Elon Musk is CEO of Tesla and SpaceX, has plans to colonize Mars, and thinks AI may turn humans into its pets. But beyond the hype and his enormous net worth and Twitter presence, here's how Musk's companies are actually taking on ... virtually every industry.
Elon Musk thinks and acts on a larger, more cosmic scale than we’re accustomed to from entrepreneurs. Elon Musk has become a household name synonymous with the future.
Whether he’s working on electric vehicles (Tesla) or sending rockets into space (SpaceX), his larger-than-life reputation attracts its fair share of hero-worship. Musk can get a hundred breathless reporters to write about him and his companies with little more than a concept drawing and a tweet.
His main projects take on almost every major industry and global problem conceivable, and imagine a disruptive fundamental rewiring of that space or sector.
Whether he can deliver on his vast promises is often beside the point. And Musk himself is more than happy to feed into this hype machine.
We’ve decided to take a different kind of look into the Musk ecosystem.
Rather than assess Elon Musk and his companies on promises and hype, we wanted to look at the ways in which his companies are or are not transforming the industries in which they live — with numbers, hard evidence, and concrete demonstrations of disruption.
To do this, we took a deep dive into 8 different industries where Musk and his companies operate to understand how they have begun to change:
- Energy: Read on to learn about how, according to a utilities lobbying group, Musk’s efforts with Tesla and SolarCity could “lay waste to US power utilities and burn the utility business model.”
- Automotive: Musk wants Teslas to not just be affordable — he wants them to do something strange: make money for their owners. They’d do this through next-generation AI and self-driving technology. We investigate how he’s making it happen.
- Telecommunications: While few realize it, Musk’s work in space could revolutionize how we get online, and provide fast, affordable internet for the 4+ billion without access today.
- Transportation: We dig into how the Hyperloop, Musk’s proposed “fifth mode of transportation” that’s a “cross between a Concorde and an air hockey table,” plans to cut down the 6-hour trip from DC to New York to 30 minutes
- Infrastructure/Tunneling: We look at how Musk’s “Boring” Company is trying to cut costs in the notoriously expensive tunneling industry, where a mile of tunnel costs $1B to dig and each additional inch in diameter costs millions more.
- Aerospace/Airlines: Find out how SpaceX plans to build a “freeway” to Mars by reducing the cost of flying a space shuttle to a fraction of what it is today, not to mention harness rocket technology for earth travel as a well, so that a “spaceflight” trip from London to Hong Kong is similarly priced to a regular flight from London to Hong Kong.
- AI: We investigate why Musk, who is certain that the race for AI superiority is the “most likely cause” of WWIII, is investing so much of his time into building better AI.
- Healthcare: We dig into the high-bandwidth, minimally invasive brain machine interfaces that Neuralink is developing to create futuristic humans.
Read on for a deep dive into just how the money, invention, and ingenuity of Elon Musk and his companies are transforming these vital industries.
First with SolarCity and now with Tesla, eliminating our dependence on fossil fuels and instead drawing energy from the “giant fusion reactor in the sky” aka the sun has been one of Musk’s priorities for more than a decade.
SolarCity, his first attempt to make solar power mainstream and ubiquitous, was at the forefront of the early 2000s “solar gold rush.” In some ways it was a failure, but it remains important to understand its trajectory to understand how Musk and Tesla plan to take on the problem of renewable energy.
SolarCity grew to become the country’s largest provider of residential solar, then suffered some very public financial problems before being purchased by Musk’s other company, Tesla, for $2 billion.
That 2016 acquisition was controversial, with many observers calling it a “thinly veiled bailout.” And yet Tesla’s continuation of SolarCity’s work has helped make a stronger case for solar than SolarCity was ever able to make on its own.
Elon Musk originally suggested the concept for the company that became SolarCity to his cousins, Peter and Lyndon Rive, in 2004.
The concept for SolarCity emerged out of a simple realization: the clock was running low on fossil fuels. The need for a replacement was emerging fast. “If they started now,” as Men’s Journal reports Musk telling Lyndon in 2004, “They might rule the market.”
Evidence that other forms of energy production were vulnerable was abundant in 2004.
Coal production had been in a plateau since the late 1990s — early signs of a decline that has continued since. Electricity generation from nuclear also hit a plateau in the late 1990s. And while some predicted a “nuclear renaissance” in the early 2000s, as of 2004, that had not arrived either.
As of 2004, a majority of the generators of nuclear and coal-based power in the United States were also starting to reach end-of-life status. They would soon need either expensive upgrades or maintenance, or to be refashioned into generators for alternate sources of energy.
At the same time, as of 2004, solar was looking quite attractive as an alternative. Prices on energy generated through solar power had been dropping for decades, going from $76.67 a watt in 1977 to just a few dollars a watt in 2004.
The price of installing solar panels on your roof decreased as well — and has continued to do so in the ensuing years.
A SHIFT IN STOCK
Musk and SolarCity took on the last-mile challenge of making solar truly accessible and mainstream.
By 2013, it was the leading installer of solar systems in residential buildings in the United States.
