Atoms and Bits

There is a framework I’ve/we’ve used over the years to think about where to invest and where not to invest that I call “atoms vs bits.”

I am not sure where I got it from but the concept is simple. Is the software being built and taken to market dealing with just bits or are atoms also involved?

The idea being that it is going to be easier to make something work if there are just bits involved. Atoms make things more complicated and more expensive.

In the 90s, when I first came across this framework, it led us/me to focus on areas like media and financial services where the product was end to end digital. And the first industries to be truly disrupted by the Internet were the ones, like media and financial services, that are end to end digital (or can be).

I’ve held on to this framework over the years and while we’ve veered from it from time to time, often unfortunately, it still holds up.

If you look at machine learning, possibly the most impactful technology right now (and I mean right now), you can see this at work.

Machine learning algorithms have massively transformed online advertising (just bits), online commerce (just bits on the UI), trading of financial assets (just bits), and our attention (just bits and neurons).

But in areas where atoms are involved, not so much. There appears to be a growing acknowledgement in the tech sector that the timeline to fully autonomous vehicles is going to be longer than some had thought. It is not that surprising. There are lots of atoms and lives involved.

I’ve been waiting patiently for the day that I don’t have to do the dishes after yet another amazing meal by The Gotham Gal. I will likely wait longer. Atoms are involved.

I am not saying that we should not work on these harder problems. We should. But we should also understand that the timelines will be longer and the road to adoption will be more challenging. That means these efforts will be more capital intensive and should ideally be investable at more attractive valuations. Sadly the latter has not been the case.

When you are investing other people’s money, you need to be mindful of where the timelines are shortest and the path easiest. And that has been bits for the totality of my investing career.

#VC & Technology

Comments (Archived):

  1. LIAD

    many understand this intellectually but you need to feel it viscerally for it to really hit home. – i grew up working in the family retail store/warehouse. endless hours shlapping boxes up and down ladders, moving stock, counting stock, collecting stock, delivering stock. learnt to hate physical work with a passion. – then in the early 2000s i wrote an ebook on how to dominate your opponents in online poker, it took off relatively well and sold a bunch of copies. but the real eye opener was that you could do something end-to-end in the digital world and get paid for it. changed my outlook on business and the path of my career.

    1. LE

      – i grew up working in the family retail store/warehouse. endless hours shlapping boxes up and down ladders, moving stock, counting stock, collecting stock, delivering stock. learnt to hate physical work with a passion.Well as always I have a theory on this based on my personal experience. I did similar work because my Dad had a wholesale company and there was a warehouse. However I also got do do office work and in particular typing invoices.So I will explain the difference as I saw it which became very helpful later in life.The warehouse work is boring but more importantly you can’t ‘be in a zone’ while doing it. The best example is taking inventory. Nothing is worse than counting inventory. You have to climb, look, write down etc. You can’t be ‘in a zone’. You have to use your brain but what you are doing is not challenging at all. There is little reward. And it seems to never end (then there is the dust and the dirt). And yes the physical labor part is not great at all. One thing though. When you are young you are not thinking about the health aspects of climbing the ladder. Would be interesting to see if the same job when older would be seen as a positive because climbing the ladder is exercise. (Interesting, huh?)On the other hand typing invoices, say, you can do for hours on end painlessly. It’s not that the work is very interesting it’s that you are in a zone when doing it and it requires zero physical exertion. So time flies. You are, for a lack of a better way to put it, on ‘auto pilot’. Another example of this is doing piece assembly work. I used to love doing that (was paid by the piece). You sit there and you also ‘get into a zone’. You setup your work with input and output and just run yourself like a machine. This also came in very handy later in life in my first business and even today I am able to setup work so that it’s done in the most efficient way possible. My experience working early on is what has helped with this.I am curious if your own history bares out what I am hypothesizing based on what I experienced when I was younger.

  2. jason wright

    Atoms vs bits is in Peter Thiel’s Zero To One.Bits do not exist without atoms in the web tech context (no cars, no Uber). They are eternally entwined.Bits have super modularity. Atoms not so much. Musk is grappling with this reality.p.s. Cuomo or Nixon?

  3. JimHirshfield

    Automatic dishwasher or paper plates.You’re welcome.

    1. Chimpwithcans

      I’ll add takeout – give the Gotham Gal a rest too.

    2. kidmercury

      Paper plates? Not if you want to live in style. Automatic dishwasher? Not if you want clean dishes

      1. JimHirshfield

        They make high end disposable plates nowadays…but not so environmentally friendly.

      2. JamesHRH

        If you spend enough $, the dishes are clean.

  4. William Mougayar

    That concept was first explained by Nicholas Negroponte in his 1995 book Being Digital where he described the movement from atoms to bits, eg books, CDs that became digital.Today, we can extrapolate further with crypto collectibles, eg cryptoKitties that are another form of conversion to digital goods.

