Blog

Back to Part 2

While individual humans seem capable of improving their abilities incrementally, both during their lifetimes (learning, building better habits, developing systems to deal with repeated situations, etc) and intergenerationally (Flynn effect?), it is not obvious that organizations currently do so. Robin Hanson lists some common inefficiencies of organizations; these problems seem endemic and timeless. Furthermore, while some organizations can last a long time and continue to do well, it is not obvious that they get “better” in a general sense over time. Rather, we tend to see an effect where organizations become overly specialized and decay. A good goal, then, is to hone in on how these issues can be dealt with systematically.

What would be the ideal theme for a culture built for an organization motivated to survive and thrive in the long run? It would seem to be some sort of growth mindset analogue at the organizational level. Specifically, great care should be taken at being aware of, and steering, the creation of systems and norms (knowing that they will tend to be sticky over time) with long-term effects and viability in mind.

Ideally we would know what healthy policies and internal systems should look like and build those from the beginning, but alternatively we can also work backwards from current failure modes (linked above); trying to identify them within an organization and finding solutions that maximize long term benefits. It seems to be that algorithms are potentially well suited to this task, especially as more data is being collected which covers the progress of work, communication, decision making and conflict within an organization. Perhaps patterns can be found across similar situations, and solution mechanisms can be tested across equally many samples. The kicker is if these solutions can be reused across organizations.

Currently we find this problem identification and solution proposition role being served on an ad hoc basis by management consultants. It seems intuitive that if organizations make a more concerted effort to record and organize as much internal data as possible, this role can be performed more efficiently with more statistical techniques. Given working models of failure modes and solutions, it is then possible to build systems which steer themselves away from the failure modes in the first place.

Back to Part 1

DNA and epigenetic effects are important and have a large impact on one’s potential and individual abilities. However, it seems clear that intragenerational adaptation, namely learning and acting via the neocortex, is the prime determinant of what an individual is capable of during a lifetime.

Luckily, there has been a whole lot of research into how to achieve these gains predictably. We have large institutions which take an organism which isn’t capable of much beyond farting and burping and turn it into a generally productive member of society. Much of this is done through the educational system, but the majority of the truly useful learning seems to happen more haphazardly, less formally. Specifically, learning to manage habits and develop personal systems for dealing with recurring challenges leads to massive individual effectiveness increases; having a growth mindset and working towards improving oneself makes life easier and makes you more effective as an individual.

Can these concepts be extended to organizations? Organizations do not possess a hard-coded DNA, which gives them tremendous flexibility but as a result also do not possess a default, time-tested foundation of decision making systems (for humans this would be instincts, emotions, etc) to fall back on. Organizations develop a set of systems, attitudes and practices which together can be considered their culture. These systems tend to be sticky once formulated, difficult to change throughout the life of an organization, and persist through generations of personnel. Similarly to how the cells of a multicellular organism will replace themselves throughout the organism’s life yet we still consider the organism intact, personnel will rotate in and out of an organization yet the organization (and more precisely, its codified systems and practices; the culture) will remain intact (but perhaps modified, over time).

As a result, the long term behavior of an organization can be said to be governed by its culture — it encompassing the role of both genetic code, instincts, deeply ingrained habits as well as systems for problem solving and learning that we see on an individual level. If we wish for better, more effective, organizations then we should be looking to intelligently design their DNA and maintain (hopefully, improve) the quality of the culture throughout the life of the organization.

Forward to Part 3

At some point on the path towards greater complexity, it becomes clear that multicellular organisms become “greater” than the sum of their parts (the individual cells). The general mechanisms and effects are clear; a combination of gains from trade due to specialization, sharing costly resources, etc. It also seems logical that such an effect should be possible with collections of organisms. Most relevantly, organizations of humans should be able to do things better than humans acting individually — to be more effective than the sum of their parts.

Not terribly controversial, but perhaps we don’t ask ourselves often enough why or when organizations fail to live up to their potential, and what that potential is governed by. We’re not very good at understanding this yet. Delving into this will be my objective over the next few posts.

To begin with, lets consider core motivations and intrinsic limitations. Humans (and most? carbon-based lifeforms) seem to be motivated most fundamentally by reproduction. This being a logical extension of their mechanism of creation. As a contrary thought experiment, we can envision a digital lifeform which is not constrained by DNA, biological cells, etc. and has no obvious motivation to reproduce [in the traditional sense]. Rather, it may be motivated to be more efficient, to learn more, or who knows what (perhaps maximize the number of paperclips in the universe?).

