Kidney Disease: A Silent (and Costly) Killer

The underlying macrotrends, current solutions and investment opportunities

Creative Ventures recently led an investment into VenoStent, an advanced-material technology company working on eliminating the 50%+ failure rate in vascular-access surgery required to initiate hemodialysis. Here we unravel the underlying macrotrends driving the development of solutions and investment opportunities in this sector.

The kidneys are among those typically overlooked organs that are crucial to our survival. While all organs are essential, the kidneys should get a bit more love. They are responsible for our waste filtration, act as powerhouses for hormones, and deposit non-waste materials such as vitamins and amino acids back into the blood flow, among many other functions.

Kidney diseases are a leading but ‘silent’ cause of death

Kidney function declines as we age. Kidney diseases also emerge more in patients with existing chronic conditions like diabetes and high blood pressure. Those who prefer a clear definition of where things stand will appreciate the step-wise classification of Chronic Kidney Disease (CKD) stages below. 


Image: http://www.kbdialysis.com/what-is-kidney-disease

Approximately 37 million people – 15% of the US population – are estimated to be suffering from CKD. The rate increases to 38% in the 65+ age group. Kidney diseases are ranked as the eighth leading cause of death in the US. It is sometimes referred to as the “silent killer,” as almost all patients with mild/moderate symptoms are oblivious to their sickness. Awareness increases with severity to about half for those with severe CKD, at which point it is almost always too late. Early detection is paramount for saving lives. 

The rise of irreversible End-Stage CKD (ESRD/ESKD) 

Low awareness and action-oriented treatment leave most patients irreversibly progress through the CKD “steps.” Given the surge in other chronic conditions, an aging population, and the decreasing mortality rate in End-Stage Renal Disease (ESRD) patients, both the prevalence and incidence of ESRD are projected to rise.

Image: https://jasn.asnjournals.org/content/30/1/127

Arriving at the dead end: dialysis or transplant

As  patients approach ESRD, they are left with two options – kidney dialysis or kidney transplant. 

A transplant using a functional kidney is typically an ideal option; unfortunately, it’s not readily available and there are significant potential complications, including organ rejection. As of 2019, nearly 100,000 patients in the U.S. remain on the waiting list for kidney donation. Kidneys are the most needed organs on the organ-donation list, with the highest gap between ‘needed’ and ‘received’ figures. 

As a result, about 70% of ESRD patients rely on dialysis as an alternative. Hemodialysis, in particular, makes up 90% of the dialysis treatment, which is delivered in the course of dreadful four-hour sessions, three times a week, for as long as it takes until transplant becomes an option. The average wait is approximately three to five years. Only one third of the dialysis patients survive that long. 

Chart comparing dialysis vs transplants using percentage of patients alive over years after transplant or start of dialysis

Image: https://www.mainlinehealth.org/specialties/kidney-transplant-program/dialysis-vs-kidney-transplant

The opportunities

The already limited supply of human kidneys from living and deceased donors is expected to worsen given the apparent decline in living kidney donors and decreasing mortality rates. This has generated a renewed focus on opportunities for developing artificial kidneys as well as improving hemodialysis in the hope that more patients can survive until either a transplant becomes available or artificial kidneys emerge as a clinical reality. 

Efforts to develop 3D printed kidney are likely five to ten years from becoming clinically relevant. “The largest hurdle is that there are currently no techniques capable of mimicking the multiscale, hierarchical architecture and complexity of the native tissue/organ which is vital to function” (Wragg et al., 2019). 

In the meantime, there is great potential to dramatically improve hemodialysis. From an investment perspective, hemodialysis solutions are of great interest, given the recurring and critical nature of the problem. Creative Ventures recently led an investment into VenoStent, an advanced material tech company, that is working on a solution which could potentially eliminate the 50%+ failure rate in vascular access surgery required to initiate hemodialysis.