    1. jason wright

      i’ve got that book.Stewart Brand’s The Media Lab: Inventing the Future at MIT as a reflection on how things have actually worked out thus far.

    2. awaldstein

      Most don’t know that these ideas were touched on earlier though in a much more obscure format in the Habitat Project.https://en.wikipedia.org/wi…The founders of this, started Electric Communities which toyed with many of the ideas now foundational to the blockchain world.Coolest group I ever worked with. So before the market was ready. Snow Crash was required reading and Neal S. was on the board.

    3. SubstrateUndertow

      Although abstracted “digital bits” created in a representational-object-space amplify/accelerate/scale everything by orders of magnitude the working value/stability created via said abstracted objects/processes completely depends on our ability to construct highly accessible social visualization tools for mapping all those abstracted “digital bits (objects/processes)” back on to the ATOMIC territory of our visceral lives.Without highly effective/accessible visualization mapping tools we will simply be pushing those “digital bit” abstractions too far beyond our visceral ATOMS and they will flip from amplification-tools into chaos-impediments. A key point made by Jame G. Miller in his ” Living systems theory” 1978

      1. Michael Elling

        All living systems need price signals acting as incentives and disincentives. We can extend that to ALL systems wrt to their generativity and sustainability. The internet lacks price signals clearing marginal supply and demand. It also lacks redistribution mechanisms sharing the geometric value captured at the core and top with the linear costs borne at the bottom and edge. Those working on distributed systems are going down the same path as those original internet steps.What conclusions would Miller have come to if he were aware of network effects; namely, it is becoming increasingly apparent that network theory is central to everything?

        1. SubstrateUndertow

          “network theory is central to everything?”Yes ideas synchronization-signalling(network-theory) is a fractal organizing function all the way to the metaphysical basement !!!Miller was very aware of network theory and even anticipated internet social structures as they are endemic to his unifying theory of visualizing everything complex as layered instantiations of reusable living system dynamics.He elucidates a vision of everything from cells to supranational systems as parallel/mappable networked living systems and attempt to map the cross level details in a some what tedious 1000 pages.Only a little dated given when it was written but fresh enough to bring your on refresh will a little effort.

  5. kenberger

    that’s exactly why wework, and scooter company valuations puzzle me.

  6. TeddyBeingTeddy

    Keep It Simple don’t be Stupid

  7. Bartosz Jakubowski

    There was also a great Uber video ad back in 2016 called Bits and Atoms: https://www.youtube.com/wat

  8. Chimpwithcans

    Is one line of investing more worthwhile than the other? I am reminded of the movie industry’s desire to play it safe with more predictable and less original investments (blockbuster sequels = bits) versus art that might push the boundaries (atoms?), all because of the economics. Is society poorer as a whole due to the relative ease of investing in bits over atoms?

  9. Richard Reisman

    Yes (as already commented) Negroponte’s Being Digital was probably the first well-known usage — it also described personalized news as “The Daily Me.” Related ideas go back at least to the early 1980s when IBM and others observed that if cars advanced at the pace of Moore’s Law (from the 1960s) we would “now” be paying 25 cents for a Rolls Royce that drove around the world on one gallon of gas.

  10. gbb6

    Bits (Media/Finance) vs Atoms (Healthcare/Education)? In education, content is transitioning to digital, but disrupting school/learning, as we’ve seen in the last decade, will require whole new ecosystems, not just a new stack/app/tablet. Has that played out for you in reflecting on your approach to investing vs education philanthropy? What about the education and health companies in the USV portfolio?

  11. Thierry Ascarez

    What about the meal to be machine learnedly prepared as well? Taking into account your day, your week, your efforts, your tastes, etc. Could be awesome no?

  12. Greg H

    When many of the world’s governments are (rightly or wrongly) looking to encourage private sector investments in technology as a way to make society better, this post makes it pretty clear that strategy will fall flat.As you state, VCs, and most who can invest meaningful amounts of cash, are stewards of other people’s money, and so have a duty to find straightforward investments which can yield returns on a predefined timeline. Atoms va bits makes sense in this context.Yet it’s the problems involving atoms which will most meaningfully make a positive impact on people’s lives. What model is there which can allocate substantial capital to these problems? Philanthropy maybe, but most charities are woefully inefficient when compared to VC-backed startups.

    1. Pointsandfigures

      the profit motive in capitalistic societies is a strong one. all you need.

      1. Tom Labus

        08 crash?

        1. Richard

          That wasn’t capitalism. The math of Tranches was akin to today’s so math of bitcoin. Fools math. Capitalism requires having skin in the game, ROA, ROE.