Its key innovation, though, was less on the technology side and more on the accounting side. When SolarCity first came about, the cost for getting a solar roof installed was between $30,000 and $50,000 — upfront. SolarCity pioneered the “solar lease” strategy, which allows homeowners to get their roofs installed for free and pay back the installation costs over time. GTM Research reports that solar leases made up 72% of new solar installations as of 2014.
February of 2014 was SolarCity’s peak in terms of stock price. But cancellation rates on SolarCity contracts soon spiked to 45% or more, according to Fast Company.
Some critics pointed to SolarCity’s aggressive sales tactics as the culprit. SolarCity salespeople would book installations using savings promises that critics say “bent the truth” of the numbers. Customers, once they realized they wouldn’t be saving as much as they’d been promised, cancelled their installations in droves.
All the while, the SolarCity sales team was growing by hundreds of people a week. They were incentivized to book installations, and they were booking them in large numbers. Revenue, however, was not increasing at nearly the same rate.
Towards the end of 2015, they promised investors they would right the ship—by reducing their growth rate. Wall Street wearied. After SolarCity announced a particularly bad quarter in February 2016, their stock price dropped by a third.
“This is a company that I regard in a first-class crisis that acts as if everything is fine,” TV anchor Jim Cramer said afterwards. “You know I’m an aficionado of conference calls. You may have found the bottom. Yes, [this is] the worst conference call of 2016.”
TESLA BUYS SOLARCITY
In February, Musk proposed Tesla buy SolarCity. Tesla was at the time developing the technology to help people charge their Teslas at home and on the road. These so-called Powerwall batteries were being installed in homes and connected to solar generators by third parties. After the deal was approved, SolarCity’s business became organized under the Tesla “Solar Roof” product offering — allowing Tesla to provide end-to-end residential solar energy rather than just the battery.
With a one-story ranch house in California, it’s estimated that Solar Roof customers would save $41,800 over the course of thirty years. That doesn’t factor in state and local tax credits and other types of subsidies and incentives, or the potential property value increase from having a Solar Roof installed.
If people can install systems which make them virtually self-reliant when it comes to energy, in other words, what utility do the utilities have?
“Solar power and other distributed renewable energy technologies could lay waste to U.S. power utilities and burn the utility business model” – Grist Magazine
At a 2017 National Governors Association meeting in Rhode Island, Elon Musk announced that (with solar technology from Tesla subsidiary SolarCity and battery technology from Tesla Powerwall) a 100 square mile patch of land (probably in Nevada, or Texas) could provide enough power to supply the entire United States.
The first Solar Roof preorders took place in May 2017. They almost immediately sold out “well into 2018” and Tesla announced it would begin installations in the summer. In August, the first installations did take place — but at the homes of a few Tesla employees.
Tesla’s factory in Buffalo, “Gigafactory 2,” has had numerous production delays getting the Solar Roofs out to their preordering customers. Tesla brought Panasonic in to help make up some of the shortfall, which in December announced that it’s “getting ready” to start producing the cells needed for the Solar Roofs.
First non-employee installations are expected to begin in 2018, well behind schedule.
In Tesla’s newest showroom in NYC’s Meatpacking District, there are no Solar Roofs on display — nor any Tesla Model 3s. Fitting, because besides the Solar Roof, no Tesla project has encountered as many slowdowns and production problems as the Model 3, which we’ll dive into below.
The Model 3’s troubles are just the latest chapter in the roller-coaster ride that has been Tesla.
First started in 2003, Tesla was Musk’s second project post-PayPal, and still one of his most ambitious.
Tesla is a car company working to make the “car company” a thing of the past. The future that Tesla envisions isn’t just one where a gasoline-fueled or non-electric vehicle looks anachronistic. It’s one where a human driving a car at all is anachronistic. It’s one where the majority of people travel short distances by hailing Tesla cab rides and sleeping, talking, or playing as the car — through self-driving capabilities —takes them where they need to go. It’s one where people who own cars frictionlessly rent them out to serve as self-driving cabs while they’re not using them, then summon them back with the touch of a button when they do need them.
Production problems have plagued the California-based company, however, causing delivery delays and concerning many Tesla shareholders. The enormity of the hype around Tesla has made the company an attractive target for short sellers, though Tesla’s shorters were punished more in 2017 than any other company’s when they lost $3.7 billion betting against the company.
The resilient growth of Tesla’s stock price is built upon investor optimism. Many believe that Musk can deliver on his vision, or at least a hefty fraction of it, despite the “production hell” the Model 3 has experienced recently. Musk articulated his vision in a 2016 “Master Plan” post on the Tesla blog:
- Create a low volume car, which would necessarily be expensive
- Use that money to develop a medium volume car at a lower price
- Use that money to create an affordable, high volume car
- Create stunning solar roofs with seamlessly integrated battery storage
- Expand the electric vehicle product line to address all major segments
- Develop a self-driving capability that is 10X safer than manual via massive fleet learning
- Enable your car to make money for you when you aren’t using it
The plan began as promised, with the creation of an expensive, low volume sports car: the original Tesla Roadster.
Musk financed the Roadster’s creation with some of the money he made from starting PayPal. That first car was the first domino of the Master Plan, a “catalyst to accelerate the day of electric vehicles.”
