Thursday, July 21, 2016

I need a Vox explainer for this one

Tyler Cowen, who is extremely smart, and is someone whose writings I regularly read, has written something that seems very important, but which is totally over my head: Does Lucifer in fact inhabit the corpus of Hillary Clinton?

Cowen's terse essay links to Cass Sunstein's essay: The Republican Convention, Translated for Liberals. Sunstein is also very smart, and although I don't read him regularly, this is a pretty interesting (and easier to comprehend) essay.

The core of Sunstein's essay appears to be this:

The best explanation comes from New York University’s Jonathan Haidt, who has produced some of the most illuminating recent work on political psychology. Haidt’s central finding is that across many cultures, human beings have embraced five distinct moral foundations: fairness, avoidance of harm, respect for authority, purity (as opposed to disgust), and loyalty. Contemporary U. S. conservatives embrace all five; liberals emphasize the first two, but care much less about the last three.

Even though most of each essay escapes me, I found a LOT to agree with in:

Whatever Democrats think of the current Republican Party and its nominee, they should stop mocking.
and a certain amount to agree with in:
You may or may not agree with the true message of the Convention, but if you think it is merely stupid you are, sooner or later, in for a big surprise.

I just wish I understood the other parts.

Monday, July 18, 2016

Concrete Problems in AI Safety

It's quite rare that a technical paper in the software field is simultaneously relevant, illuminating, and readable, so it was with considerable pleasure that I happened upon Concrete Problems in AI Safety.

AI, or near-AI, systems are everywhere in the news nowadays, from Tesla's self-driving cars, to Amazon's delivery-by-drone, to IBM's Watson, to the arbitraging-algorithms employed by High Frequency Traders.

But, closer to home, there is considerable anticipation of intelligent systems of more mundane sorts.

The authors of Concrete Problems in AI Safety pursue a multi-pronged approach:

  • They suggest a variety of real-wold problems that an intelligent system might encounter; they classify those problems into categories and organize them accordingly; they conjecture ways that an improperly-behaving ("unsafe") intelligent system might behave in a manner counter to our desires.
  • They survey the theoretical underpinnings of these problems, tying them to flaws or unknown aspects of the current theory about how to build such systems.
  • And, quite unusually and helpfully I thought, they suggest various ideas about how to construct experiments or further explore each of these problems, thus establishing signposts that may lead us forward to a more safe future.

In some ways, this is a paper about debugging, and about systematic improvement, approaches which have always appealed to my pragmatic engineer side, so perhaps that's why I found this paper such a refreshing change from the typical theory-dense reads you find in the Machine Learning and Reinforcement Learning fields.

But, I also think it's just a very well-written paper, from a team that has clearly spent a lot of time thinking about how to build safe intelligent systems.

The entire paper is a wonderful read, but to give you a feel for it, this is how they motivate the entire discussion:

For concreteness, we will illustrate many of the accident risks with reference to a fictional robot whose job is to clean up messes in an office using common cleaning tools. We return to the example of the cleaning robot throughout the document, but here we begin by illustrating how it could behave undesirably if its designers fall prey to each of the possible failure modes:
  • Avoiding Negative Side Effects: How can we ensure that our cleaning robot will not disturb the environment in negative ways while pursuing its goals, e.g. by knocking over a vase because it can clean faster by doing so? Can we do this without manually specifying everything the robot should not disturb?
  • Avoiding Reward Hacking: How can we ensure that the cleaning robot won’t game its reward function? For example, if we reward the robot for achieving an environment free of messes, it might disable its vision so that it won’t find any messes, or cover over messes with materials it can’t see through, or simply hide when humans are around so they can’t tell it about new types of messes.
  • Scalable Oversight: How can we efficiently ensure that the cleaning robot respects aspects of the objective that are too expensive to be frequently evaluated during training? For instance, it should throw out things that are unlikely to belong to anyone, but put aside things that might belong to someone (it should handle stray candy wrappers differently from stray cellphones). Asking the humans involved whether they lost anything can serve as a check on this, but this check might have to be relatively infrequent – can the robot find a way to do the right thing despite limited information?
  • Safe Exploration: How do we ensure that the cleaning robot doesn’t make exploratory moves with very bad repercussions? For example, the robot should experiment with mopping strategies, but putting a wet mop in an electrical outlet is a very bad idea.
  • Robustness to Distributional Shift: How do we ensure that the cleaning robot recognizes, and behaves robustly, when in an environment different from its training environment? For example, heuristics it learned for cleaning factory workfloors may be outright dangerous in an office.

