Uncertainty Wednesday: A New Tack

After a six week hiatus I am resuming Uncertainty Wednesday. Much as I have done at varying points with my prior Tech Tuesday series, I have decided to head off in a different direction, at least for some time. Instead of the theoretical approach so far, I will focus more on the philosophy of living in a world full of uncertainty. How should we live our lives, make business and personal decisions, judge performance, etc. given pervasive uncertainty?

My plan is roughly as follows, but I am open to suggestions. First, I intend to write a bit about just how much of our lives is impacted by uncertainty (hint: all of it) despite us largely not acknowledging this reality. Then I plan to look at examples that illustrate how poor our intuitions are when it comes to dealing with uncertainty. With that in place, I will share the answers I have arrived at for myself for how to live with uncertainty.

An interesting exercise is to look back at one’s life and think about how events that had a big impact came about. Here are three examples from my life: the first was my father mentioning rather off-handedly when I was young that a year abroad might be a good idea. Without his comment I might not have applied for a 1-year stay in the United States – none of my childhood friends did anything remotely similar. Yet living with a wonderful host family in Rochester Minnesota when I was sixteen years old was a crucial formative experience in my life.

The second example is meeting Susan, my now wife of 21 years and mother of our three children (and co-founder and CEO of Ziggeo and so much more). I was sitting in a cafe in Paris in the Carrefour Odeon when I heard two non-native French speakers conversing diagonally behind me. As I was in Paris to practice French myself, I decided to turn around and say hello. That’s how I met Susan, who herself was there by chance having abandoned a trip to Belgium to visit a relative of hers in Paris!

My third example is how I wound up at Union Square Ventures. My now partner Brad had been an investor in and board member of an accelerator I had co-founded in 1999. Following the implosion of the dotcom bubble, Brad and I tried to raise a fund (together with another friend) but nobody wanted to invest in Venture Capital in the “nuclear winter” that ensued. I went off to work on another project, which also failed about two years later. By that time Brad and Fred had raised the first Union Square Ventures fund. They asked me to take a look at del.icio.us with them and after the purchase of that company I wound up joining USV as a venture partner.

All three of these events had a massive impact on my life. And yet in each case it is quite clear that there was a huge amount of uncertainty involved, meaning each of these events might also not have happened and my life would be quite different today.

Labor Day: AI & Labor

We are finding ourselves in a strange place this Labor Day: the impact of AI is simultaneously being overestimated and underestimated. There is still massive hype surrounding self driving cars, which will almost certainly take much longer to arrive than the public appears to believe. At the same time, there is mostly quiet but massive AI progress on tasks such as reading radiology images.

The key thing to understand – if you want to form an opinion on AI progress – is to distinguish between “open” and “closed” domains. The difference revolves around how much knowledge is required in adjacent domains to be effective in the domain itself. In an open domain you need to know a lot about other domains as well. 

Driving is an example of an open domain: to be an effective driver, you don’t just need to know about the road and the car and the physics of acceleration and breaking, but you also need to know about people and their intentions (eg is this person about to cross the road or not?), stuff that can be found lying on a road (eg a plastic bag is nearly weightless and you can drive over it but a piece of metal might puncture your tire). 

Reading a radiology image, by contrast, is an example of a closed domain. The image will have a part of the human body on it and you are supposed to find something that’s wrong, e.g. a broken bone or a tumor. In order to find that you do not need to know anything beyond that; for instance, you do not need to know how the bone was broken or how the tumor might be treated.

AI is making rapid progress on closed domains, such as playing a game or reading a radiology image. As it turns, out much of what humans do in the workplace today operates on closed domains. That’s in fact so by design and has been the hallmark of Industrial Age organization (by this I don’t just mean manufacturing but also services, such as banking, insurance, etc.) We have over a long time carved up jobs through specialization so as to reduce the knowledge required for carrying out the task.

The hype around self-driving cars will likely disappear and that could happen any day. Don’t let that fool you about the prospects of AI to radically change how much and what kind of labor will be required going forward.

World After Capital: The Knowledge Loop

NOTE: I am continuing to post excerpts from my book World After Capital. The following is on the Knowledge Loop. Unfortunately I am dealing with a gitbook issue, so this revised text is not yet live on the book website.

The Knowledge Loop

Already today knowledge has made possible something extraordinary: by means of the innovations of the Industrial Age we can, in principle, meet everyone’s basic needs. But we cannot stop here. We need to generate additional knowledge to solve the problems we have introduced along the way, such as climate change. Knowledge is powerful, but only if we have enough of it. Where will that additional knowledge come from?