Then came the Tesla Model S. It won 2013 “Car of the Year” awards from both Motor Trend and Automobile Magazine. In 2015, it won “Car of the Century” from Car & Driver. It went on to become the best-selling electric vehicle worldwide in both 2015 and 2016 (among models that plug in). But at about an initial $70,000 base, it still wasn’t the affordable “mass-market” car Musk wanted to build.
Betting on electric vehicles becoming mass-market always did make sense. Great Britain and France voted to ban diesel and gasoline auto sales starting in the year 2040. China has made it a point that one in five cars sold in the country should run on some alternative source of fuel by 2025. GM plans to have 20 electric vehicle models on the road by 2023. Volvo has decided to get rid of traditional fuel-powered cars entirely by 2019.
In this landscape, owning the electric vehicle market begins to look a lot more like one day owning the entire automobile industry.
And the economic incentives to use — if not own — an electric car are already obvious. Today, according to the Department of Labor, Americans pay something like $2,000 a year in gasoline and “motor oil expenses” alone. Freight companies pay as much as $200,000 a year to fuel up each of their gas-guzzling semi trucks.
Then there’s the AI component. In 2016, Tesla announced that it would outfit every single Tesla rolling off the factory floor with the constituent elements of a machine learning self-driving car program:
- Eight cameras
- Twelve ultrasonic sensors
- A forward-facing radar
- A computer
As people drive their Teslas around, these sensors work together to create a lifelike model of the surrounding environment. Those models are uploaded to Tesla, where they’re studied and compared with millions of hours of footage picked up from other Tesla vehicles.
The resulting “Autopilot” technology has already been rolled out to Tesla vehicles, though a driver can’t fall asleep while their Tesla drives for them — yet. Musk anticipates that functionality will be ready around 2019.
This self-driving functionality also includes the ability to control the car via smartphone, as in this example below. A user “summons” his Tesla to come pick him up under an overhang during a rainstorm in Miami — through his iPhone:
For now, however, Autopilot and summoning technology are only available in the Tesla Model S and Model X. It’s also supposed to be preinstalled in each Model 3— the locus of Tesla’s attempt to disrupt the automotive industry and simultaneously its biggest boondoggle.
Within 48 hours of being announced in March of 2016, the Model 3 — Tesla’s first true mass-market electric vehicle — had almost a quarter million pre-orders. That amounted to over $10 billion in potential sales. But production problems would plague the rollout of the killer app.
Musk promised 1,500 Model 3 units in the third quarter of 2017, up to 20,000 per month by December.
In reality, only 260 units were produced in the third quarter.
In November 2017, the date for hitting 5,000 Model 3 units a week in production was moved from December to March of 2018.
Meanwhile, Musk was on Twitter, announcing that Tesla would soon develop “intelligent windscreen wipers,” a “disco mode” for its interior lights, and a pickup truck.
Some analysts have advised Musk to “stop over-promising and under-delivering.” But while Tesla’s stock price hasn’t flourished amidst the Model 3’s production problems, the stock as a whole is still up about 50% on the year.
That’s because Tesla investors, by and large, trust Musk despite the challenges that lay ahead for the company. Even setting aside the Model 3’s problems, there is a chance that the Tesla machine learning program simply won’t be successful. There’s a chance the auto dealer lobby will be able to legislate Tesla out of business (Tesla bypasses traditional dealer networks that are supported by legislation in some areas), or that Tesla’s factories will never get up to snuff.
And there’s one which few talk or write about, but which is an existential threat on par with all of those, and which Musk is addressing head-on already — data.
Every Tesla car on the road receives software updates from and sends data about its environment using an AT&T LTE network. Each one sends and receives several gigabytes of data every month. As Gavin Sheridan points out, that’s a degree of reliance on another company that Musk is classically very uncomfortable with. Usually only the “full-stack” approach makes sense for his companies.
That’s a big part of the thinking behind what Musk is now calling “Starlink”— his plan to leverage his success with SpaceX into providing cheap, fast internet for every human being (and Tesla) on Earth.
For all the talk of how Musk and his companies innovate, the average Musk project actually seems to revolve around a different formula: find an old idea that failed because of lackluster technology, and attack it with some of the world’s best engineers.
That’s exactly how Musk and SpaceX are going after the satellite internet industry. The idea of beaming the internet down from satellites is an old one. Teledesic was founded in the early 1990s to build a constellation of satellites that could provide a wide network of broadband internet. It, and a few other companies founded to do the same thing, all failed and went bankrupt amidst the huge logistical challenge of getting so many satellites into space and maintaining low latency connections.
Elon Musk first talked publicly about his desire to blanket the world in satellite internet in early 2015. In November 2016, SpaceX filed an application with the FCC requesting to launch more than 11,000 broadband satellites over the course of six years and “provide robust broadband services on a full and continuous global basis.”
By the mid-2020s, this new satellite-driven internet service — “Starlink” — has the potential to become the world’s largest telecommunications provider on Earth. That’s potentially a $1 trillion prize. Not bad for what amounts to a SpaceX side project designed to help finance our eventual colonization of Mars.