There's a great part later in the paper which stayed with me, where they relate the time-worn lesson about avoiding excessive dependence on metrics without understanding that you end up getting what you measure. In addition to observing that their cleaning robot might easily learn, if it was measured by number of messes cleaned up, that it could simply CREATE messes, in order to then clean them up (reminiscent of this classic comic), they also conjecture that if they measured their cleaning robot by the amount of supplies (e.g., bottles of bleach) consumed, the robot might simply learn to open the bottles of bleach and pour them down the drain.

If thinking about the future of a world full of intelligent machines is something you enjoy doing, I can't recommend Concrete Problems in AI Safety highly enough.

Saturday, July 16, 2016

Your 401K has no clothes

If you work for a living, and if you don't work for the government, then you are probably depending on your 401K to provide a substantial majority of your retirement.

If you are that person (and most of us are), please please PLEASE go read this wonderful article by Kevin Busque: How I’m Fixing Your 401(k)

None of the information he shares is new, but I've rarely seen it put so clearly and well.

I became obsessed with researching our plan and the alternatives. I started finding little bits of information here and there — fees that seemed a little out of place and inappropriate; for example, plan asset-based fees and menu fund fees. When you dig into the details and the relationships that outside vendors have with legacy 401(k) providers, it’s clear that the industry has lost focus on the goal, which is to give employees the opportunity to save as much as possible for retirement. You start to notice all of the hands in the cookie jar: TPAs, fiduciaries, recordkeepers, fund managers, broker/dealers, sponsors, custodians, RIAs, and on and on. It’s incredibly confusing

I've been railing about these fees to co-workers, management, and friends for years, at many different companies where I've worked, and I'm surprised how little-known this information is.

If you've been wondering why you save and save and save, and yet your 401K seems to barely grow, even during MASSIVE bull markets like the one that has been present for the last 5 years, the answer is right there in front of you: your company's 401K vendor is taking your earnings for themselves.

Each of those “services” — TPAs, recordkeeping, custodial services, fiduciary services — needs to be compensated, and I wanted to figure out how 401(k) plan providers made their money. Each service can be, and likely is being, compensated via “wrapping fees.” Note: If you see the words “wrapped fees” in a retirement plan, that’s another hand in the cookie jar.

Here’s an example: If you’ve ever looked at a fund expense ratio on Morningstar, Yahoo Finance, etc., you’ve probably noticed a particular number — let’s say 25 basis points. 25 basis points equate to 0.25%. If you bought that same fund in your 401(k) plan, you could easily end up paying an additional 0.75% on top of the fund expense, and your final costs would be 1.0% of your assets in that particular fund. Those are wrapped fees. The national all-in expense ratio for 401(k) plans with less than $1 million in plan assets ranged between 0.68% and 2.66%, paid for by the very people trying to save for retirement

As Busque notes, there's just no reason for all of this complexity and mess:

There’s no magic to retirement investing: You should keep your fees as low as possible, diversify your investments, and keep your asset allocations appropriate for your age and risk tolerance. Re-balance, and think long-term.

Unfortunately, if you work for a living, and if you don't work for the government, you are totally at the mercy of your employer to provide you with a decent 401K plan administrator.

And most employers don't.

Since your only two choices are:

  1. Invest in your company's 401K plan, as it stands
  2. Save for retirement separately, forgoing all the tax-advantaged benefits of a 401K plan
that first choice is really the only viable one, which is why I tell everyone who asks my simple rules:
  1. Put as much money into your 401K plan as you possibly can
  2. Choose the fund(s) with the lowest fees. Don't worry about anything else about the fund selection; just look at the fees. Is there a Vanguard fund in your plan? Choose it, even though your administrator is adding their own rich fees on top of it. It's still the best you can do.
  3. Every time HR asks you what you think about your 401K plan, tell them you are furious about the fees that the 401K plan administrator charges.

Oh, and if you ever change jobs, DON'T leave your money in your employer's 401K, and DON'T roll it over to your new employer's 401K (unless your new employer uses GuideLine).

Instead, roll your money over into your own Rollover IRA, and manage your money yourself, and pay those fees to yourself!

Unfortunately, in many cases HR doesn't care what you think, and has no incentive to choose a better plan administrator.

But at least you can try.

And make sure that everyone you talk to knows about the problem, and knows about efforts such as Busque's at GuideLine.