Follow the Money

Chronic healthcare management is expensive, and the costs will only continue to grow: as average life spans increase so too will the compounded effects of multiple chronic diseases. 

In the US, the cost burden associated with CKD is largely borne by Medicare, which accounts for about 80% of total ESRD expenditures. This is hardly a surprise, given that the Centers for Medicare & Medicaid Services (CMS) deliberately includes patients with ESRD under Medicare, in addition to those in the 65+ age group.

C:\Users\005861\Downloads\F1.large.jpg

Image: https://jasn.asnjournals.org/content/20/10/2093

It is a costly, though arguably necessary, move on the part of CMS. ESRD is disproportionately expensive. Under Medicare, average costs per patient were approximately $8K per year for a non-CKD control group, $46K per year for stage 4 CKD patients, and $87K per year for ESRD patients. After accounting for an additional  $90K in hemodialysis costs, ESRD patients cost, on average, 22 times more than those without CKD. ESRD patients make up about 1% of Medicare beneficiaries but account for 7.2% of Medicare-paid claims, $35.9 billion in 2019. The majority of this costs ($28 billion in 2019) goes to hemodialysis patients. 

Image: https://www.ajmc.com/view/all-cause-costs-increase-exponentially-with-increased-chronic-kidney-disease-stage-article

The burn is so direct that it fuels logical incentives for CMS to look for cost reduction. Initiatives – like this one launched in 2019 – specifically target ESRD patient groups, proposing that incremental improvement on dialysis may be reimbursed on an add-on basis. Such payment policies will encourage solutions that improve patients’ quality of life while decreasing CMS’ overall costs.

The cost burden on Medicare also leads to an interesting dynamic, one in which innovators benefit if they provide improved cost effectiveness in ESRD patients. According to CMS’ final rules (Medicare Coverage of Innovative Technology or MCIT – Jan 14, 2021), Breakthrough Devices are covered under Medicare for up to four years after the FDA approval date. 

Aligning incentives to encourage adoption

In Healthcare “just” improving quality of care is not always sufficient. Innovators need to convince their customers – the hospitals or providers in this case – why it’s beneficial (usually economically) to adopt their solutions. Hospitals are largely compensated by payers or insurance parties, so, without reimbursement codes, providers would have to bear the costs. Additionally, one needs to account for how well the solutions fit into the physicians’ and healthcare providers’ workflows. If they’re disruptive, they may never be put into use. Lastly, from the payers’ perspective, everything needs to add up. They aren’t likely to agree to pay for something extra unless their overall economics improve. This is the harsh reality: improving patients’ lives alone is hardly ever enough.

When it comes to ESRD, the CMS’ MCIT policy is exceptional in aligning incentives among payers, providers, and patients. For the payers – primarily Medicare in this case – solutions can address the inefficiency in the system, improve patients’ quality of life, and cut down on growing costs, are worth investing in. For providers with a reimbursement plan in place, implementing new innovations proven to improve patients’ lives is not only ethical but also generates much needed revenue for the low-margin hospital industry. Finally, his policy provides patients with increased access to life-improving solutions, and that makes it a win-win-win.  

Myths and Realities of Automation and Labor Shortages

This is the introduction to a series of articles Creative Ventures will be publishing on the topic of Labor Shortage and Automation. 

The pandemic has brought discussions of automation front and center once again. In our new contactless, socially distant world, automation has allowed operations to continue and has accelerated trends in industries where labor shortages have become a persistent problem. 

There is, of course, a case for recruiting more humans into these industries to address these problems. The US now has 10 million fewer jobs than before the pandemic. Why aren’t  businesses taking advantage of this abundant labor supply and hiring on the cheap? People need jobs, and it seems counter-intuitive that with such high unemployment, labor shortages would still persist.

In many cases, however, the specific workforces needed for the specific job do not exist. 