A few months after they filed this application with the FCC, SpaceX flew — for the first time — a used rocket into space. It was a big step in the SpaceX “Master Plan,” part of which is to perfect the technology of rocket reusability such that a spacecraft can be landed and sent back into space within hours of releasing their payloads.
It is a technology which, when combined with SpaceX’s broadband aspirations, has the potential to massively disrupt the way telecommunications companies do business.
SpaceX has already brought the cost of a satellite launch down to $300~ million under what it costs to fly one with Boeing or Lockheed — about $85-95 million compared to $420 million. Their first reusable rocket launch, the Falcon 9, cost less than half of its original launch. There are still various pieces of the puzzle that SpaceX is working on making rockets fully reusable, a project which Musk projects will be done by late 2018. Then there’s only one aspect of launching that can’t be reused — the fuel — which costs about $250,000 per mission.
A unit cost of under a million dollars per mission would make it possible to launch 4,000+ internet satellites with ease. And those satellites, once in space, would blanket the entire Earth — including areas without internet currently — with persistent gigabit, low latency broadband.
There have been a number of prominent satellite internet company flame-outs in the last few decades — Iridium and Teledesic to name two. The Starlink project differs in some significant ways:
- Cost: As discussed above, SpaceX has brought (and is bringing) the cost of launching a satellite down to a fraction of what it once was
- Speed: Traditional satellite internet caps out at about 25 Mbps, while SpaceX’s would reach 1 Gigabit
- Latency: The amount of time it takes for a data packet to travel between Earth and a satellite — current providers clock latencies of 600+ milliseconds, while SpaceX is aiming at about 30ms
The first prototype of a SpaceX internet satellite is scheduled to go into space around the beginning of 2018. The first actual satellite should follow in 2019. As the company drives down the cost of launching, and more of its satellites get into space, the odds get higher and higher that SpaceX will, in the words of Gavin Sheridan, “bring about the death of land-based networks.”
That would be a huge coup for the very idea of satellite internet, which has been stagnant relative to its potential for decades. And it’s not the only “old idea” that Musk and his companies are working on restoring.
One of the oldest ideas in transportation, for example, is transportation by vacuum tube. In 1812, an Englishman named George Medhurst was the first to propose building tunnels underground and shooting passengers through them pneumatically, in pods.
In 2012, Elon Musk was one of the first to convince people that he might be able to bring that vision to reality.
Musk first started talking publicly about the Hyperloop in 2012, at a PandoDaily event in Santa Monica.
This “fifth mode of transport” (after cars, planes, trains and boats) would be a “cross between a Concorde and a railgun and an air hockey table.” Riders would travel in a low-pressure tube, inside pod-like capsules supported by air and powered by a “magnetic linear accelerator.”
In a whitepaper, he worked with the SpaceX and Tesla teams to test the idea’s feasibility and understand its economics. They found that “pod” would be able to travel a distance of 30 miles in just 2.5 minutes, cutting (for instance) the six-hour trip between LA and SF to just a half-hour. And it would only need to cost about $20 USD each way to sustain itself.
It would be cheaper than the high-speed rail California was planning to implement at the time.
Combine pressurized pods with a depressurized tunnel, and you get a form of transportation that’s much faster than any mode conceived before.
As far as speed, the Hyperloop would be the fastest (on average) mode of transportation in existence. Commercial airlines are the next fastest, traveling at an average speed of 575 mph. The Hyperloop would travel at about 600 mph, or about 3x as fast as the Shinkansen (bullet) train in Japan.
The Hyperloop could have a major impact on a few different industries. First of all, there’s the $660B airline industry. With the exception of travel over oceans, the Hyperloop could transport passengers faster and for less money than an airplane.
That speed would change a lot about where and how we live in America, which could dramatically change both residential and commercial real estate. One could easily work in Manhattan and yet live a six-hour drive away, in Burlington, Vermont — with a 30-minute Hyperloop commute back-and-forth each day.
It could change politics too, as politicians serving in DC could visit their constituents in their home states once a week, or once a day.
And it could revolutionize freight shipping. Almost half of all American import goods flow through the ports at Los Angeles and Long Beach. 14,000 truck drivers bring those goods to “warehouses and rail yards” all across Southern California, according to SCPR. They move about 11,000 containers a day and burn about 68 million gallons of fuel every year, according to PricewaterhouseCoopers.
While you would still need trucks and their human drivers for last-mile delivery, a Hyperloop-like system could transport goods an order of magnitude faster at much lower expense (with far less pollution).
Of course, the Hyperloop has its detractors. Not least of whom are those with an obvious protestation — where are we going to put it? Achieving the right-of-way necessary to build a train above-ground and the cost of construction has doomed high-speed rail projects for decades. And tunneling technology just isn’t there yet.
One day, when Musk was sitting in traffic outside Los Angeles, he tweeted out a complaint that became the impetus for the company that would attack this problem head-on.
Traffic is driving me nuts. Am going to build a tunnel boring machine and just start digging…
— Elon Musk (@elonmusk) December 17, 2016
And so started The Boring Company.