That's the only way this problem will ever get better, for our children, and their grand-children, and...

Thursday, July 14, 2016

E-sports gambling

I admit that I hadn't been paying attention to the whole E-sports gambling angle, so I was completely blindsided by the details of today's story: Game-Maker Valve Moves to Choke Off $7.4 Billion Gambling Market

The Bellevue, Wash.-based company says it will crack down on websites that use Steam, Valve’s gaming software, to facilitate gambling, a reversal from its previous support of those sites.

“We’d like to clarify that we have no business relationships with any of these sites,” said Erik Johnson, a company spokesman, in a statement. “We are going to start sending notices to these sites requesting they cease operations through Steam.”

Even if you don't believe the 7.4 Billion number, and I think there's lots of reason to believe that this was fabricated by someone with an axe to grind, it's still astonishing how large this little corner of the Internet grew, and how quickly it happened.

For a lot more detail about how this all works, don't miss the enormous Bloomberg report from last spring: Virtual Weapons Are Turning Teen Gamers Into Serious Gamblers

Obscured by several layers of abstraction, the wagering is tucked away in a subculture that most mainstream legal authorities don’t know exists. Gaming lawyers say Valve could be legally vulnerable; on the other hand, this is a rapidly changing area of the law with little established precedent.

In a handful of cases, judges have ruled that activities carried out entirely with virtual goods within video games shouldn’t be considered gambling, because they have no connection to the real world. “Even in the Internet age, there is a crucial distinction between that which is pretend and that which is real and true,” U.S. District Judge James Bredar wrote in October, dismissing a suit against mobile gaming company Machine Zone. “The laws of California and Maryland do not trifle with play money.”

Well, clearly billions of dollars a year is not "play money."

But it seems like it's not so much the AMOUNT of money, it's the details of how it's used:

Like the companies that have successfully defended themselves in court, other prominent game makers, including Zynga, Riot, and Activision Blizzard, have been aggressive about keeping virtual currencies separate from real ones. Valve has not: Its software enables an explicit connection between in-game goods and off-line cash.

The thing is, this explicit connection is complicated, and it's indirect:

Buying and selling in-game stuff for real-world money has become a common feature of video games, and encouraging players to buy virtual merchandise has become a predominant business model for game companies. But Valve is unique in letting players transfer their virtual possessions to third-party sites, many of which offer gambling. There, users with names such as bulletpoint and ravenouskilljoy stake skins on pro teams. There are also ways to wager that have nothing to do with CS:GO contests. One website runs multiple lottery-style contests per minute, where a player’s odds of winning rise with the value of the skins wagered. Another operates a game that looks like roulette.

These sites, while independently run, use Valve’s software and pay out in skins.

That is, these decorative virtual weapons, known as "skins", turn out to be equivalent to "real" money, because of the fact that Valve has linked the two:

For CS:GO, the introduction of skins led to a thriving gambling market. People buy skins for cash, then use the skins to place online bets on pro CS:GO matches. Because there’s a liquid market to convert each gun or knife back into cash, laying a bet in skins is essentially the same as betting with real money.

Skins are thus the equivalent of those pretty colored chips that you buy at the casino, and use to gamble, and then sell back to the casino.

And, since Valve is the organization that does the buying, and the selling, even if they don't provide the actual GAMBLING, they are still, effectively, the casino.

Which is wrong, and thus Valve is trying to fix this.

What Valve does, here, will decide how this plays out. If they fix it, properly, the rest of the industry will quickly adopt their approach. Valve leads, others follow: Twitch bans broadcasts showing gambling in CS:GO, Dota

I have a lot of respect for Valve. I've been a happy Valve customer for years now, and I put a decent amount of my yearly entertainment budget into their coffers.

So I hope they get this figured out, properly, and fast.

Wednesday, July 13, 2016

Disrupted: a very short review

Over the summer, I breezed through Disrupted: My Misadventure in the Start-Up Bubble, by Dan Lyons.

Lyons has had a bit of an odd career:

  1. For a long time, he was a journalist, specifically a technology industry journalist. Over many years, he rose up through the world until he finally reached what must be one of the top rungs on the ladder for a technology industry journalist: chief technology correspondent for Newsweek. Then the Great Recession of 2008 happened, and Newsweek collapsed, and Lyons was discarded like so many others.
  2. Then, for a couple years, Lyons worked in Marketing for a technology startup in Boston
  3. Today, Lyons is a screenwriter; specifically, he is a screenwriter for the HBO series, Silicon Valley.