The airline, hospitality, and retail sectors were heavily impacted by the pandemic, and we’ve continued to see layoffs there. Meanwhile, a number of other sectors have been experiencing tightening labor markets as a result of secular demographic changes decades in the making. At an aggregate level, both the civil labor force of workers between age 25-54 and labor productivity have been stagnant since 2000. This suggests that there must be a labor shortage somewhere in the system. 

Many of the jobs experiencing this labor shortage fall into the category of “dirty jobs” – work that is often labor-intensive, repetitive, and – occasionally – unsafe. This includes jobs ranging from e-commerce packing and crop harvesting to assembly line work and even carpentry, to name but a few. The most acute labor shortage is in semi-skilled work that requires 18-24 months of training. Many of these positions are currently being filled by aging workers who are on the cusp of retirement. 

As a result, labor as a share of total cost has begun picking up, hurting low margin sectors such as construction, logistics, and food services. This shift will only continue to accelerate. It’s not only that older workers aging out and leaving these industries. At the same time millennial workers are shying away from these jobs, creating a vicious cycle.

The truth is that we are at an inflection point, spurred by the pandemic, where automation has become not simply a viable option for many industries but increasingly essential. Consider that the construction sector is facing 250,000 unfulfilled jobs vacancies, which represents more than $10 billion in market value. Demand in sectors like e-commerce grew 44% in 2020, but warehousing and fulfillment centers remain bottlenecks and the source of delivery delays. Of the 6.1 million food-service jobs in the US lost to the pandemic only have returned by July 2020. 

For industries like construction, automation offers an immediate solution to a persistent labor shortage problem. In warehousing and logistics, automation adoption has drastically accelerated. (Even as Amazon has hired an unprecedented 500,000 positions over the past year to fill demand, it’s easy to see that these jobs are not a sustainable long-term solution for the company). For food service companies and others in low-margin industries, the pandemic downturn in labor provides an opportunity to further deploy automation initiatives already underway.

Most importantly, in striving to maintain the balance of labor and automation, we have a generational opportunity to train the workforce we will need to support automation. Companies can invest in training for new positions rather than retraining for jobs that will likely be automated away within the next decade, if not sooner. 

This is the first in a series of articles Creative Ventures will be publishing on the topic of Labor Automation. Over the next few months, we will explore where venture investment opportunities lie within automation and how to evaluate these opportunities.

We will focus on industries that are most acutely susceptible to labor shortages and consider how disruptive technologies play a role in helping these industries meet rising global market demands. We’ll also look at how industries are retooling as a result of the pandemic, how this has accelerated automation, and what new markets might emerge as a result.  

Follow along as we dig into the data of automation acceleration, and identify what we consider to be unique opportunities for investment. 


Continue to the first post in our Labor Shortage and Automation series: The $400 Billion Market Part I: The Case for Construction Automation

What is Method Driven VC?

In order to truly address global problems facing society, many of which stem from resource shortages, we turn to technology, a resource multiplier. However, technology needs capital–for successful access, execution, and deployment at scale. 

Thus, the question becomes, how do we enable new and potentially impactful technology to reach the broader commercial market? Capital is both a crucial barrier and enabler at every point throughout the commercial lifecycle. It is through the lens of early-stage venture – from inception through scalable revenue – that we approach this question. By exercising the intelligent application of capital, we maintain that there is a higher probability of overall success as opposed to a random approach.

Venture investing is an environment with ambiguity, uncertainty, and rapid change. Decisions must be collapsed into binary outcomes with less than perfect information and a number of practical realities. How do you make the best decisions in an environment that is inherently noisy and uncertain? What steps can be taken to systematize this decision-making process so that it is repeatable, scalable, and can be measured and improved over time? 

At Creative Ventures, we’ve built our method-driven approach to address these questions.

In the rest of the post we’ll address the what, why, and how of method driven VC.

What is Method Driven VC?

At its core, method-driven investing is a system of investing that puts repeatable rules and processes in place to improve decision quality over time. These rules, and the process itself, are not set in stone. Far from it in fact, as the goal is to modify and update rules and processes as more information becomes available. 