Compared to rocket ships and auto-piloted cars, The Boring Company is aptly named. Infrastructure isn’t cool. But it is a vital field, and one that the US is currently not a world leader in. When it comes to spending on construction projects, in 2012, the US only just outspent Greece as a proportion of GDP. It ranked #143 in construction spending, or 13% of GDP, one of the lowest globally. When you look at the largest infrastructure projects currently running, Asia and Europe are the big spenders.
US construction development has plateaued. With The Boring Company, Musk wants to not only dig a lot of tunnels, but also get better at digging tunnels, bringing infrastructure capabilities back to the US.
The Boring Company has three active projects. The first is the test tunnel at SpaceX in Hawthorne, California. This is a tunnel to nowhere, built just as an R&D site.
The problem with tunneling is cost. London’s current Crossrail project, tunneling directly under central London from east to west will add 26 miles to the London Underground and is projected to cost $23B. In the US, New York’s Second Avenue Subway consists of two 8.5 mile-long tracks and will cost $17 billion. The Los Angeles subway’s 2.5-mile extension cost nearly $2B.
The cost of tunneling is thus approximately $1B per mile. Musk considers that this needs to fall by an order of magnitude — to $100M per mile — for tunneling to be economically viable.
Reducing cost comes down to two things: size and speed.
The cost of a tunneling is proportional to the cross-sectional area of the tunnel bore. The wider the tunnel you want, the more you have to pay for it. The Second Avenue Subway tunnel is 23.5 feet wide. A one-lane road tunnel has to be 28 feet. The two-lane A-86 West tunnel in Paris, completed in 2011, is 38 feet wide.
The Boring Company intends to build tunnels of just 14 feet initially. This is half the diameter of the current required road tunnel, and leads to approximately one-fourth of the cross-sectional area.
|Tunnel||Diameter (feet)||~ Area (feet sq)||Area increase (%) vs The Boring Company|
|The Boring Company||14||154||–|
|2nd Ave. Subway||23.5||434||181%|
|1-lane road tunnel||28||616||300%|
|A-86 road tunnel||38||1134||636%|
Doubling the diameter quadruples the cross-sectional area and quadruples the cost. Reducing the diameter can save millions of dollars.
The Boring Company can drill smaller tunnels because of how they will be used, which becomes clear when you consider its second project: an inner-city tunnel in Los Angeles to relieve the city’s insane traffic congestion.
These will be tunnels for cars, but not for them to drive through. Instead, each car will be transported on an electric skate, catapulting it through the tunnel network at 125 mph. Musk believes that you could cut the travel time from Westwood in northern LA to LAX from 30-45 minutes to just six minutes.
The tunnels can then be small as these will be electric cars on electric skates. No internal combustion engines on site. If you look at the cross-section of the A-86 West tunnel, you can see why this makes a difference.
With all the fumes from the combustion engines, you need the majority of the space in the tunnel for ventilation. Additionally, you need to add extra space for larger vehicles, and make sure there is space for emergency vehicles to maneuver. By only allowing specific electric vehicles in, these problems are negated.
The other factor in the cost of tunneling is speed. Tunnel boring machines (TBMs), used to drill holes for tunnels, are excruciatingly slow. The Boring Company has a pet snail, Gary, who can currently outpace its machines, moving14X faster. “Victory is beating the snail,” says Musk.
And victory is achievable. Firstly, TBMs currently bore only 50% of the time. The other 50% of the time, they are adding concrete reinforcements to the walls. Optimizing the machines to do both simultaneously would immediately increase speed by a factor of 2. The company believes that TBM power can be increased without damaging the equipment, and power output could be tripled with the right power source and thermal management. Sprinkling in automation, and outpacing a snail looks likely.
The third project is more radical still — a tunnel all the way from Washington, D.C. to New York City. This is a journey that currently takes over four hours to drive, and almost three hours by the “high-speed” Acela Express train. Acela is supposed to be high-speed, but high-speed travel is restricted in high-density areas. Acela is only high speed in portions.
By tunneling, you can go high speed the whole way. Coupling a Boring Company tunnel with a Hyperloop train, Musk thinks this journey could be made in just 29 minutes. This turns the entire mid-Atlantic region into a massive metropolis.
For Musk, The Boring Company is little more than a hobby, taking just as “2-3 percent” of his time. He bought the TBMs second-hand, and staffs the company with interns. But that shouldn’t downplay the importance of The Boring Company to his other projects. The Boring Company is silently laying the foundation for a very exciting, and fast, future alongside three of Musk’s other endeavors.
The first is Tesla. The cost projections for the inner city tunnels are low because they will be exclusively for electric vehicles, shuttling thousands of Teslas around cities. This will alleviate traffic congestion on the surface streets, transferring traffic underground. When the first tunnels hit capacity, the company will just bore more, creating a 3D network of tunnels under each city, expanding to meet demand. Musk expects more traffic from autonomous, electric vehicles as driving costs plummet due to cost sharing.
The second is also obvious: Hyperloop. These tunnels will have to be larger, but with the advancements learned through the smaller tunneling projects, The Boring Company can increase efficiency in these tunnels as well.