          1. Girish Mehta

            If Capitalism requires skin in the game, a lot of what happens in capitalistic societies today gets excluded ?

          2. Richard

            capitalism over time eliminates the free lunch of the skimmers

          3. Girish Mehta

            Isn’t the trend in the opposite direction – of decreasing skin in the game in the last few decades ?

          4. Richard

            Yep, it’s the price of handouts / universal income

          5. SubstrateUndertow

            I humbly disagree !The evolution of scarcity under AI / Automated / Network-Effect conditions bottoms out as a critical scarcity of non-debt driven “purchasing power” or “Effective Demand”. All other inputs are arguable elastic or are amenable to huge gains in scalable efficiencies.Karl Marx was disastrously naive about all his solutions but his primary contention that capitalism as presently formulated lacks the endemic mechanisms required to support production/consumption homeostasis was and still is right on the money!As AI driven automation accelerates “universal income” will, by the sheer causal necessity of it all, become a pivotal tool for all modern economies.The true challenge is how do we get there in an incremental fashion while dealing with the ball and chain that is our vestigial work-ethic based economic reward system ?

          6. Michael Elling

            Conceptually it is simple. Everything, and I mean everything, is driven by network theory. Knowing exactly how networks and internetworks behave in a multi-dimensional fashion is something few have come to understand, let alone master. The same principles can hold across all socio-political and economic institutions, living organisms and matter in the universe. There is no real difference between bits and atoms in the end.The answer is for the larger body/ecosystem/network/actor/component to provide a slightly larger settlement to the smaller actor, but in return get a bigger share of the resulting broader network effects.Developing these models won’t be simple, but we can start with some gigabit broadband and IoT sensor systems that achieve universally affordable solutions across infinite and infinitely divisible demand and supply that is obsoleting rapidly.

          7. SubstrateUndertow

            Great comment !I thing a key component for developing such models is contingent on the emergence of a widely adopted set of new narratives/metaphors and language memes that expressly maps onto the organics of network-synchronization-effects and the concept of “Non-Trivial-Causal-Spreed” that underpins the complexity of all living systems.Without such an “Organic-Process-Literacy” set of mass-culture memes/tools it will be near impossible to embody the distributive social collaboration(signaling) required to collective discover that oh so complex homeostatic coherence required to transition human social-economics into a functioning instantiation of the living-system organizing dynamic. This assumes those living-systems dynamics are reusable at the social organism level ?We are now fully entering into that stage of “Meta Evolution” where our “collective-introspective-cognitive-imagination” becomes simultaneously the evolutionary variant and selector.It is no longer about individual organisms being selected for fitness by and external environment. It is now mostly about the collective social organism creating new variances of its own self-generated socio-economic environments then circularly self selecting for the best fit new socio-economic variant. Best fit meaning optimal alignment with the underlying reusables endemic to all stable/coherent living system instantiations.If you can humour me here on the “Meta Evolution” and its circularly self selection thingy? Then here is some food for thought.If both the evolutionary variants and fitness-selection mechanisms are now largely driven by volitional cognitive self-interest as determined by a previous “Meta Evolutionary” stage then it follows that those socio-economic variants are no longer random but driven by the self-interest of the dominant political/economic players and on the flip side the selector for homeostatic-economic-fitness is no longer an objective selection by an impartial external environment but again driven by the self-interest of the dominant political/economic players. This would mean that there in no longer a functioning evolutionary feedback loop, no functioning mechanism to drive an evolutionary strange attractor?Which could imply that we have now reached our biology based evolutionary hull-speed? We were not built for collaborative self-interest/intelligence as the primary DNA directive.

          8. Michael Elling

            I think it starts with an understanding that Darwin’s work supports survival of the most adaptable as opposed to the strongest. The latter is capitalism’s mantra of winner takes all. In fact, it should become fairly apparent that the strongest and most intelligent are least likely to adapt and survive.Then we must view everything (all socio-political and economic institutions and all living organisms and universal constructs) through a network effects filter and gain a new understanding of how things came to be, as well as where things will lead if we stay the current course. Disaster.Lastly, if “digital” is disrupting everything and becoming the basis and driver for everything, we can build these settlement models into the root fabric of digital networks. These will be more sustainable than the current winner takes all platforms that have come to dominate today’s digital networks. They will then virally infect everything.

          9. SubstrateUndertow

            We are thinking along similar lines. I hope your optimism about the evolution of digital filter base social nervous systems is the winning conjecture.I of course was being a little more pessimistic about the limiting factor that may work against that evolutionary pathway 🙂

          10. Tom Labus

            That was leverage and keeping the worst CDOs on your balance sheets. Everyone was all in, subprime and a blind eye.