New knowledge does not spring forth in a vacuum. Instead it emerges from what I call the Knowledge Loop. In the Knowledge Loop, someone starts out by learning something, then uses that to create something new, which is then shared which in turn is the basis for more learning. And so on.

The Knowledge Loop is not new. Given my definition of knowledge, it has been around since humans first developed written language, some five thousand years ago. Before that humans were able to use spoken language, but as I have noted previously that puts tight limits on both time and space for learning and sharing. Since the invention of written language we have had breakthroughs that have helped accelerate and broaden access to the Knowledge Loop. Those include moveable type (about one thousand years ago), the printing press (about five hundred years ago), and then more recently the telegraph, radio and television. Now we are in the middle of another fundamental breakthrough: digital technologies, which can connect all of humanity to the Knowledge Loop at zero marginal cost and are allowing machines to participate in the Knowledge Loop.

It is easy to underestimate the potential of digital technologies for further accelerating and broadening access to the Knowledge Loop; to many, it seems as if these innovations have under-delivered. As a line on the Founders Fund website once complained, “We wanted flying cars and all we got was 140 characters.” Since that lament we have made great progress on flying cars in no small part because digital technologies, including the maligned Twitter, have already helped accelerate the Knowledge Loop.

Speech and Power

There is a healthy debate going on now about the role of Twitter, Facebook, and others (Google, Apple, …) with regard to moderating speech on their platforms. Rather than writing something entirely new, I decided to go back and look at what I have written and whether my opinion has changed since then. As I did that I was happy to find that I have had a fairly consistent approach based on who has power.

Over the years I have written a lot about keeping the government out of regulating content on the internet. For instance, in a post from 2010 titled “We Need an Internet Bill of Rights (And Fast)” I wrote:

If you care about freedom and democracy you do not want to give the government a wholesale way to shut down access to sites on the Internet.  The potential downside from abuses of such as system far outweigh the upside to copyright holders.

Much of my writing about various content bills, most recently about FOSTA, has been around limiting government power to interfere with speech on the internet. Government has a monopoly on hard power within a country (e.g. throwing you in jail) and so government intervention in speech is particularly problematic.

What about the platforms themselves? Here I have written for some time that they have a responsibility to society to moderate content. Here is a quote from a post that I wrote shortly after Trump was elected President, titled “We Must Talk About the Role of Facebook, Twitter in Society.”

Imagine a system that at the push of a button allows its users to deliver a physical object into the homes of hundreds of millions of other people. Now imagine the operator of that system saying: “we don’t ‘edit’ what people send via the system.” You could put in a hand grenade with the pin removed or a bouquet of flowers. “We just deliver it. We are only in the delivery business.”

We would find this position preposterous. As society we would have none of it. Yet we have exactly that when it comes to information. The system operators are making no distinction between uplifting content and propaganda. They allow for “mind grenades” to be lobbed into millions of homes.

So how to square the positions from the first and the second post? I don’t want government regulating speech on online platforms but I also want online platforms to take responsibility. The answer is take power away from the platforms so that competition is possible. Rather than antitrust, which is an industrial age tool, I would like to see the large platforms become programmable via APIs. Again I have written about this extensively, for instance most recently in a post titled “We Need Mandatory Enduser APIs for Social and Search Systems.”

[…] with an API key I can have an intermediary software layer that operates on my behalf. And that layer can connect me with friends and family that are split up across multiple social networks. This would allow for real competition to Facebook to arise. And once there is competition there is a strong check on behavior as a future #DeleteFacebook campaign would be far more impactful.

Even just the potential of competition (known as “contestability”) will allow users to exert real power over what kind of behavior is acceptable on a platform.

Of course much of the work on decentralized systems is meant to provide alternative platforms that do not have a central point of control. While I am supportive of this and have invested in a number of projects, such as Blockstack and Algorand, It would be a mistake though to think that these kind of systems will automatically lead to good outcomes for speech (or for anything else). Here is a talk I gave at Blockstack Summit making that point

In a follow up post I will write more about the problems with speech that we can already tell will arise with decentralized networks. While they do have the potential to guard against government power, they will definitely invite the kind of manipulation I flagged as problematic in the context of the election.

World After Capital: The Power of Knowledge

NOTE: Today’s excerpt from World After Capital dives deeper into human knowledge and why it is so powerful.