The third is a bit less obvious: SpaceX.
Musk’s plan is to put 1 million people on Mars sometime in the not-so-distant future, and tunnels are central to this vision. The atmosphere on Mars is rough. People need some place to live, after all. Plus, all of the fuel you need to mine to run a twice-daily Earth-to-Mars space shuttle route is underground. If Musk is going to build a colony on Mars, building a network of tunnels is essential. Without expertise in carving out underground infrastructure on Earth, it’s unlikely we’ll see the vision of Elon’s most ambitious project completed.
On December 15, 2017, SpaceX CRS-13 launched from Cape Canaveral on a resupply mission to the International Space Station. This was the 13th resupply mission on SpaceX’s NASA contract, and the 45th launch of a Falcon 9 rocket to date.
With 18 flights in 2017 alone, even landing a rocket on a moving boat is now routine. But this mission was different. It was the first to exemplify the core feature of SpaceX and how it plans to get us to Mars — it was an entirely reused rocket. The Falcon 9 Full Thrust first stage had previously flown as part of CRS-11 in June. The Dragon capsule had first flown as part of CRS-6 in 2015. Falcons and Dragons had been reused before, but this was the first time an entire spacecraft had comprised previous flying components.
For Elon Musk, this is the only way space travel makes sense. If every rocket is a one-off, we’ll never leave the planet. If rockets become like airplanes, being reused time and time again for multiple trips, then space can become the next air travel — a way to span great distances, open to all.
It comes down to the cost-to-weight ratio. The cheaper it is to put tons of equipment into space, the more space seems a good place to go. If you have to build an entirely new spacecraft each time, however, costs stay sky-high.
At the top end of the price spectrum are the “expendable launch systems,” such as Arianespace’s Vega launcher and Boeing/Lockheed Martin Atlas V (manufactured by United Launch Alliance, ULA, a joint venture between thetwo companies). These are big rockets that can put a lot into orbit, but cannot be reused. The Space Shuttle (NASA) sits in the middle of the cost range. The shuttle was designed to be cheap and reusable, but the cost of the solid rocket boosters and main fuel tank that were expendable added to the cost and ultimately restricted the value of the program.
At the bottom of the spectrum sit SpaceX’s Falcon rockets, which have already shown a 3X-5X decrease in cost for getting a spacecraft into the sky. But it’s not where the economics need to be for Musk’s end-goal. He doesn’t just want to send a singe spacecraft to Mars.
Musk wants SpaceX to put 1,000,000 people on Mars. To do that, he says we need to “improve cost per ton by 5 million percent.”
From Musk’s perspective, leaving humanity as a single planet species is crazy, a surefire path to extinction. The further we explore and get away from Earth, the more anti-fragile we become and the less susceptible we are to superhuman AI or the destruction of Earth’s natural resources.
Mars isn’t exactly hospitable, but it is the best of the local options.
The Martian day is similar in length, the temperature range is roughly the same, and the amount of land is almost identical. There is water under the surface and an abundance of important elements in the land and air. The atmosphere is different, but that is why we’ll need The Boring Company and its tunnels.
There is another reason for tunneling, again related to cost. Getting to that “5 million percent” cost improvement requires not just reusable rockets. That is just the first of four components that are needed to get to Mars economically:
- Re-usability of all rocket technology. This is what SpaceX has been focused on so far. CRS-13 shows that this is already a reality.
- Refill the rockets in orbit. So much fuel will be needed that launching with all the fuel needed for a trip to Mars isn’t feasible.
- The ability to produce propellant on Mars. If we can’t even launch with the fuel to get there, it definitely isn’t cost effective to take the fuel to get back as well. The first thing the new colonists will need to do is build a gas station.
- The ability to produce the right propellant. All of this is predicated on being able to make the right fuel on Mars.
The SpaceX vehicles will use Methalox, a combination of methane and oxygen. To make the methane, SpaceX will collect CO2 from Mars atmosphere (96% of the atmosphere is CO2) and mine water from the surface. Through this the company can produce all the fuel its needs for the return trip.
The vehicle won’t be the Falcon/Dragon combination currently in use. Instead, SpaceX is developing the BFR — the Big Falcon Rocket. Whereas the Falcon 9 can take 22,900 kg to lower earth orbit (LEO), the BFR will be capable of taking 500,000 kgs to LEO. With the Raptor engines the company is currently building, the trip to Mars will take just 80 days.
But before we become Martians, we are still terrestrial-bound. And before BFR is built, the other Falcon rockets are still working to improve life on this plant. This is part of Musk’s overall strategy that you see throughout his companies: build something really helpful to use now that finances the crazy stuff of the future.
Switching to reusable rockets is allowing us to put stuff into LEO for considerably less than we have previously. It is even starting to open up space exploration to commercial realities. A great point of comparison for the commercial feasibility of SpaceX’s plans is air travel. If Boeing had to write off each 737 after just one flight, a trip from LA to Las Vegas would cost something like $500,000 per person. Because we don’t crash or trash every plane after it’s been used once, Boeing can charge just $43.