          11. Richard

            Sounds like Bitcoin and many other ponzi schemes

          12. ShanaC

            sort of. hedging instruments are as old as society itself (the first derivative comes from mesopotamia in order to hedge on long distance trade.) When they become increasingly abstract and based primarily on theory, you get problems

        2. Pointsandfigures

          that has nothing to do with innovation. if you look at it closely it has more to do with a government backstop (bad policy) and bad banking behavior responding to bad policy.

          1. Tom Labus

            We disagree here.

          2. SubstrateUndertow

            ORmaybe having a banking backstop is just good economic risk management that simply requires the weight of the taxpayer to be credible ?ANDthat taxpayer banking backstop should have government policies that block corporate banking malfeasance

    2. Girish Mehta

      Strong comment.

    3. cavepainting

      Great comment.Food is AtomsWater is AtomsPower is AtomsPharma is AtomsHealthcare is Atoms.Climate Change is AtomsThese things have the biggest impact on society and need long term investments. Bits of course help manage these atoms better, but the mot substantial impact will come from the innovations that deal directly with atoms.The only entity that can afford to invest on a longer time frame on problem sets with no immediate monetization opportunities are nation-states. State driven R&D has a huge role to play here (Think of where we would be without GPS as an example).

      1. Twain Twain

        People are atoms.Machines are bits.There are some Techno-Utopians who’d prefer it if the world was just the machine + its bits — perhaps with some bit implants in our brains to connect us to the Cloud with (that’s the Utopian version; the dystopian one is to control our minds and make us go insane like the neural lace in Iain Bank’s book).Humans are so complex (intellectually, emotionally, culturally) that Facebook’s decided to train us like Skinner’s rat-in-the-box to only click 6 emoji:* https://medium.com/s/story/…The irony being that now, apparently, Silicon Valley has no idea how to cope with President Trump’s anti-tech rhetoric.Venturebeat: “Now it is just emotion, mostly fury, that matters.”* https://venturebeat.com/201…Maybe this doesn’t matter: “For many Facebook posts, measurement of the emoji reactions is likely to be a bell-shaped curve that reveals no distinct audience sentiment.”* https://glean.info/can-face….Maybe brands don’t care that FB engagement is on the decline:* https://www.thedrum.com/new…None of it matters. It’s all just bits.

      2. Pascal Aschwanden

        You wanna talk about biggest impact? Why is no one talking about the Greatest need in the world: Housing! a multi-trillion dollar market that desperately needs innovation (in every sense of the word, not just how it’s physically produced/it’s component parts, but also it’s regulations and all the immense barriers to entry). As a category in desperate need of innovation, I don’t see how this gets so frequently overlooked.Here’s your potential impact, just look at the immense needs: Even in America, supposedly the richest country in the world, we spend over 1/3 our income providing shelter (over 1/2 in backwards states like CA, etc). Most people are just a few pay checks away from homelessness and eviction. the CA SPA adjusted poverty rate is 19% thanks to housing. almost 2% of SF is homeless, over 40K homeless in LA.Housing is Atoms, lots of atoms, and regulation too, of course.

    4. PhilipSugar

      I think people are missing the point:”That means these efforts will be more capital intensive and should ideally be investable at more attractive valuations. Sadly the latter has not been the case.”Doesn’t mean not investable. But it means when you look around and demand (and possibly get) a really high valuation because of valuation envy from somebody who is new in the investing market, the returns don’t justify it.Not that it’s bad. Not that you can’t make tons of money. But at the right valuation and the right timeframe.I have seen this: I want $10mm pre for my idea on a napkin. Why? That is what other people get!Well but your deer hunting trail camera might have a sprinkle of high tech “pixie dust” on it but you know there are lots of issues: Distribution, inventory, manufacturing, etc.Hey your scooter company seems pretty cool, but you need a “unicorn” valuation. Maybe not so much.A great example is Midway USA started in 1977. Nothing wrong with how big that business is, not wrong on how he moves atoms and is unbelievable at digital:”MidwayUSA received the Malcolm Baldrige National Quality Award in 2009,[4] and 2015,[5] and is one of only seven organizations to earn the award twice since the first award cycle in 1988. The 2015 Baldrige National Quality Award was a follow-up to MidwayUSA’s receipt of the Missouri Quality Award from the Excellence in Missouri Foundation in 2015, an award the organization also received in 2008.[6]”

      1. LE

        I bought one of those trail cameras and mounted it on a tree near my house to get evidence for the township that the property renter was using the house for his business. Next thing I knew a bit later the tree mysteriously died and I suspect that the guy running the business cut it down. He had a tree business and the issue was all the undesirables that were congregating at his house in the morning. Various neighbors would call the police. In true ‘under the radar’ fashion he knew exactly how to skirt the law. The township basically had no way to document what he was doing other than an occasional drive by. So I bought the trail camera to film and document the activity. The the tree died. Just like that.Anyway this is a pretty cool concept. I especially love the ‘not cell service not wifi’ part of the very effective video. This is leagues ahead of sell sheets being done for tech products (where by sell I mean any internet marketing) . Direct and to the point. (Just a spot check of Midway USA website things like this get me excited for some reason..) Very old school. It works.https://www.midwayusa.com/p

