Can Washington Be Automated?
politico.com
politico.com
But I've come to understand these industries are NOT built on rational, dispassionate decision making. The reason they only have a foothold in automation is because at their core, they are deeply relational industries. Even for incredibly sophisticated, "rational" individuals.
People simply want a person to talk to and walk them through things - even when they have all the facts generated from sophisticated algorithms giving them an unequivocal "correct" answer.
It seems like lobbying is the same. There is an important role for automation. But I believe it will simply augment the existing offering rather than completely supplant it.
Politics is the most relational of professions. Rational policymaking is an idealists dream - and I say this as a metric-driven, lover of numbers who once believed in that dream.
We need to think of automation as more of a cybernetics concept than just a program on a computer. In the front line you will have humans doing the Sales and that probably won't go away. In the back you will have computers which make decisions and in the middle you will have engineer(s) working on the interface. Once we can replace the sales experience thats when we will get there.
Actually, I just looked, and Wealthfront and Vanguard invest in the exact same funds (Vanguard's VTI and foreign equivalents), but Wealthfront's expense ratio was around 0.12% vs. 0.05% for Vanguard.
What are you investing for? The targetless fund has higher expected earnings; the whole point of the targeted fund is to get less money in expectation than the normal fund will.
Target retirement date funds are very different from reverse mortgages. In a reverse mortgage, you're borrowing money, which you pay interest on, which is secured by a large asset you own (your house). The reason they're so dangerous is because you're paying interest on the borrowed money, so it compounds, and can eat up your home equity very quickly.
With a target retirement fund, it's a normal investment which you receive investment returns on, but the asset mix moves from stocks to bonds as you get closer to retirement. The reason for this is that you generally don't want to put any money at risk in the stock market that you can't afford to lose within the next 5 years. Market crashes are unpredictable and often take several years to recover from; if you're in retirement and forced to live off that money month-to-month, you may be forced to liquidate stocks at very low prices, often lower than you paid for them, wiping out your investment gains. You're supposed to move assets from stocks to bonds or other more predictable income streams as you get closer to retirement, so that you know that the money you need to live on will be available when you need it, and can just let the money you don't need to live on sit in the stock market until you do need it.
Target retirement funds automate this rebalancing, dollar-cost averaging out of the stock market as you get closer to retirement so you aren't tempted to panic and pull everything if there's a crash.
You're confused; that's a regular mortgage. In a reverse mortgage, you sell your house to the bank for a defined monthly payment over X duration. It's a way of letting you spend the value of your house while figuring "hey, I'll die shortly before I lose the house".
Similarly, a target retirement fund is a way for you to pull your money out of the fund and hope that (1) you die just as you spend the last of it and (2) you didn't want to leave your family anything.
You can't live off of invested money by spending the principal. If that's what you're doing, you're not living off your investments, you're living off, and losing, your savings.
You're probably confused about reversed mortgages: my understanding of them is that you do not actually sell your house to the bank, you take a line of credit for a certain monthly payment with your house as collateral, and then rather than making payments on the loan, the interest compounds against the equity in your house. My stepfather was a former banker who did financial counseling for seniors (often including whether or not to take a reverse mortgage), so I do have some knowledge in this area...
I'll accept the correction on target retirement funds, but I still think "lose your principal preemptively now to avoid it possibly losing value later" is a misguided approach. The only justification offered for these funds is that you want to pull out of the stock market, with its volatility, before you die, so that you can enjoy spending your money (in my model) or lower returns (in yours). Assuming you have any family members, the point of doing so disappears -- you can't realize the future growth in your crashed stock portfolio after you die, but they can. All the same reasons you want stocks when you're 20 and don't have children suddenly apply again when you're 80 and do.
I will note that kelnos's sibling response to yours takes the same view I've been taking - that the point of a target retirement fund is that you withdraw the money so you can spend it:
> You don't want to be 85 and have to do a major withdrawal due to illness right when the stock market is experiencing a periodic down-swing
The targeted fund also gives you a (theoretically) predictable outcome, dependent on when you want to retire. You don't want to be 85 and have to do a major withdrawal due to illness right when the stock market is experiencing a periodic down-swing, causing you to burn through your retirement savings too quickly. The point of investing for retirement for most people isn't to attempt to end up with as much money as possible (which could easily backfire and turn into much less than you need), but to do a reasonably predictable job of having enough money to last you until death, based on your lifestyle and expected expenses.