This is the kind of cost structure Musk wants to bring to space flight. It’s not going to cost $43 to get to Mars, but it is going to go from the infinity it currently is to $300k-$500k. Expensive, but doable.
When we start to lower the cost of orbital spacecraft, the company comes into the economic reality of not just inter-planetary travel, but intra-planetary travel. Boeing should be worried that it is providing the model for Musk’s business plan. As well as using BFR to get from Earth to Mars, Musk also sees a viable business in using BFR to get from Earth to Earth, just a lot quicker.
A spaceflight route, even sub-orbital, around the globe is going to be quicker than a regular flight. Musk contends that with such a flight trajectory, you can reach anywhere on earth in under an hour. The economics then follow that of commercial flight — originally something only open to the rich, as more people take advantage, the price will come down until a spaceflight trip from London to Hong Kong is similarly priced to a regular flight from London to Hong Kong.
That question of our long-term future is one that Musk takes seriously, and beyond space travel. Between climate change, nuclear war, and various other types of man-made disaster, few threats loom larger in Musk’s imagination as a problem for the long-term viability of the human race than artificial intelligence.
In September 2017, Musk announced that he believed AI (and the competition for superiority “at [the] national level” would be the “most likely” cause of World War III. He said he believed it was a bigger threat than North Korea. Of all the people to doomsay about AI, of course, Musk has strong credibility — his company OpenAI had just accomplished, one month prior, something no other AI company had ever done before.
7. Artificial Intelligence
In August 2017, during Valve’s Dota 2 tournament, a new top player emerged in the world of online gaming. Over the course of a week, this player beat a string of other top players, including world champions, in one of the toughest online games. And the player had only been playing for six months.
OpenAI first ever to defeat world's best players in competitive eSports. Vastly more complex than traditional board games like chess & Go.
— Elon Musk (@elonmusk) August 12, 2017
This tweet paints Musk and his “non-profit AI research company” OpenAI as following in the footsteps of Google with AlphaGo and Facebook’s DarkForest. But for Musk and OpenAI, this isn’t about playing games. This is about life and death and the future of humanity. As far as he sees it, if AI research continues down its current path, humanity has no future.
Artificial Intelligence is now a core component of tech. It is prevalent not only in the obvious places — Siri’s natural language processing, Google’s RankBrain— but in almost all tech sectors.
AI research is progressing at a significant rate, and it is this that Musk sees as an existential threat to humanity. Google, Facebook, Amazon, Apple, and all the companies in our AI 100(featured above) are each contributing to the upside of AI: higher efficiency, higher productivity, less work for humans, and, ideally, a higher quality of life for humans — freeing us to do the more fun, creative things.
But the race for these upsides is also a race towards a massive potential downside — a super-intelligent general artificial intelligence that is vastly smarter than humans and sees no use in keeping them around.
The purpose of OpenAI is to strengthen AI research. The above companies working on AI are naturally secretive. There is a commercial imperative, so, though you can read research papers from the DeepMind or Google Brain teams, or even Apple that has started its own machine learning journal, the work is behind closed doors.
OpenAI wants to not only perform research, but also “occupy the meta level, such as platforms and infrastructure that enable faster research for everyone.” To accomplish this, the company has two core components:
- Research: The foundation has attracted some of the best researchers in the field, promising them the opportunity to work on some of the biggest problems in AI. The group regularly publishes its own research into AI and machine learning. In addition, the team publishes broader ideas on its own site.
- Systems: The team is building platforms to help other AI researchers understand the machines it is building better. For example, the team has built an AI Gym, “a toolkit for developing and comparing reinforcement learning algorithms.”
The overall concept of OpenAI is to bring high-quality, Google/Apple/Facebook/Amazon-level AI research into the open with no commercial restraints. As the company says in its introductory blog post, “Since our research is free from financial obligations, we can better focus on a positive human impact.”
Is a super-intelligent AI a real problem? It sounds too sci-fi, even for Musk. Imagining colonies on Mars or self-driving cars is fairly easy. Imagining an AI-induced apocalypse isn’t. This is the criticism of Musk’s worries from others in the field. Machine learning leader Andrew Ng said, “worrying about superhuman AI is like worrying about overpopulation on Mars”.
But that is kind of Musk’s point. No one is thinking about this. Instead, they are all too focused on the commercial possibilities of AI. They can’t see the potential problems.
Those problems are two-fold:
- An AI will unintentionally do harmful things
- An AI will intentionally do harmful things
The first could be a problem even with current narrow AI. Say we build an AI cleaning bot. All this bot wants to do is make sure the world is as clean as can be. If the bot just wants to make sure everything is clean, it has a few options. The first option is to clean up all the mess. This is the outcome we want and that the AI developer is expecting.
But that isn’t the only option. Another possibility is that it will try and stop the mess occurring in the first place. Humans cause mess. “If there are no humans, there is no mess, so let’s get rid of all humans” increases the AI’s utility function and is a perfectly legitimate solution to the AI’s problem.