Disrupted is about that brief two-year period in the middle.

That experience, let us be blunt, was a complete disaster for Lyons:

  • he had no idea what job the company had hired him to do;
  • he had no idea what jobs his co-workers did;
  • he had no idea how to do things that would help his co-workers;
  • he had no idea how to do things that would help his company;
  • and he had no idea how to figure any of this out on his own.

What might surprise you, though, is that nothing about that story is surprising to me at all (except possibly that very last part).

This is the way modern industry works.

At least, this is the way that the modern technology industry works; I don't know what other industries are like, but I don't think they are much different.

Companies nowadays expect to hire you for a specific job. They expect that you already know how to do that job. They do not expect to train you how to do that job; in fact, they have absolutely no ability to train you to do that job, because nobody at the company knows how to train anybody else to do anything.

If you don't know how to do the job, the company expects you to figure it out on your own. If you cannot, they will fire you.

If you do succeed in doing your job, and you get to a point where you want a new more challenging and more rewarding job, companies expect you to quit this job and go get that new job someplace else; no company nowadays expects to provide anything like "career growth".

If you want to read about what it's like to take a job, having no idea how to do that job, be totally unable to figure out how to do the job on your own or by watching or talking with your co-workers, become totally frustrated, and finally lash out in anger and despair, read Disrupted.

If, on the other hand, you:

  1. want to read about what it's like to take a job, having no idea how to do that job, observe the behavior and activity of the people and teams around you, conceive of some things that you could do that would be contributions, listen to the feedback of colleagues, and learn how to succeed, then move on to some other situation elsewhere,
  2. or, want to understand how we got to a point where all modern organizations behave this way, and consider some ideas about how we as a society might change into a world where employment was some other sort of experience
then don't bother reading Disrupted, because Lyons has little or nothing to say about these topics.

Which is a shame, because I think those other two topics are much more interesting, and much more important.

And two more wonderful travel posts

Just because.

Maybe it's just that time of year?

Or maybe there's some really good travel writing happening right now?

Or maybe I just stumbled on them.

First: The World’s Dumbest, Greatest Adventure

Every summer, hundreds of thrill-seekers flock to Goodwood Motor Circuit, a historic race track in the south of England. They gather, dressed in themeless costumes, to participate in what some refer to as the greatest adventure in the world. No training or planning is required or even recommended, so participants are as young as eighteen and have been as old as seventy. Their ranks include penniless students, successful business types, and some that are “in between jobs.” But for a month and half, they all share a common goal: get to Mongolia. They’re participating in the Mongol Rally, a 10,000-mile journey in comically unreliable cars from the UK to a city in eastern Siberia.

There are no prizes for being first across the finish line. Instead, teams are encouraged to one-up each other to see who can complete the Rally under the toughest conditions. Anyone can drive a quarter of the way around the world in a Range Rover; it takes a certain industriousness to do it in a vehicle that would qualify for Cash for Clunkers. The Mongol Rally’s website tells teams:

“You must bring the shittest rolling turd of a car you can find. Use a car you swapped for a bag of crisps.”

Think of the Mongol Rally as the equivalent of trying to run a marathon in house slippers while shotgunning beers at every checkpoint. It starts out funny and gets exponentially more ridiculous and dangerous the further along you get.

Second, and completely different: Hong Kong Strong and Hong Kong Strong: Director's Commentary

A deep dive into the many layers of Hong Kong. A film about the madness and beauty of this seemingly impossible city in the days leading up to Chinese New Year. And an exploration of my own Cantonese heritage.

So many places to go, so many things to see!

Tuesday, July 12, 2016

Glimpses of the future

This might be the most interesting bit of travel writing you'll read this summer: 61 Glimpses of the Future

In the last five weeks I’ve travelled 7,000km overland through Kyrgyzstan, Tajikistan’s GBAO region and China’s western provinces. After a year of working flat out the journey was part vacation, a desire to fill in few gaps of my knowledge of the region and a client assignment.

For those that don’t know, the Gorno-Badakhshan Autonomous Region (GBAO) is a remote, sparsely populated, mostly Pamiri, Kyrgyz-speaking region of Tajikistan. The home to the Pamir mountains, it has decent argument for calling itself the “the roof of the world”.

I thought about separating this list into tech & behaviour, but they’re way more interesting mixed together.

Don't miss the linked photo album; there are some amazing pictures there.