious things, but that only gets us so far. Maybe 2 or 4 orders of magnitude gain in research efficiency.
Research gets diminishing returns, so inputs scale steeply sublinearly with outputs. In order to make it 18 gigabits, we’re going to need at least 4 to 10 orders of magnitude more productivity per unit resource. So right now we have to find 2 to 7 orders or magnitude somewhere.
This gets tricky but we have ideas. The silver bullet, if there is one, is in Artificial Intelligence. Sadly it’s not really one single silver bullet, but rather more like some alien life form that feeds off a complex infrastructure. Assuming advances in ML happen, and that’s a big assumption, we figure we can give us a few orders of magnitude. But we need to know how to use it.
”
This orange dot refers to the simulation hypothesis. The color refers to ”
We did a test run on 5 randomly sampled narrow questions in
“