        1. PhilipSugar

          Look at his warehouse. Not old school. Now his commercials…..old school…He does half advertising, half showing you how to do something. At the end of the advertising he says: “I’m Larry Potterfield, thanks for your business” At the end of a demo he says: I’m Larry Potterfield and that’s the way it is”

          1. LE

            That guy is the real deal. ‘Potterfield’ is kind of a country like name for that matter. Never heard of him or Midway until you pointed out the site. That’s the type of thing I like to have in my brain (as I fancy and elevate myself on knowing all things about business as opposed to, say, sports or music).I like that he does these stories. I am compiling stories to maybe one day do a similar thing.https://www.midwayusa.com/l…Other than the fact that they sell guns, this is the type of business that Walmart or Jeff should buy. Sure Larry doesn’t want to sell. Obviously many have approached him and he says ‘not for sale’. So what you do is structure an option to buy the business and then you cozy up to the family members or work some other magic to get him to sell. Dicks would have been a buyer but they apparently have an issue with sales drop because of their views of guns after that knee jerk reaction to the school shootings. Big mistake.

      2. ShanaC

        there are tons of areas where atoms are involved without the pixie dust. Think healthcare. tailored drugs! tons of bits, less atoms.Basically, this is a scale

    5. Dasher

      Amen. And the world’s most valuable company deals with atoms and bits. Sticking to just bits is conventional VC wisdom, I know.

  13. Tom Labus

    Timelines are the most difficult element to control in any investing scenario. You can be right but still wrong in the markets. False starts, market indifference can all throw off a legit scenario.

  14. ErikSchwartz

    I don’t know where it came from but I first heard it at the Media Lab back in about 1990.

  15. PashaPS

    Interesting that you consider e-commerce as bits on UI. I know my business required a lot of atoms when it came to warehousing & logistics. Amazon’s clearly does too, and that’s one of their biggest (IMO) competitive advantages today. Am I missing something?

  16. awaldstein

    Reality is you are right.UnfortunatelyThat is why our food supply and environment are so screwed up and innovation so challenging to fund, implement and market.And why we need government to put rigors into things that are necessary but not always the easiest change for businesses to do the right thing with.

  17. Alan Gershenfeld

    Although it is early – the breaking down of barriers between bits and atoms could impact this filter over time. It is interesting to track the range of technologies that are enabling data to turn into things and vice-versa – which are accelerating in performance and reach – enabling some of the network effects of bits-centric platforms and services. This is the focus of my brother’s research at MIT in the aptly named Center for Bits and Atoms (http://cba.mit.edu/), and the global fab lab movement. The acceleration of these technologies (on a track similar to Moore’s Law) will also have significant societal implications (e.g. fab divides, fab access, literacy, risk mitigation etc.)

  18. Twain Twain

    Blockchain = bits. Matt Blaze is Professor of CS and Information Systems at University of Pennsylvania.https://uploads.disquscdn.c

    1. sigmaalgebra

      In physics we know that, first cut (setting aside black holes, “spooky action at a distance”, dark matter, dark energy, gravitational waves inside black holes, and more we don’t understand) energy is conserved. So, we can’t create or destroy energy but just move it around, e.g., from oil to gasoline to a fuel tank to a car engine to the wheels and exhaust to kinetic energy, etc.Well, in the US, mostly, except for the Fed, money is conserved — we can’t create it or destroy it but just move it around, e.g., from some people who have it to some other people who want and try to get it.Commonly when a share of stock on the NYSE is changes hands, one of the buyer or seller is wrong!More generally, money moves from the dumber to the smarter, maybe to a fool from a bigger fool!So, for crypto, just need to find some bigger fools with money!!!Not a new story!!:-)!!

    2. LE

      It’s like having a degree in binary search treesThis is an almost perfect example of how someone with a highly advanced brain (in one area) fails to see or understand the differences in the comparison that he is making.Just study computer science (or business or embezzlement or whatever).Note the juxtaposition of ’embezzlement’ next to ‘business’. Clearly he thinks his profession is of a higher calling and therefore fails to realize that w/o business (and more importantly what is created by people in business (which he enjoys every single day no doubt)) is quite essential to life as we know it and all the advances in society that we have made.And guess what? In the end he is just like everyone else when actually confronted with a decision that pits money vs. some stated principle. Take the fact that for years he steadfastly refused to sell crypto.com but finally sold the domain. No surprise. And make no mistake about it at all. In the end it was about money despite what appears in these and other stories and what he has said on his blog post about the subject. At least admit you sold because you got an offer you couldn’t refuse.https://www.theverge.com/20https://www.theverge.com/20…[1] Driven by jealousy after all the name school at Penn is not the one where he is an associate professor it’s the business school, Wharton followed by Medicine and Law.