Similarly, most organizations (henceforth, specifically referring to typical formal and informal organizations of humans) are not motivated by reproduction. For-profit businesses tend to be motivated by profit; the creation of more money for itself, in order to extend the life of the business and, ideally, ultimately to enrich the owners monetarily. I will focus on this specific type of human organization in later posts, mostly because they tend to be the drivers of innovation and have the largest tangible impact on our lives. That seems like an important thing. Other organizations can be equally important to our lives, specifically social structures which satisfy important emotional needs for humans, and similar analysis can be applied to them as well.

Forward to Part 2

A world where privacy is non-existent can be a very beneficial one. Scott Adams writes about this once in a while and it is eye-opening to those that aren’t familiar with the rapid progress in data collection and analysis technology, or haven’t connected the dots to what can be done with that analysis. The feasible short-run achievements would include drastic reduction in crime (physical, as well as white collar and identity theft), drastic reduction in adverse selection in insurance and financial services (rewarding good actors with lower prices, better service, etc.); you get the idea. In the long-run, businesses that succeed will be those that can use the available data on your life and personal preferences to customize products and services to meet your needs in more granular ways and with less effort required on the part of customers.

Most people tend to get queasy and defensive when confronted with giving up privacy, even in exchange for a better life. Dystopian portrayals are available aplenty, see Minority Report. And unfortunately for those people, there seems to be little that can be done to stop the erosion of privacy. Laws and regulations could certainly slow down the process but the information will only become easier to obtain with better technology, and pushing it into black markets and towards non-transparent usage will only increase the likelihood of a dystopian outcome. There are other ways…

The amount of data being collected on individuals, licitly and not, is reaching a tipping point. At the moment, we find ourselves surrounded by growing walled gardens hell-bent on knowing everything about us. Google, Facebook, Twitter, etc. have collectively built hundreds of billions of dollars of market value providing us with convenience and better access to information about the world and other people we care about. In exchange, we let them harvest our data and sell it to interested third parties (mostly businesses selling us things). There are two major issues with the current model. We cede ownership of this data to services which, while potentially trustworthy and having good motives, are not infallible. They are also not terribly transparent, so we must take them at their word. The degree to which our data is protected from theft by less benevolent actors is unclear. And there are less obvious but potentially just as harmful issues. With the data locked up inside walled gardens, we stifle the innovative potential of new companies which could potentially use this data in even more beneficial ways.

In the long run, with the right technological foundation, cultural norms will change and we will not only be eager to allow others to access our data and use it, but we will have better knowledge and control as to how its used, as well as potentially being compensated directly (perhaps even monetarily — this would rapidly increase the pace of widespread adoption since frankly most people don’t care much about the security of their personal data, but everyone likes money). For this, we must align incentives.

I can’t propose a precise solution, but the general features of one come to mind.

1. We should insist that the data we generate, whether its from posting status updates, letting our phones track our location, or in the future allowing devices in public spaces to identify us and analyze our activity (actually, its happening already), is not only in our possession but also under our control. More precisely, it seems like what we need is an open source framework for encrypting data we generate and placing it into a storage system which we can allow conditional, piecemeal access by selected parties.

2. The open system will allow for transparency in understanding the security features surrounding our data. I’m not sure to what extent this would be feasible, but ideally you prevent bulk transfer of data out of the encrypted store (unless by the owner, obviously). This is tricky. Off the top of my head, the data could be categorized and stored in a query-friendly format. External access would be through the form of queries and the amount (as well as the specific data points) decrypted and exported as the query result can be tracked. If you can make the argument that personal data decays quickly — e.g. data from a year ago isn’t nearly as valuable as data from yesterday — an alternative approach could be as simple as changing the private key for your data’s encryption at some reasonable frequency (monthly). Then when your data is accessed by a third party, they will need a different key for each piece of history, as well as perhaps for different data types (a key for your status updates, a separate key for your pictures, etc.)