The Power of Knowledge

Have you watched television recently? Eaten food that had been stored in a refrigerator? Accessed the Internet? Played games on your smartphone? Driven in a car? These are all things that billions of people around the world have access to and often use daily (there are over 2 billion smartphone users). Many of us take these capabilities for granted and rarely do we ask where they come from. And while these are produced by different companies using a wide range of technologies, none of them would be possible without the existence of knowledge.

Knowledge, as I use the term, is the sum total of all information humanity has recorded in a medium and improved over time. There are two crucial parts to this definition. The first is “recorded in a medium” which allows information to be shared across time and space. For instance, stone tablets were some of our earliest ways of recording information. The second is “improved over time” which separates knowledge from mere information, provided that the process of critical inquiry is allowed to operate (we first encountered this process in the chapter on Humanism).

A conversation I had years ago but didn’t record cannot be knowledge. However, if I write down an insight from that conversation, or even the conversation verbatim, and publish it on my blog, I’ve potentially contributed to human knowledge. The conversation isn’t accessible to anyone who wasn’t there at the moment it happened. Even my own recollection of the conversation will fade. The blog post, by contrast, is available to others across space and time. Some blog posts will turn out to be important and become part of human knowledge. As another example, the DNA we carry in our cells isn’t knowledge by my definition, whereas a sequenced and recorded genome can be. Every person’s DNA sequence is ephemeral, i.e. disappears with our bodies. Recorded sequences though can be maintained over time, shared and analyzed. Ones that turns out to be medically relevant, such as the BRCA mutation that increases breast cancer risk, become part of human knowledge.

This definition of knowledge is intentionally broad and includes not just technical and scientific knowledge but also art, music, literature. But the definition is also narrow in that it excludes anything that is either ephemeral or not subject to improvement. Computers these days produce tons of recorded information, such as logs of activity on a system, that are mere information, unless they are subsequently analysed.

I started this section with examples of everyday technologies that would not exist without the power of knowledge. An even stronger illustration of its power is that without knowledge many of us would not be here today. As we saw in the chapter on population, Malthus was right about population growth but wrong about its most dire consequences because he did not foresee technological progress powered by knowledge. It is useful to go through one specific example to show just how powerful knowledge is and how it improves over time.

Humans breath air. But for the longest time we did not know what air consists of. Both oxygen and nitrogen, the two primary components of air, were not identified and isolated as elements until late in the 18th century (around 1770). Separately the systematic study of manure as a fertilizer, which had been used in agricultural practice dating back to Egyptian and Roman times, didn’t start until the early 19th century. That study led us to understand that ammonia, which consists of nitrogen and hydrogen, is a powerful fertilizer. Progress in chemistry and industrial processes eventually resulted in the so-called Haber process for nitrogen fixation, which means converting atmospheric nitrogen into a form that can be available to plants. The Haber process, which was invented in the early 20th century, became a crucial ingredient in raising agricultural yields globally and thus averting the Malthusian dystopia.

How successful has this been? For most humans today, about half the nitrogen in our bodies has been touched by the Haber process on its way into plants and animals that we subsequently ingest. Put differently: knowledge is so powerful that we are now made from knowledge.

What my much compressed history of nitrogen fixation doesn’t capture are the many false starts along the way. It seems hilarious to us now, but at one point a leading theory as to why some materials can burn had nothing to do with oxygen but was attributed to the material containing “phlogiston” which was thought to be the part of the material that “disappears” into the air when burning. Without the improvement of knowledge over time, we might have remained stuck at that theory.

When thinking about the power of knowledge, we must remember that a year, or a decade, or even a hundred years are all trivial in the time scale of humanity, and in turn, humanity’s time scale is trivial compared to that of the universe. When considering longer time frames, we should regard as possible all speculative propositions that don’t explicitly contravene the laws of physics—a line of thinking inspired by a new theoretical foundation for science called Constructor Theory [57].

Consider for a moment what knowledge might allow humanity to do in the future. We might, through further discovery, rid ourselves of fossil fuels, cure any disease, take care of every human’s basic needs, and travel to other planets in our solar system (organizations like SpaceX and NASA are already working toward this goal [55]). Eventually we might even travel to the stars. We could, of course, also blow our own planet to bits before any of that can happen or be struck by a massive asteroid (this is why allocating our collective attention properly is so crucial). Now, you might say: “Travel to the stars? That’s impossible.” Actually, it isn’t. Extremely difficult? Yes. Requiring technology that doesn’t yet exist? Yes. But impossible? No. Interstellar travel is definitely not imminent, but with the further accretion of knowledge, it will become possible.