Money management can be compared to a long game of poker. Though an imperfect analogy, especially in that we don’t think of our investments as ‘bets’, it’s illustrative from the standpoint of assessing decisions that involve both luck and skill. You can be on a long winning streak, but if it’s driven purely by luck, it’s more likely than not that by the end of the night you’ll end up going home with a much lighter wallet. Professional poker players curtail the unavoidable factor of randomness by following a certain philosophy: as long as you play your hand correctly, it’s okay if you lose it. Play with enough skill and small losses won’t matter. Long-term, strategy equals success. 

Ironically, in venture capital, thinking long-term increases the likelihood of more short-term returns. Instead of betting on the next unicorn that may drive profits over a decade down the road, method driven investment leads to much shorter turnaround times on profit. 

Why Method-Driven VC?

Successful investment has strong elements of both skill and luck. You can’t take luck or randomness completely out of the picture, but you can constrain the boundary conditions such that success is less contingent on being ‘lucky’ and more contingent on being ‘skilled’. Randomness will always be present, but there’s a way to shift the distribution of outcomes in our favor. Developing repeatable and scalable systems to maximize our skill results in ’curve shifting’ our distribution of probable outcomes as far to the right as we can, generating a higher overall probability of success. 

Some may ask “why not invest in the technology?” Our simple answer is that it just isn’t reliable. Every investor salivates over finding the next Facebook, Uber, or Twitter, but it’s incredibly unlikely. We find success in steady investments that consistently generate revenue, as opposed to one miracle company.

How to Successfully Implement Method Driven VC

That leads us to our most important question: How do you know if a decision you make is the ‘correct’ one? In venture, it’s not often immediately clear whether you are succeeding, as there’s a long feedback cycle (especially in early-stage startups). 

Method driven venture-capital investment embodies the same strategy as that of a seasoned poker player: it’s all about playing the correct hand for the right reasons. If a company we choose to invest in succeeds, but the reasoning behind our investment ends up being incorrect, we still consider that a failure of strategy. The opposite is also true–it’s not helpful to us if a company we declined to invest in does poorly, but not for the reasons we projected. 

The idea is to improve enough in skill and research to accurately predict a prospective company’s success and failures, potential challenges that the company will face, and even issues we anticipate a company will easily overcome.

It’s All About the Process

Overall, method-driven investment is all about accountability. We keep our strategy in writing, creating a record for ourselves that we can learn from and be held accountable for. For us, creating a method that’s not only accurate, but repeatable and scalable is where we view the true success. 

This is why we say at Creative Ventures that we’re not in the business of predicting the future, of placing bets. We don’t take on R&D risks with unknown horizons. We invest in scalable technologies primed to meet existing market demand. 

When investing, ask yourselves this: at the end of the day, are you riding on the coattails of one big hit that made your entire fund and your reputation? And if so, can you depend on that luck, or like most poker players, will you lose your hand?

Invest in “Product-Ready” Technologies, Not Moonshots

How to avoid being a sucker in deep tech

At Creative Ventures, we often hear from our peers, “You seem really fearless about technical risk.” Ironically, this could not be further from the truth.

It’s understandable why people feel that way. We have invested in everything from next-generation LiDAR, to synthetic biology platforms, to quantum computing. From that perspective, it really does look like we invest in things from some crazy science-fiction future. It probably doesn’t help that most of our team have been scientists and engineers of some sort. 

And to be fair, it’s easy to fall in love with exciting visions of technological marvels that are so advanced that they seem like magic.

However, we explicitly do not invest in technologies at Creative Ventures. We are a deep tech fund out of necessity because if existing technologies could tackle the massive and worsening secular trends we invest in (e.g. healthcare costs from chronic disease, global impacts of climate change), they would have already been adopted.

For a technology to be investable, we require two things:

1. A real, genuine market. Not a pile of maybe markets. Not a “look at how many applications we have.” Not “we could be used here.” A real market where the technology can be used and useful today.