3. To get the full benefit, the system would need a secure audit trail of access (what data and by whom) as well as a gatekeeper mechanism which controls this access to some level of granularity. Interestingly, the recent innovation of a publicly cryptographically signed chain of transactions (blockchain, i.e. Bitcoin) seems potentially applicable here. In a nutshell, different pieces of data owned by an individual and stored at a trusted service running the open source protocol would by encrypted by various private keys (which could even change over time for the same data set). Access to data could be enabled by a transaction which is logged to a publicly-visible block chain. You could use a unit of account which must be transferred from accessor to owner in exchange for this access. Think of it like a cryptocurrency for your private data. You can see where that brings us…

A system similar to one described above seems to enable an alignment of incentives across many parties:
– makes individuals eager to rent the data, but critically to have it be their choice and be potentially compensated for
– creation of an ecosystem which will be incentivized to improve the infrastructure of the security, storage and analysis of this data
– enabling new businesses, outside of the walled gardens of current internet giants, which need personal data to work well
– a clearer accounting of who is accessing your data, including government agencies. if they want to know everything about us, lets at least make that transparent

Adoption of such a mechanism seems like a chicken/egg problem at first, and also seems vulnerable to exploitation by nefarious parties by collecting data through the system and reselling it themselves (or using it for purposes that you didn’t intend to allow). It seems to me that grass roots adoption of such a framework by individuals currently eager to share their personal data with innovating companies (in exchange for whatever service they offer for free, basically) will seed the system and begin drawing adoption by new entrants as a selling point as well as a data source to feed their business. Eventually, we should see increased pressure for large incumbents (google, etc.) to move their data into such an open system as a show of transparency and good faith towards their users — especially if people start to see that they can be compensated directly for access to their curated data. As far as access by nefarious parties, I would argue it’s not any worse than the system (or lack thereof) that currently exists in the form of walled gardens and incompetent government agencies. At least with an open source mechanism we can have a concerted and transparent effort towards preventing our data from being abused.

The ideas here are half-baked and create plenty more new problems beyond the ones addressed (how to prevent bulk extraction, how people can manage large sets of keychains, how to manage large permission lists, etc.) but the key is the to create a platform of transparency within which these problems can be incrementally chipped away at. Currently, we’re flying blind.

At the beginning of the year, with BTC around 600 to the dollar, I made a bet with a friend (I was given 3:1 odds) that BTC/USD would be below 150 at the end of the year. I think there are some core flaws with cryptocurrency that, in order to alleviate them, effectively make their benefits moot.

As a target for theft, Bitcoins are basically a dream. Once transferred away from an owner they can be mixed with other (fungible) BTC and be no longer feasibly traceable to the a single thief. To harden the target, you can put BTC into a wallet that’s in cold storage (put it on a hard drive, take hard drive out of computer, hide under bed). This is all well and good, and there are even fancy services popping up that will do all kinds of things to ensure the security of your coins, but at this point you can ask yourself — why? Is it really that good of a store of value that it’s worth going to such an effort to preserve their possession? In fact, I see few sensible arguments for BTC being a useful store of value other than as it’s value as a transactional tool…

I think the concept of a block chain is interesting and has potential for many applications (that’s a whole other argument, here I only refer to Bitcoins and other currency usage). If the exchanges gets super liquid, you may even be able to use one for transactional purposes for lower total cost than a credit/debit card (debit is really the only applicable comparison here). I could feasibly see, with a large enough transaction flow, the profit potential from market making into fiat resulting in tighter spreads than the ~2% paid for interchange fees using plastic. How much lower though? A few basis points would be reasonable I think — similar to liquid electronic financial markets. But for market makers to be able to get it that tight, you would need pretty low underlying volatility — a high confidence in the stability of the value.

Chicken and egg? It seems like a solid bet that continued highly publicized thefts will become the norm and this will degrade confidence in the value. Without that confidence, it’s unlikely to gain the significant liquidity necessary to succeed as a transactional platform.

What purpose citations? It seems like they are extremely susceptible to being a manipulative, rather than informative, tool for any discourse other than in hard sciences where reality is close at hand. A reference on a topic or opinion that you are deeply familiar with is of little use to you. However, the majority of the time we deal with what we are not experts in. Usually there is some inferential distance that needs to be crossed, or at least some learning of the inevitable multitude of points of view surrounding an issue. Referencing particular arguments or external sources, unless done extremely carefully and in a well-balance manner, seems to be an even greater signal that the original document aims to persuade rather than inform.

For the past 8 months or so, I’ve been making a 1 liter dense smoothie about 3 times a week. On those days it makes up a little more than 50% of my caloric intake I would imagine. I tried to come up with something that would cover most of my micronutrients as well as have a low glycemic load, taste decent and be easy to consume (drink). It’s nice, I get lots of energy and never get that tired feeling that comes after most meals. I also think it tastes good, but not everyone agrees 🙁

I’ve been curious as to the actual nutritional breakdown of my typical recipe. The ingredients vary slightly but are usually something very close to the ones listed below. I suspected I wasn’t getting much protein, and that appears to be the case. I’m not worried about the relatively low carb intake as it’s  pretty easy for me (and most people) to eat carbs. My dinners (and lunches on non-goop days) are vegetarian about 75% of the time with probably more carbs than necessary in a single meal.