If you want to shoot for the moon, that's up to you, but most people are more risk-averse, and a target-retirement fund fits that risk profile much better.
I'd love nothing more than to be able to buy stocks, real-estate etc through an online interface with 0 human interaction. Everyone I know above 40 seems to be the exact opposite.
Sometimes progress is made one funeral at a time.
I for one hate to run errand where i need to talk to someone. It is very inefficient, and socially awkward.
Even when you "have all the facts generated from sophisticated algorithms giving you an unequivocal 'correct' answer", how do you know it's not pulling a fast one on you? Do you trust whoever gave you the algorithm? Especially that it's usually not some diverse group of open-source developers, but a particular financial institution with a product to sell you.
I feel that the desire for human interaction in matters involving money has a lot to do with having someone you can look in the eyes and see yourself if they're trying to defraud you.
1. Automation can reduce transparency, by moving open discussions and reasoning into obscure, sometimes proprietary computer code. Effectively, the person writing the code, or the person who controls the coder, decides the policy. For example, some courts now use software to determine prison sentences; what was once decided based on established principles and transparently detailed by the judge is now decided in some cubicals by developers and written in C (which is often prioprietary). Why are you going to jail? The software (i.e., the developer) said so. Also, remember that code is mysterious and obscure to most people outside HN; they have no understanding of what it is, of its significance, what it can do, etc. Many can't define the word "algorithm" or even "web browser"; even open source code provides no transparency for them (and doesn't for us, unless someone makes the extraordinary effort to review the code).
2. To whom will automation shift power? The lobbyists with the best algorithm? The people developing the technology and solutions (see #1)? Will automation democratize power or further restrict it? Note that the developers of one solution in the article went to Harvard. Sounds to me like the same people running Washington, and the same gateways to power.
3. Lobbyists who have more data about constituents in general and about major players in particular will have an advantage. It could increase competition to erode privacy in its most important context - politics.
4. Will it result in better policy - more justice and better welfare - or just more power for the few with the data and algorithms?
If I were devious and wanted to change judicial policy nationwide, for example, I'd buy the companies making software for courts. I'd be surprised if I was the first to think of that.
It's an issue of individuals: People are neither completely altruistic nor completely greedy. A very important factor is, where are the elite on that continuum? My humble hypothesis is that that's the main difference between a banana republic and first-world civilization and society. IMHO, the elite in the U.S. have swung far toward greed; the tax bill is a good example: There is no policy reason for it, not even a figleaf discussion of economics, and even the principles of conservatives (deficit reduction) are thrown by the wayside: They just want more money.
It's also an issue of systems: How do we design a system that marries power, on one hand, to justice and welfare on the other? Then those who seek power also seek justice & welfare, intentionally or not. Democracy is an (incomplete) solution to that problem: Power flows from making the people happy.
Despite the problems, the U.S. and other major democracies work and have worked for generations, and they work better than any other system of government. Somehow, I'm not sure how, that happens.
So instead of writing laws in lawyer speak that nobody can understand anyways they would be written in C!
Lawmakers would essentially become the programmers
Heh, thinking more about "speed over correctness": we can't even get that right with people making the decision. Consider the secretive, rushed GOP tax bill that just passed...
I can't see any reason why my an application for a spot in school should have anything like the rigor afforded to
99% of government automation is of the "handle the easy stuff and dump the hard stuff to the operator to use their big fat brain" variety. In theory, well factored systems and machine learning should let more and more of that over to machines happily :)
Is that based on experience in the industry?
I think as politicos realize the power in controlling the software, they will use it as another way to exercise control outside the public eye.
The vast majority of dialog between governments and citizens is kinda boring with occasional exception handling. Think tax handling... The exceptions are demanding, and require expertise, and looots of time, but most other things get 'rubber stamped'.
Politicos get lots of control through the tax code and supplier agreements. Software will just make them better at it ;)
I'm not sure if automated decisions are a good idea. But it would be really cool to improve productivity of politicians.
How can we expect our politicians, let alone our lay people, to understand all of the law's in a country? There is just too much text to grok.
Software developers build abstractions, write tests, refactor and simplify older code, yet no such trend seems to be occurring for law. Why not?
Politicians already do use references and bucketization, i.e. write a law for schedule one drugs, then add or remove entities from the schedule one bucket. Can't we abstract even more?