2. Engineering, not R&D. Related to the “today” part of 1. Not something that could still use some marinating in a research lab — something that is ready to go today.

The first requirement is more self-explanatory. The second probably deserves more explanation.

Engineering, not R&D

Technical due diligence is one of the shortest parts of our investment process. Every time I’ve told someone that, it tends to shock them, but it’s true.

It’s either possible today, or it’s not.1

We don’t invest in things that may be useful. We don’t invest in things that with just this one breakthrough will become economically viable. We don’t invest in things proven at lab scale, but have no currently known way to go production scale.

Although we have peers who do invest in those kinds of companies, these technologies are all what we would classify “R&D”—as in, technologies that probably should be funded by grants or in research universities because the time it will take until they are commercially viable is entirely uncertain. It could be one year. Or it could be ten. Or it might not happen at all.

Now, given we invest in seed-stage startups, we obviously can’t expect our companies to already be ready to massively scale tomorrow. What we do expect is that the path to getting there is clear and bounded. Basically, you can—with a straight face—describe how many engineers and how long it’ll take to be “product-ready.” We call this “engineering” stage, not R&D.

Ten years from now, ten years from now

Gene therapy and AI are great examples of why we take this approach. In both cases, we see a lot of excitement today but those with longer memories will recall that this is not the first time for both. In fact, both of these technologies have been theorized since the 1960s-1980s and were in recognizable in “modern” form since the 1980s.

We have been able to insert genes into mammalian cells using viruses since 1984. Most of the base toolkit for modern deep learning (multi-layer networks, back propagation) was developed between 1982 and 1986.

In both cases, there were high hopes that faded by the 1990s.

In the case of gene therapy, there were still serious questions in targeting, immune response, and durability. These came to a head and brought the first gene therapy boom to a screeching halt from the death of 18-year-old Jesse Gelsinger.

The collapse of AI was less dramatic. By the late 1980s and 1990s, the massive limitations of expert networks and other “practical” AI technology became clear and basically killed off the market (deep learning was infeasible with the hardware of the time). This has been called an “AI winter”, and it wasn’t the first, just the first with more modern techniques. One could argue that, we’ve been disappointed by our expectations of what AI would be able to do in 10 years, every 10 years.

Lisp machine for AI from the 1980s. Michael L. Umbricht and Carl R. Friend (Retro-Computing Society of RI) CC BY-SA 3.0.
Lisp machine for AI from the 1980s. Michael L. Umbricht and Carl R. Friend (Retro-Computing Society of RI) CC BY-SA 3.0.

We don’t predict the future, just the present

We like to say that we don’t try to predict the future. We think that it’s sufficient to have a very strong grasp of the present.

It is epistemologically impossible to know how long it will take for some technologies to become product-ready. We take enough risk investing in early stage, deep technology startups; we don’t have to make things harder by needing to be right about when inherently random breakthroughs occur.

We have written and will continue to write posts about why technologies like AI now fit our two criteria and are actionable today. Meanwhile, the history of past booms and busts of frontier technologies will continue to function as a cautionary tale on how projecting forward can go wrong.

Footnotes

[1]: To be fair, our team does have an advantage given our advanced technical degrees and STEM backgrounds. We aren’t starting from zero; however, we apply that knowledge to understand what is and is not possible — not to imagine endless unrealized possibilities.

Credits

(Feature Photo by tableatnyCC BY 2.0)

The Fall of Data Moats

Data is the new oil: a commodity on its way out

“What is your data moat?” It’s one of the most common questions asked of any AI or data-driven startup.

It’s also an increasingly irrelevant question. Thanks to better representations and algorithms it’s now possible to do a lot more with a lot less data. Outside of a limited number of fairly specialized applications, data moats are becoming less and less secure.

It’s worth looking back at the history of big data not only to better understand how we got here but also to help identify those increasingly rare circumstances where data moats continue to create defensibility.