Knowledge is the essential human project. We are the only species on planet earth that has created knowledge. This is also why I include art and music in my definition of knowledge. Art has allowed humans to express our hopes and fears, and its accretion into culture has helped motivate the large scale coordination and mobilization of human effort. We can broadly think of technical component of knowledge as underpinning the “how” of our lives and the artistic component the “why” And if you have ever doubted the power of the art portion of knowledge, just think of the many times throughout history and the present when dictators and authoritarian regimes have banned and destroyed works of art.

Browser, Wallet or Something New? Looking for Crypto Ease of Use

The internet was around for about two decades before the arrival of the web supercharged its growth. The ease of using a web browser made broad consumer adoption possible. Mobile payments were around for at least a decade before they started to take off. Again the key factor to adoption was ease of use – this time in the form of wallets that are actually integrated with in-app payments and widely accepted at the point of sale such as Google and Apple Pay (in other parts of the world adoption was driven by carrier billing or pre-payment – again questions of convenience).

Similarly crypto is approaching a decade of existence. And we are still looking for the massive increase in the ease of use that will enable broad adoption. It is possible that clinging to the browser and wallet metaphors may be holding us back. The browser is strongly associated with content. The wallet is strongly associated with payments. But crypto is so much more than both: it is a new infrastructure for decentralized systems.

Lots of smart people are working on browsers and wallets that are crypto enabled. And I am glad that’s happening, as the right answer may come from that work. But it would also be good to see experimentation with a new metaphor. Candidates that come to mind are: agent, proxy, and authenticator. A piece of software that I, the enduser, control and that represents me vis-a-vis the decentralized infrastructure.

What would be included in such a “representation” app? Authenticating myself and granting/revoking access to my data. Storing and managing my private keys (possibly in a highly abstracted manner, where I don’t even need to know what a key looks like). What does representation *not* have to include? The UI/UX of decentralized applications can be separate and handled by each app, including payment and investing applications. Some of the apps can be web based and others native (or the same app can have a web and a native experience). The closest thing we have to this representation concept today are password managers.

If you are aware of any projects that are explicitly pursuing this direction, I would love to hear about them.

Blogging Hiatus: Reading More

Despite making progress on the recovery from my shoulder surgery, I have not been blogging and today will be another missed Uncertainty Wednesday. 

I have been spending the last few weeks of summer reading more which has been enjoyable. I am learning about Zero Knowledge Proofs from various blogs, including Vitalik Buterin’s excellent posts on the topic. I am also nearly finished with Michael Pollan’s “How to Change your Mind” about psychedelic drugs. And I am halfway into Judea Pearl’s “The Book of Why” and about a third into Kurt Andersen’s “Fantasyland.” All three of these are interesting reads.

As for updates on World After Capital: I am stuck on a section that I am unhappy with but am hoping to get that resolved by next Monday, so I can continue posting excerpts. I have also received a large number of emails with comments on the book and am slowly getting back to people. My thanks to everyone who has written and apologies for the slow replies.

Wishing everyone a good summer and if you have topics you would like me to write about, please let me know.

Facebook’s Travails and the Decentralized Future

Facebook is in the unique position today to face both the problems of centralization and decentralization. On its centralized core platform it is confronted with making content decisions, while on WhatsApp it struggles with slowing down the spread of rumors and calls for violence. Just to be clear, I have no sympathy for Facebook which has been arrogant about these issues and has put growth above anything else. Nonetheless everyone who is building new decentralized platforms would do well to think about these issues NOW.

The Internet itself is quite decentralized relative to Facebook. That’s of course why InfoWars and many other conspiracy websites are out there. Facebook has long wanted to convince people around the world that it is effectively the Internet (after all time spent outside of Facebook is a lot harder to monetize). But of course it is not and it can easily censor content on its network. That is quite obvious based on its long suppression of even vaguely sexual content. So of course now it needs to rightly answer the question as to why it lets Holocaust denial be spread through its connections.

But Facebook also already has a foot in a semi-decentralized future due to its acquisition of WhatsApp which provides end-to-end encrypted messaging. Here content travels in a way that cannot be observed or censored. That is of course what many people building new decentralized systems are dreaming of. But that dream too has a nightmare version of it which played itself out in Sri Lanka and Myanmar among other places, where WhatsApp was used to spread rumors and calls for violence that amplified bloodshed. Here Facebook is reduced to trying to diminish the viral coefficient by limiting the number of forwards. 