Recently I've been wondering if it is feasible to create a syntax for lawyers. Like a law language with a compiler, package manager and similar tooling to compile laws rather than write them from scratch.
Anyone familiar with such research?
Just like the cynical reality that C/C++ programers don't want to encourage package managers. The higher the barriers to entry the more you're talents are worth.
The reality is, regardless of any one groups desire, the productivity wins of package managers are too great and the momentum of the larger software development community is moving towards package managers for every language and will one day include C/C++.
Similarly, lawmakers are a select few of a larger community of domestic but more importantly international lawyers. You give the larger community a 10x productivity win, and just sit back and watch the domino effect work.
Draftsmanship is not the hard part of law. The hard part is everything else.
_Question_: > what do you think can be done to take back US society and make it a little closer to what our seemingly more critically-thinking predecessors had lived through?
_Logic_: An individual can only critically think about or evaluate things that are presented to them.
_Issue_: If something is glossed over, for whatever reason (be it too difficult to understand), it has not been [and likely will never be] presented for evaluation unless ____(Reason #1).
The issue doesn't end with the lack of presentation -- instead it ends with the consensus that it's OK or necessary to gloss over things because ______(Reason #2).
Why do you feel the need to tell me this? I wouldn't have commented if I thought otherwise. P.S. There was no intention to be demeaning. Just intention to present the issue clearly.
As for the cutesy symbols, they're there to make clear that this isn't a problem because it's DIFFICULT to solve, as the solution is perfectly clear. Give individuals a voice by informing them and allow them to offer an evaluation.
The reason it's still a problem is because of a lack of leverage on the uninformed individuals' behalf.
No need to get worked up about simple things.
Please don't cross into personal rudeness, regardless of whether someone else did.
Edit: unfortunately you've broken the HN guidelines with incivility quite a bit in the past too. We ban accounts that do that. Please read https://news.ycombinator.com/newsguidelines.html and clean up your act if you want to keep commenting here.
That said, this is the secret. We market this to the law writers!
Getting to a point where we have 10x productivity gain will be hard but after that any tool with a 10x productivity win will be adopted. That is a no brainier.
Now we are in a future where all laws are written in this new syntax, but shared as long compiled documents full of redundant legalese English.
In such a future, lawyers can write law and contracts easily but cannot read them efficiently. Imagine writing JavaScript but debugging the assembly...
Lawyers would naturally start asking their peers to share the "source" rather than the compiled "binary", for evaluation. It is just a no brainier.
Now you have a future with a flurishing new syntax dominating the entire ecosystem. Hopefully better languages are created, etc. You even have an entire generation of lawyers who only know these higher level languages.
This isn't an overnight solution but by the time lawmakers realize the trend of higher level syntax is going to hurt their ability to hide truth, the game will already have been changed.
Just in case it isn't; we'd all benefit from laws that ordinary people can read, understand and follow.
When the quantity of law grows so great and complicated that nobody can feasibly read it becomes hard to be a law abiding citizen. Lawyers, despite rumors otherwise, are ordinary people too and benefit from clearly written law, in English for English speaking countries.
I agree that society benefits when we have succinct, easily understandable laws for lay people.
That said we are currently moving father away from that target, rather than nearer.
I would argue software is doing the opposite. It started extremely obscure and has slowly but surely moved in a direction where a majority of people can learn to understand it in months.
Example, a 3 month "bootcamp" can now teach a lay person to write HTML, CSS and JavaScript and accurately predict the output of the interconnections of millions (billions?) of lines on code.
You might argue the bootcampers don't truly understand what the kernel is doing. This is the point though, thanks to abstraction, they can grok it without needing to dive deep.
I don't think there will ever exist a world where lay people will understand legalese. However, I do believe we can build a world where people can be trivially self taught and then grok new laws, and even write new laws of their own.
HTML, CSS and Javascript are not comparably complex.
In software engineering we can set scope and conceptual boundaries.
Law cannot. It is complex because the world is complex. Unless you have a mechanism for simplifying human nature, the law will continue to require specialists.
Actually, the Common Law[1] (which the US, England, Canada, and other English-speaking countries) very much have a legal system that includes building abstractions, refactoring, and simplifying (though there are no tests). Judicial decisions are precedential and over time rules and tests are discarded or re-written to be easier to use and more applicable.
I think the parent comment is discussing legislation (which politicians enact), but the process by which judge made law develops and improves is very similar to how code evolves and improves.