Defensibility Through Data

Data moats aren’t a new thing. In the internet age, they stretch all the way back to Amazon (the e-commerce bookstore) having ambitions of capturing general consumer purchasing behavior. Even before the internet became ubiquitous, IBM had vendor data lock-in and specific data on customer business requirements.

Defensibility through mass quantities of data really took off around 2010 as progress in machine learning/AI began to accelerate. Starting in 2015 breakthroughs from convolutional neural networks and other techniques increased opportunities for practical applications of AI, increasing the need for large data troves and data-network effects. 

From EFF AI metrics, https://www.eff.org/ai/metrics

The irony of this data land grab is that many of the same breakthroughs that enabled practical applications of AI/ML have introduced innovations that make massive data sets less and less valuable.

Before getting into that though, let’s discuss why mass quantities of data have been so treasured that investors once regularly funded companies that did nothing but collect data.

The Unreasonable Effectiveness of Data

In the not so recent past, when given the choice between a better model or more data, you’d have been smart to take more data.

On the one hand, you could carefully interview experts in a specific application area and find the best engineers to carefully craft the best algorithm. In contrast, you could simply dump a massive amount of data into an off-the-shelf algorithm—one that didn’t even fit the assumptions of the problem area—and very likely outperform the better model trained on less data. Even worse, when sophisticated and simplistic models were both trained on extremely massive data sets, they performed more or less equally.  

This was pretty much the finding in Michele Banko and Eric Brill seminal 2011 paper. Given the natural language task of figuring out what was the appropriate “disambiguated word” (e.g. principal vs principle), Banko and Brill  found that more complex models that took into account some aspect of grammar/structure really didn’t do substantially better than an extremely simplistic/dumb “memory” learner that just literally memorized the words before and after a word.

Chart of accuracy from Banko-Brill 2011.

The more complicated models eventually did better, but we didn’t actually see model choice making a huge difference. And, in fact, the really dumb model was still in the ballpark of accuracy even if it didn’t do as well when trained on  the largest data set. This is the core insight behind Peter Norvig’s widely quoted (and misquoted) paper “The Unreasonable Effectiveness of Data.” 1

So, why isn’t data still king?

It’s because innovations in representation and algorithms. The ImageNet performance chart embedded above. illustrates that, around 2015, we started to see the ability of machine-based image recognition reach human levels of accuracy.

Since then, of course, we’ve had things like AlphaGo beating the human world champion in Go. Or AlphaStar doing the same in StarCraft II.

We’re essentially seeing something that’s been predicted  since the 1980s: machine intelligence—in more and more general ways—is finally starting to match or exceed human performance.

It’s worth understanding why. Are the many advances in computer vision the result of data sets growing that much larger? Are AlphaGo and AlphaStar just mass-mining game data to beat the best players?

The answer is no, and no.

If we really sat around hoping that we’d get enough data to be able to recognize a cat from pixels, we’d need billions of examples with cats in all kinds of orientations, environments, and contexts. Similarly, if we wanted AlphaGo to use mined data to beat a human player, we’dneed to have an exponentially, impossibly larger data set exploring almost every iteration of the game board—which works in the more bounded chess, but not in Go (let alone StarCraft).

When it comes to image recognition, what we actually got were CNNs (convolutional neural networks), which “convolve” or “move around” images in a lot of ways, and feed them into a neural network that “compresses” data by throwing out overly specific data and retaining more generalizable data.

On the AlphaGo/AlphaStar side, they didn’t use “real” data at all. Both algorithms essentially played themselves over and over and over again, and used deep reinforcement learning to synthetically generate “data.” Simply put, Alpha Go and Alpha Star taught themselves how to play these games.

In neither case are the algorithms either hand-tuned like the laborious crafted algorithms I described above (which most closely resemble “expert networks” from the 1980s and 1990s) or reliant on ridiculously large datasets..