Here’s the ingredient breakdown. There’s obviously variation between this and what I’m actually eating, but it’s pretty close. I checked the nutritional info on the packages of the stuff I have that comes in packages and it seems close enough to the numbers below.

CALORIES		QTY	MEASUREMENT	FOOD
80	4.90%	2	1 individual (packet)	Almond milk  Blue Diamond Unsweetened
227	14.00%	1	1 fruit	Acovados, raw, California, without skin and seed
105	6.50%	1	1 medium	Bananas raw
42.2	2.60%	0.5	1 cup	Blueberries raw
33.5	2.10%	1	1 cup	Kale, chopped, raw
114	7.00%	0.71	1 ounce	Nuts, Almonds
130	8.00%	0.71	1 ounce	Nuts, Brazilnuts, dried unblanched
110	6.80%	0.71	1 ounce	Nuts, Cashew, raw
125	7.70%	0.71	1 ounce	Nuts, Hazelnuts or filberts
137	8.50%	0.71	1 ounce	Nuts, Pecans
130	8.00%	0.71	1 ounce	Nuts, Walnuts
389	23.90%	1	100 grams	Oats
1.9	0.10%	1	1 pepper	Pepper, serrano, raw

Data is from the USDA database, and the extra metrics + pretty pictures is generated through http://nutritiondata.self.com. The site makes pretty graphs but is very painful to use. Most of the time when updating the meal composition it would fail and load stale data or miss some new data that was saved already. I would not want to try to use that to try variations on a combination of ingredients but I wasn’t able to find a better tool. I also don’t like building GUI’s, otherwise I’d make my own.

 

More or less what I expected. My guess is it may be worth putting a scoop of protein powder in and maybe cut down a little on the nuts to keep it from getting too viscous.

A pretty good overview of paleo nonsense ends with a good summary of what science can agree on at this point:

Having talked to all of these people and read their work, here is how I walk away from this. Oxidative stress will increasingly be the target of medical treatments and preventive diets. We’ll hear more about the role of blood sugar in Alzheimer’s and continue to focus on moderating intake of refined carbohydrates. The consensus remains that too much LDL is bad for you. We do not have reason to believe that gluten is bad for most people. It does cause reactive symptoms in some people. Peanuts can kill some people, but that does not mean they are bad for everyone. I agree with Katz that the diets consistently shown to have good long-term health outcomes—both mental and physical—include whole grains and fruits, and are not nearly as high in fat as what Perlmutter proposes.

The amount of hype over a major dietary change with so little scientific evidence is not surprising; this ain’t the first time

A very interesting potential solution to the Fermi paradox which turns the Kardashev scale on it’s head. It suggests that intelligent lifeforms will have a tendency to evolve and expand inwards rather than outwards. It seems like this balances on the idea that transforming biological lifeforms into a denser substrate (computational on silicon, initially?) and then acceleration of energy efficiency gains by increasing the “computational density” will occur quicker than expansion into space (very time consuming without energy to reach very high speeds) to find more energy.

In an Aeon article from last year, David Deutsch seems to be taking a dim view of Bayesianism as a core component of an AGI:

The doctrine assumes that minds work by assigning probabilities to their ideas and modifying those probabilities in the light of experience as a way of choosing how to act. This is especially perverse when it comes to an AGI’s values — the moral and aesthetic ideas that inform its choices and intentions — for it allows only a behaviouristic model of them, in which values that are ‘rewarded’ by ‘experience’ are ‘reinforced’ and come to dominate behaviour while those that are ‘punished’ by ‘experience’ are extinguished.

Or rather, since the Bayes AGI just does what rewards it the most, it will align its ‘values’ to correspond to that. I think the problem is with the hidden meaning in the word values. What he is referring to here is probably something along the lines of a compressed version of the AGI’s internal decision making model. In fact, you could say that this is in theory what we consider the personal or ethical values of a human being to really be. Of course in practice a human’s actions frequently depart from his advertised values — probably by design.

The core problem here is not with Bayesianism but rather that an unfortunately designed AGI could find itself with a fitness function (and corresponding set of values, ethics, etc.) which are not necessarily friendly to humans. Or perhaps he’s suggesting it’s a problem with any AGI which has a goal system. What is the alternative, in that case?