Due to the messy uncertainty of life, and our seldom perfect information, there would still be a need for humans to make the final decisions, but an automated recommendation with a confidence indicator would really be something.
Aside from that, one of the saving graces in the American system is all of the friction points between various jurisdictions that prevents government from being ruthlessly efficient. I’m not sure that it would at all be a good thing if we could actually correctly codify and automatically enact the will of the people.
The law does do this.
There was a period during which this was so rigidly applied that the law forked into Common Law and Equity. The latter reintroduced fuzziness partly because of the increasingly absurd injustices of the Common Law at the time.
The law is filled with linguistic variables: "similar person similarly circumstance", "person having ordinary skill in the art", "buyer at arm's length without prior notice" and so on and so forth. These terms are all defined and understood, but they are not crisp, discrete concepts that can be safely reduced to a real number.
The law is a workable and incredibly reliable system, given the scale, scope, timeframes and consequences involved. Introducing something that flatters our own profession would not add net value.
The only thing I want to introduce is consistency through abstraction.
eg. If you want to write a law pertaining to minors, you need to define what a minor is. I'd like for the lawyers to just `import legal_entity::{minor};` and the definitions would self populate. Additionally, if you're writing a law about minors, you might need to contrast to non-minors, so you could just `import legal_entity::{minor, ...};`.
I'm sure even today definitions are relatively trivial to share, but sadly I'm not a lawyer and can't quite think of better examples of abstractions.
Maybe: eg.
current_entity.equals_with_specificity(person, Specificity.LevelThree)
&& current_situation.equals_with_specificity(situation, Specificity.LevelTwo /* Similar Circumstance */)> The only thing I want to introduce is consistency through abstraction.
By implication you're saying this doesn't already exist. Judges and legislatures undertake a constant, dynamic process of organising and reoranising the law into workable abstractions, seeking a level that renders a problem domain tractable. Typically these are called "rules", "tests" or "doctrines", depending on their scope and purpose.
> I'd like for the lawyers to just `import legal_entity::{minor};` and the definitions would self populate.
This happens when they say "minor" in regular legalese; a lot of boilerplate is generated from templates as-is. A lot of what solicitors call "bread & butter law" is repetitious, well-worn transactional law: wills, trusts, partnerships, real estate conveyancing and so on.
I do get some sense that you are mixing together statute and caselaw. These are both The Law, but the are created in different ways and have quite different forms. Statute is amenable to formalisation and indeed many legislatures have adopted "Plain English Law" drafting rules. Many legislatures also have reform acts intended to replace, consolidate, restate or remove various historical Acts resting on the books (you can find some amazing stuff in the attic).
But caselaw is very different. It is composed of judges giving reasons for their orders. It is a continuing conversation. There are no drafting rules and in general, there can't be, because the entire purpose of the system is that two parties can submit any legal question under dispute -- any question -- and be guaranteed to receive an answer.
Just as code often has "bugs" in the face of unexpected situations, so too can the law. But the law can be patched on the fly by judges (and later replaced by statute). It requires humans speaking in natural languages, which admit indefinite nuance in a way that artificial languages don't and shouldn't.
[0] http://clubtroppo.com.au/2007/08/25/programming-in-legal/
The only time you see the politicians opposed their own party is when the party have enough votes AND the politicians is from a state dominated by the other party. For example, a Republican from California.
If politicians have to do their own tax return, I can guarantee you the tax code will be a lot simpler than the 70,000 pages we have.
No way in hell is it producing probabilities calibrated to 3 significant digits. If some kind of testing of calibration produces proof it's even within 1 percent calibrated I will be shocked.
Toyota I can understand & accept, but is it appropriate for a branch of the executive to hire lobbyists to lobby the legislature? Just feels a bit unseemly to me.
> Quorum says its system also powers one well-known late-night comedy show—it won’t reveal which one—helping staffers scan YouTube to pull together almost instantly a selection of footage about topics they might use on that night’s show. Says Wirth, “They’re like, ‘What kinds of Roy Moore clips do we have?’” The software has already machine-read heaps of video footage, turning them into tagged transcripts, and sends off exactly the correct clip to that evening’s show. Millions of Americans, without knowing it, are getting their political news curated for them in part by a Quorum bot.
Sounds a bit like Last Week Tonight.