      1. Twain Twain

        Yes, intellectual ivory tower snobbery versus on-the-ground in-the-money pragma has existed for a long time.My share wasn’t focusing on that.Just putting into context why investors get and like blockchain. It’s bits.

        1. LE

          Security researchers also think they are doing god’s work and somehow have a place in heaven. The truth is they are a huge part of the problem by spending time and resources discovering and effectively disclosing in great detail weaknesses that would never be known or exploited if not for their efforts. The cost to society of preventing these obscure discoveries dwarfs the cost if there was not public disclosure. All so they can get their little awards, speak at conferences and get pats on the back. Which is a necessary part of being a nerd; recognition by peers since nobody else honestly understands or cares what it is you are doing. This more or less dovetails what his tweet implies.

  19. sigmaalgebra

    I’m surprised at the timing, only current or recent, for the successes of machine learning (ML).The main reason for the surprise is “only current or recent” since as far as I can tell all or nearly all the successful ML applications have been using only applied math techniques that go way back, 100+ years, and were attempted back to IBM’s SSP (Scientific Subroutine Package), SPSS (Statistical Package for the Social Sciences), SAS (Statistical Analysis System), Matlab, Mathematica, R, etc., that is, as soon as the computing hardware was available. Back then, that applied math made a ripple in the pond but no big splashes, no “big box” of gold coins, no 100′ yachts, etc. Early in my career, I did a lot of consulting with that math — I never saw any yachts.As far as I can tell, the only new parts have been (A) using much more data (thousands of variables, billions of observations) and (B) more use of sigmoid curves (e.g., as in neural computing elements). But the use of sigmoid curves, and their potential to “fit” better than just simple arithmetic, goes way back in the subject of categorical data analysis and some uses for representing non-linear relationships, e.g., in some core parts of stochastic optimal control. I’m reluctant to believe that (A) and (B) represent a significant fraction of the practical ML successes so far.Net, I have to believe that nearly everything mathematical and at all useful about ML now has been feasible and just as reasonable to do going back decades.Maybe the old bottleneck and recent opportunity have been more use of computing: That is, no doubt it’s a lot easier for Google, Facebook, Netflix, and Amazon to do recommendations and ad targeting now that the users are connected via Web sites instead of just walk in retail stores with no Web browser digital user interface. I.e., the additional computing includes the Web for collecting data on the users and server farms for storing that data, including particular to individual users, and processing it with the old math.I do believe in the potential of some applied math for recommending movies, video clips, music, books, Web sites, clothes, …, in general for recommending to Internet users, one user at a time and particular to the user, Internet content and associated “ad targeting”, but I see much more potential in some applied math techniques, user interface techniques, and new data well beyond what apparently is now in ML.So, apparently so far the “big new thing” is that so many people are doing their shopping, getting their information and entertainment, socializing, etc. using Web browsers that permit readily collecting digital data particular to each user and their interests. That is, we had the math and basic software of ML decades ago but now have very wide use of Web browsers and server farms with lots of processors and lots of data on individual users.But as I mentioned to LE here recently, just because someone likes movie A is not much to go on, old math or new, for knowing how well they will like movies B, C, …, and products X, Y, …. In particular, IMHO Netflix and the Netflix Challenge world-wide competition to find more accurate means of recommending movies, along with huge swaths of Google, YouTube, Facebook, Amazon, etc. were from poor understanding of the basic principles, although from some relatively advanced math, of the applied math and, thus, fell into some easy traps of poor formulation of the practical problems and, then, inevitable from their initial poor understanding, some weak applied math techniques and much less good results than should be possible now.On the “bits” versus “atoms”, apparently the difference was a chain of data links and data processing that were, say, end to end digital without non-digital links such as in traditional retail shopping, movies in theaters, libraries based on books on paper, news media based on paper or just TV, etc. So, the difference is not so much “atoms” but just “end to end digital”.An opportunity recently has been and now is to do better exploitation of (A) the “end to end digital” and (B) data processing.For the processing, we don’t have to be stuck with the applied math and software going back to, say, SSP and SPSS! Gee, IBM did SSP (early in my career, I improved on that code, got one piece down from O(n^2) to O(n ln(n)) with no more storage and got rid of some nasty numerical problems in another piece) and recently has been pushing consulting for SPSS! Change their name from Big Blue to Old Blue???For “trading of financial assets”, I’m surprised: At one time I tried to get Morgan-Stanley, Goldman Sachs, Lind-Waldock, and some others interested in such things. The situation was like I was offering gallons of ice cold lemonade (maybe, in French terms, Romanée-Conti at 68 F or some such in Baccarat) with lots of ice, and they were ignoring it because it was not the water, or dirty water from a well, they were used to. Eventually I concluded that likely James Simons harvested nearly all that fruit. Later it appeared that trader Paul Tudor Jones II lamented on the effect of the computers. I’m guessing that Paul’s father was minister of a Presbyterian church just west of the campus of Rhodes College in Memphis!Okay; okay; okay: I know; I know: Morgan-Stanley, etc. wanted “proof”, i.e., a “pattern” they could clearly see in the rear view mirror!!!! Well, now, they can call up Simons and maybe get him to explain some of how it was done. Of course now it’s too late!!! That seems to be a fundamental situation — unable to evaluate projects early on and then waiting until too late! Thankfully for US national security, we didn’t wait for the Soviets/Russians to do the SR-71, GPS, Keyhole, …!Gee, it appears that on one computer, I have a questionable hard disk drive. Looks like I should do a full reformat of the drive and then restore the data from backup! Back to it!!