In a fully decentralized system neither type of intervention will be possible. There will be no central censor (Facebook) and no obvious throttle for virality (WhatsApp). That is the explicit goal of building such systems and that is great to the extent that the censor is asserting market or political power to the detriment of competition or the voice of the people. But it would be a mistake not to anticipate that these systems will also be used to spread conspiracies and incite violence.

So what is to be done? Unfortunately there is no simple answer. There is not “just add some AI” quick technical fix. Instead we need a massive transformation. We need to shed the Industrial Age and enter the Knowledge Age. That requires changing pretty much everything about how we live and how our societies work (just like we changed pretty much everything when we went from the Agrarian Age to the Industrial Age). If you want to read my take on all of that, you can find it in my book World After Capital.

Uncertainty Wednesday: Updating (Conclusion)

The last few Uncertainty Wednesdays had us look at how to model beliefs using probability distributions and then update those with a specific example of using the beta distribution. You may have noticed something odd about the way we updated the parameters of the beta distribution: add 1 to α when we observe heads and add 1 to β when we observe tails. This wipes out any and all ordering information. So let’s say you have a total of 100 observations. With this update rule the only thing that matters is the total count of heads and tails respectively. Let’s say that happens to be exactly 50 each, which gives us this beautiful looking distribution:

Which is quite tight around the probability 0.5 for heads.

Yet clearly there is a huge difference between observing some fairly random sequence of heads and tails versus say first 50 heads and then 50 tails, or maybe 5 heads, followed by 5 tails, followed by 5 heads and so on. Say for the last one if we had observed this 20 times in a row, our belief for the next toss – expressed as a distribution – surely shouldn’t look like the picture above. Instead, we would probably put a very high probability on the next toss being heads (the last 5 were tails and so we are due for a reversal).

Why does our current approach not detect that at all? The reason is that embedded in our updating approach was the assumption that the coin tosses were independent of each other. If we wanted to allow for tosses to be influenced by prior tosses (or, more likely, by some underlying system that determines the tosses) we would have to use a more complex initial setup and updating procedure. In a future post I will show how to do this.

World After Capital: Limits of Capitalism (Self-Conservation)

NOTE: Today’s excerpt from World After Capital rounds out the section on limits of capitalism. We already saw the issue of missing prices, the problem of power laws and today talks about how the self-conservation of capitalism through the political system keeps attention trapped in the job loop.

Self-Conservation

Toward the end of the Agrarian Age, when land was scarce, the political elites came from land ownership. Their influence really wasn’t substantially diminished until after World War II. Now we are at the end of the scarcity of capital, but the political elites largely represent the interests of capital. In some countries, such as China, this is the case outright. Senior political leaders and their families own large parts of industry. In other countries, such as the United States, politicians are influenced by the owners of capital because of the constant need to fundraise.

A study conducted at Princeton analyzes how much public support for a policy influences the likelihood of that policy being enacted [51] in the United States. It turns out that for the bottom 90% of the population their preferences have no influence on outcomes. Only the preferences of the wealthiest 10% of the population matter. Even within the 10% whose preferences matter, there is a huge concentration. For instance, over a 5 year period the 200 most politically active companies alone spent nearly $6 Billion on lobbying.

Individual and corporate lobbying results in policies favorable to owners of capital, such as low capital gains tax rates (or in the case of venture capital and buyout funds the taxation of General Partner profits as capital gains instead of income). Low corporate tax rates with lots of loopholes, including the accumulation of corporate cash in low tax countries is also favorable to owners of capital. So in 2018 we are finding ourselves with some of the lowest corporate tax rates, the highest stock prices and the highest share of profits in national income.

In addition to preserving and creating benefits for owners of capital there are also outright attacks on the sharing and creation of knowledge. I have written more about these in the chapter on Informational Freedom, but want to give one example now. Corporations lobbied heavily over the years to lengthen copyright and strengthen copyright protections. Scientific publishers such as Elsevier have used these protections to make access to knowledge so expensive that even universities as wealthy as Harvard can no longer afford the subscriptions. [52]

The existing political and economic system thus acts to conserve the scarcity of capital past its expiration date. As long as that is the case we will not be able to solve the attention allocation problem outlined above. We will heavily over-allocate attention to the job loop (work and consumption) and under-allocate attention to the individual need for purpose and the collective growth of knowledge.

How then do we overcome these limitations? That is the subject of Parts Three and Four of World After Capital. But first we will take closer look at the power of knowledge and the promise of the digital knowledge loop.