In the cases of lobbying & clip-selection alike, what I worry about is how we'll groom the higher-level people without running them through their lower-level paces (since the lower levels will all be automated). What I mean is that a lower-level job has at least two purposes: one is to perform a job, and the other is to audition for the next higher job. The former is what AI excels at. It's really, really easy to automate the job of an 18-year-old who answers phones and hands them to the next person. It's not so easy to automate the job of a 24-year-old who answers phones and hands them to the right person. It's really, really difficult to automate the job of a 36-year-old who answers phones and decides whether the topic is important enough to hand to his boss. It's really supremely difficult to automate the job of a 48-year-old who answers phones and makes decisions.
The problem is that every 18-year-old answering phones isn't just answering phones; he's also auditioning to be that 24-year-old finding the right person (experience he gets from the job itself). Likewise, every experienced 24-year=old is auditioning for the job of the very experience 36-year-old. And all those 36-year-olds are auditioning for the few slots as decision-making 48-year-olds.
We can really easily automate those lower levels away. But what will be our 'farm team'? Where will we grow the talent and depth of experience necessary for the higher levels?
Historically, we've seen the more tedious jobs get automated away to create better jobs that require more "human" skills. This is a good thing.
Having interned in a political office, the other problem is that the 18- and 24-year-old described often doesn't exist, because there aren't enough to people to have the heirarchy you describe exist at all; the 18-24 year old answering phones is doing much more what you describe for the 36-year old (answering phones, logging constituent positions appropriately if it's a position call, directing constituent services calls (and any other calls they don't handle themselves) to the right staff person. And, on top of that, doing a lot of non-phone tasks like drafting correspondence, assisting legislative staff with research, etc., etc.
Of course, lots of staff are never college interns and do their first staff job as a graduate fellow, or as professional staff often after completing a law, business, or public policy and/or administration degree.
So, I'm both not worried that the lower level will be automated easily or much concerned that there won't be an entry path if it is.
Hopefully no Silicon Valley startups think this is the next multi billion dollar industry to 'disrupt'.
I don't attribute any sort of sainthood or trustability to academics over other segments of society. It's certainly not clear that programmers are especially qualified to craft public policy. Again, instead of one group of elites that can understand what's going on you're substituting another.
Look, with Open Source software, programmers are mostly the consumers of that open source software. The customers of that software can more or less switch and aren't forced to use it at the barrel of a gun (and in the cases where big companies DO have monopoly power, the fact that they use Open Source software doesn't really stop them from exercising it in ways that are clearly for their benefit but less clearly for the benefit of their users). But if you're saying that software is now going to determine who can be arrested and who can go free, how long prison sentences will be or who can go on welfare or receive public money, there is HUGE incentive to game the system. Right now I might find Academics trustworthy because they don't get a huge financial benefit depending on the outcome of their research. But if public policy is controlled by software the incentives to cheat are huge. Now you can say they'll patrol each other, but that just means the incentive will be to collude collectively, as an entire group, for their own benefit against society at large.
If someone running for office says that they will make sure the local firehouse gets an X% increase in budget and that unemployment in the district will go down by Y%, there should be an unambiguous way of tracking, over time, how they are doing on their promises.
This "goal verification" seems like an important first step if one wishes to introduce meaningful computational processes in Washington. Ideally, over time, we would have a system against which we can run such promises: "The ministry of health says increasing the budget of X will lead to outcome Y, and our model agrees with this decision with a confidence of 90%".
Eventually, this system might even build excellent models of what happens in various parts of society when various parameters are modified. It could predict where new schools need to be opened based on the natality data provided by hospitals and local employment rates, what the plausible long ranging effects of lowering or raising corporate tax by x% would be, whether the opening of a certain factory requires increasing the budget of the firehouse in the neighboring district because the district is now over a certain threshold for fire risks, and so on.
As for modeling the effect of policy on the real world, lots of people already try to do this, and their results do not agree. It's a very hard problem, perhaps impossible.
It doesn't threaten lobbyists, it's a product for lobbyists.
However I bet it's one of the last places that actually gets automated. The value of the government is to give people reason to believe that everything is fine, that someone will deal with the big problems, etc. This story cannot be sold equally well if it's done by robots/AI. So even if we are really governed by AI some day these jobs will probably still exist. The show must go on.
[1] https://roadtrippers.com/trips/16260761?lat=40.80972&lng=-96...
Deleted comment