    1. Twain Twain

      Re “much more potential in some applied math techniques, user interface techniques, and new data well beyond what apparently is now in ML,” already invented this system.Investors don’t get it. They just want swipe L/R and a button click. They don’t care that this results in systemic data biases. It’s what they “pattern recognize”.So rinse and repeat.

      1. sigmaalgebra

        The difference between (A) poor data and math techniques and (B) better ones quickly becomes some very, Very, VERY big bucks.Indeed, what is commonly meant by “better” in the math and data, e.g., in the “optimization” and, in particular, in “statistical decision theory”, is plainly, bluntly, exactly, no compromise, really as best as possible from ANY means of processing the data, making more MONEY. No exaggeration. E.g., the stuff in optimization of an “objective function” and about minimizing expected “loss” is, in most applied contexts, including this one, about, couldn’t be more bluntly crass, MONEY. Apparently not many people working in, managing, or applying ML or writing equity checks really appreciates this point — and from their goals and values this is EXACTLY the point they should most appreciate.

    2. Richard

      Doubt Fred knows Big O

      1. sigmaalgebra

        O(n^2) means that asymptotically execution time grows proportional to n^2. So, O(n ln(n)) is, asymptotically, faster proportionally to the ratio n/(ln(n)). So, instead of digging my HP calculator out of my brief case, I’ll type in a few lines of Rexx code and illustrate that in practice the improvement can be from millions to billions times faster: n n/ln(n) —————— —————– 5 3 25 8 125 26 625 97 3,125 388 15,625 1,618 78,125 6,935 390,625 30,339 1,953,125 134,838 9,765,625 606,772 48,828,125 2,758,056 244,140,625 12,641,092 1,220,703,125 58,343,500 6,103,515,625 270,880,534 30,517,578,125 1,264,109,160 152,587,890,625 5,925,511,690 762,939,453,125 27,884,760,893 3,814,697,265,625 131,678,037,550 19,073,486,328,125 623,738,072,606 95,367,431,640,625 2,962,755,844,877Ah, to stick in the commas, I used some old code of mine!

        1. Lawrence Brass

          beautifully formatted!

          1. sigmaalgebra

            So, we get awarded “beauty points”? No doubt the first ever, likely the last, for me!! :-)!!!

          2. Lawrence Brass

            I could bet that your math publications in LaTeX are beautiful too.

          3. sigmaalgebra

            I hope the math is beautiful, but I know the typesetting is!! I use just TeX, so far never LaTeX, with Knuth’s macros Plain and about 100 of my own macros — dynamic storage allocation conforming to the nested execution dynamic descendancy (i.e., the run time call-return push down stack), table of contents, bibliographic references, cross references, e.g., “See theorem 5 starting on page 55 in section Measurable Selection” which automatically remains correct with paper revisions and page number and section title changes, annotation of figures, i.e., for an inserted figure, say, some line drawing diagram, put TeX formatted math symbols, etc. on the figure, etc.For anyone who writes math, TeX was and remains one of their favorite tools — amazing, unbelievable time saver.I have a friend, good mathematician from Courant in NYC, who got interested in computing but needed some help understanding what Knuth had in mind with TeX: The friend kept comparing TeX with other, more recent approaches to text formatting and graphics more generally and kept thinking that TeX didn’t compare well.Finally I explained:Just what Knuth tried to do with TeX is something you know very well: He was trying to get some good word processing for the pages see in the AMS journals. He was not trying for anything more or different. He was not looking forward for word whacking, graphics, information presentation, etc. but looking backward at JUST the AMS, etc. journal pages. So, no sound, no moving images, no check boxes, radio buttons, roll-overs, pull-downs, pop-ups, no spline curves.He kept suggesting that I write a translator from TeX to HTML. I kept explaining that that was impossible because TeX is an executing programming language, no doubt Turing equivalent, and HTML is neither. Or to “translate” TeX to HTML, just use the TeX program to format the TeX document, get the output file, device independent file, .DVI, use what TeX calls a “device driver” to convert the file to images, say, GIFs, and have the HTML pull in the GIFs. But that is not translating the programming language TeX but just working with the output of the TeX processor.