The advances we’ve seen are the result of CNNs (a combination of representation and models) and—in the case of AlphaGo/AlphaStar—advances in synthetic data.

From an investor and, well, just human perspective, it’s really exciting that we no longer have to hand-craft algorithms as much because if that were the case using AI/ML to solve real-world problems would never have worked at scale.  

The fact is that these seminal breakthroughs didn’t rely on more data. If anything, we got better at using less data. With CNNs we can now do with thousands of data points what once required millions.

And in the future, we’ll likely see similar order-of-magnitude shifts.

Bad news for data moats

If your business is built on the assumption that no one else can catch up to you after you’ve amassed a critical mass of data, you might be in a rough spot. .

It’s true that in the AI/ML business, the idea of building a  data moat can make a lot of sense. . Let’s assume:

  • It takes 10 million data points (say, something non-trivial, like hospital visits) to have a useful prediction from ML
  • Through giving away your not-so-great product for free (or even pay people to use it) and buying datasets and just manually creating data, you get there first
  • Now, you are just like any other software business, with near-zero marginal cost. Except in this case, people who want to catch up with you need to also need to do the same thing you did, but you can always undercut anyone who’s willing to go with a not-so-great product.

You can practically give away your product to prevent anyone from getting datasets. And as you keep accumulating data, you might have the ability to do even more, making competition even more lopsided.

You are better, cheaper, and continue to get more and more superior to any potential competition.

Given all this, it makes complete sense why VCs would be willing to fund the kind of business that will have a durable (and growing) monopoly.

Unfortunately, the reality is that representations, synthetic data, and broader machine-learning techniques are decreasing the value of data moats. Meaning you can spend a ton of money to get your dataset today… and a competitor in the near future will be able to match you with an order of magnitude less data. 

At some point someone  playing around with off-the-shelf libraries will be able to take some toy dataset and rival you.

So what still works? How should you proceed?

Option 1: You don’t need a data moat. Although this would seem to defeat the value of an AI company, there are plenty of other types of lock-in you can rely on: switching costs in a sales-heavy environment, for example.

Option 2: Accrue a dataset will always be extremely niche and somewhat painful to gather. If some researcher might be interested in getting grant funding and running a study with a hospital or something, you may be in for it. Some random part of the hotel/hospitality cleaning process? Well, that seems a bit more insulated from PhDs. Alternatively, bet on data that is specific and can only be collected by your device.

Any big interesting research problem will eventually have some high quality research dataset. A big, huge, general problem will have tons of startups grinding it out to gather  data. Some “boring” niche in the industrial universe? That’s safer. Something that requires your super-specialized equipment (or super-specialized equipment in general)? That’s a good deal safer as well. 

The lesson here is that applications mean a lot more than algorithms, which is something I’ll say more about in a future post.

The future is one where AIs can learn more like humans

A human infant doesn’t need 10,000 examples of snakes in various orientations, color patterns, and environments in order to figure out what a snake is—which is roughly what a good CNN might need to be able to handle somewhat arbitrary snake data. One, or maybe two, examples will generally suffice.

As a whole, we should expect researchers and engineers to continue to make AI/ML work more and more generally and behave more and more like humans.

Of course, “when” we will actually get to a true general human level is pretty uncertain. In the meantime, we should expect that accruing large quantities of data to become less and less of a barrier to replicating results. If you’re an entrepreneur looking to start a company, it’s worth keeping this in mind. You should be looking to develop an application that either doesn’t need a data moat or that relies on one no one will likely ever bother to challenge.

Footnotes

[1]: For nerds, this is a callback to the 1960s paper “The Unreasonable Effectiveness of Mathematics in Natural Sciences,” where the author essentially muses as to how mathematics had proven shockingly effective in physics (and other natural sciences) in predicting how things work—and where scientific intuition and math collided, math often won. In this case, it’s now data that provides the unreasonable and unintuitive effectiveness.