  20. jason wright

    perhaps what’s needed is atomic engineering to fuse the two together. over blockchain networks that my eventually be possible, sort of. it would equalise the velocity of each and make financial capital investing less weighted to one over the other in the rampant pursuit of returns. we might then see the built environment of a city like NYC evolve to serve the needs of all its residents and not the elite few (432 Park Ave. being an example).

    1. Michael Elling

      Is there any differentiation between marginal differences in supply and demand in the crypto world? Seems like the law of averages and majority hold wrt incentives, rewards, etc… Everybody wins and loses equally. Furthermore, are there any universal service or equilibrating mechanisms that mirror gravitational forces in the universe? All networks tend towards centralization (core and top). If there aren’t equilibrating forces that distribute value to the edge and bottom the system will collapse.

  21. Andrew Cashion

    aren’t NFT’s viewed as digital atoms?

    1. jason wright

      fixed bits? i wonder if binding value relationships could be established between fixed bits and atoms arranged in a complimentary form? so to access the former an investment needs to be made in the latter. to unlock the value of the bits forem requires an investment to be made in an atoms form.

      1. Andrew Cashion

        I don’t understand what you are saying.I have no problem saying that I don’t get some of the new business models here.

        1. jason wright

          neither do i. it’s just open ended contemplation of what a future bits and atoms fusion might look like. one has to start the thinking somewhere, kick it around, and see if anything comes of it. no other way.

  22. James Ferguson @kWIQly

    Take energy efficiency as a good confirmatory example- AI is very useful – but its about changing operations in the real world, furncaes lights chillers etc. Saving 30% on energy operating costs in the average office block is trivial – But energy is not core business,We conciously decided to bootstrap when we realised that though we had an absolutely massive potential market (determined by the rate of roll out of IoT devices, compliance legislation (eg smart meters) and politics (eg market deregulation.There is no way to catch the right bus – you simply need profitable sustainable technologies and happy customers and to be prepared to grow organically until the inevitable.If you get to the bus stop with a rucksack, sleeping bag, and enough food and water and you are first n the queue – you will get on the bus.Conclusion – It is not the slowness of atoms vs bits or lack of potential, but the timing. It seems this is a market VC does not serve except for fadish consumer hardware products (juice anyone).Exceptions AirBnB + Uber – In the real world but delivering pre-proven technology (beds , rides) via different channels.

  23. Sebastian Wain

    It depends of your thesis or strategy advantage: China does the reverse and invest in, or are good at, atoms but are terrible with bits. The Shenzen stories where small companies build their own tools and create new devices sounds like science fiction for people who live in bits.

  24. Amar

    I like this a lot 🙂 Good classifying frameworks are invaluable in moving decision making along (product manager bias anyone)I will add this to my list of handy to have frameworks:1. vitamins or pain killers (mint.com vs quickbooks)2. atoms or bits (see above)3. (is this a) new or better (headspace vs orange fitness)4. (saving) time or money (uber vs wow airlines) yes, yes i am aware that they are interchangeable5. visibile or invisible (bitcoin vs blockchain)What other frameworks do people use?

  25. Adam Parish

    I used to work with a super smart guy that always called software “bits”. I love that analogy and I need to start using it more.

  26. ShanaC

    At some point, bits become atoms. How much more can media and finance improve? Meanwhile, areas like healthcare, which can be majority bits (but not completely) have a huge amount of space to grow. Maybe we should think about amounts of atoms involved.

    1. Michael Elling

      There is a symbiosis that most in the bits world don’t appreciate beyond software/hardware tradeoffs. And based on the chromebook discussion the other day, many will never know the difference.But it is VERY important to understand the interdependencies. The entire internet runs on a physical layer that is complete separate from the control and application layers. That is wrong in a world of rapid technological change (obsolescence) and infinite and infinitely diverse demand.What’s needed are settlements (price signals that provide incentives and disincentives) that run north-south and east-west in the informational stack. These do not exist at scale today. Once they exist, then the atoms will influence the bits as much as the bits influence the atoms in EVERY vertical.

  27. David Todd

    The book Machine Platform Crowd talks at length about this. Great book.

  28. Hutchy

    Out of interest has anyone got any good references or cases where organisations have gone from atoms to bits to atoms?