This AI safety research is the main focus of OpenAI. In 2016, the company co-authored a research paper into these issues titled Concrete Problems in AI Safety. The paper identified five areas of research that AI researchers need to strongly consider as they push forward with any type of AI:
- Avoid negative side effects. How can we make sure that they AI won’t be following its programming too exactly, so that it will do anything to perform its function? For the cleaning robot, this could be destroying the room in an effort to clean faster.
- Avoid reward hacking. If the AI uses a reward function to determine the right course of action, how can we make sure it doesn’t just try and maximize that reward function without performing the action? For the cleaning AI, this could be range from switching off its visual system so it can’t see the mess.
- Scalable oversight. How can we make sure than an AI can train safely even when training examples are infrequent? The cleaning robot would know that it has to clean up coffee cups, but how does it learn not to “clean up” the cellphone that’s been left overnight on the desk?
- Safe exploration. Can the AI explore possible outcomes and train without serious repercussions — say, learning how to mop the floor without trying to mop an electrical outlet?
- Robustness to distributional shift. As the data or environment changes, can the AI continue to perform optimally, or at least define its ambiguity and “fail gracefully”? Can the cleaning AI try to clean a factory floor if it learned to clean in an office?
There are already attacks to test the limits of AI. Robustness is a particular concern for narrow AI. How well do they work when you test them outside of their comfort zone. The answer is not well. Image recognition machine learning algorithms often misclassify adversarial examples — images that have specific noise injected into them.
This is a benign example. It’s not hard to imagine a malicious implementation of this kind of hack, however. Imagine an adversarial attack on the AI in your self-driving car that changes “stop sign” into “green light” in its programming. Not only would it be potentially more deadly than something like cutting the brake lines in someone’s car, it would be a virtual attack and therefore (hypothetically) highly scalable.
The core problem with AI safety comes down to one simple question: How can we make sure the AI wants what we want? OpenAI is trying to lead research in this field. Though it is dedicated to this field, it is not working in this alone. The concrete problems paper included researchers from Google Brain, Stanford, and UC Berkeley alongside OpenAI.
But with the non-concrete problems of a super-intelligent general artificial intelligence, OpenAI is on its own.
The core behind this worry is the learning rate for AI. The bot that won Dota2 is a prime example of this. From when it was switched on in April, it steadily increased its ability with each iteration.
What took humans years took an AI months. DeepMind’s recent success with its AlphaZero chess AI took this a step further. It learned how to beat the best chess computers in hours.
Starting from random play, and given no domain knowledge except the game rules, AlphaZero achieved within 24 hours a superhuman level of play in the games of chess and shogi (Japanese chess) as well as Go, and convincingly defeated a world-champion program in each case.
AI learns through reinforcement learning. The AI plays thousands of games, learning incrementally from each one. AlphaZero ran simultaneously on 5,000 tensor processing units, specially-built processing units designed to run machine learning algorithms using Google’s TensorFlow framework. The learnings from each are combined to produce “a superhuman level of play.”
These are still narrow AI implementations. But an artificial general intelligence, an AGI, could use these techniques to bootstrap itself.
AI is already learning to develop itself.
“A few months ago, we introduced our AutoML project, an approach that automates the design of machine learning models. … [we] found that AutoML can design small neural networks that perform on par with neural networks designed by human experts.” -Google Research Blog
An AGI could test millions of newer, better AGIs, picking the best parameters from each, combining them and immediately becoming smarter. That smarter AGI then starts the process anew. This is the law of accelerating returns. The future is approaching quicker. AI that learns quicker is being developed quicker.
Musk’s point is that we are the emperor at the chessboard. We won’t realize our mistake until it’s over. One minute we are the supreme beings on the planet. The next minute (literally the next minute) we are overtaken. Within seconds, the AI vastly transcends our abilities. There is no way back for humanity. We become ants to AI’s supremacy.
With OpenAI, the plan is to make the public sufficiently aware of the threat that AI could represent so that it will be regulated and controlled proactively. OpenAI isn’t, however, the only iron Elon has in this fire. He’s also investing in a hedge against the bet that humanity will save itself from AI in time, like an insurance policy on humanity’s future.
It’s called Neuralink — and the idea is to digitally augment humans before we get replaced.
Most of Musk’s endeavors exist on a big scale: spaceships to Mars, tunnels from DC to New York, electric car-producing factories all across the globe.
Neuralink is an utterly different beast. It’s about the microscopic rather than macroscopic, and the mental rather than the physical world. But because of that, it has to rank as the most challenging, and thus most exciting, of Musk’s current companies.
Neuralink was also unveiled without the fanfare of many of Musk’s other companies. The project was announced in an 1,100-word article in the Wall Street Journal in March of 2017.
“Building a mass-market electric vehicle and colonizing Mars aren’t ambitious enough for Elon Musk. The billionaire entrepreneur now wants to merge computers with human brains to help people keep up with machines.“
Neuralink is Musk’s project to build a brain-machine interface (BMI) that will link human brains directly to computers. BMIs have existed in research for decades. But even though human trials have started, two big problems still exist with current BMIs:
- The bandwidth of the systems is low. We have billions of neurons but BMIs only record a few neurons at any given time. This makes using them for any high-fidelity system difficult. You could move a cursor across a screen with your brain, but you couldn’t play the violin with your mind.
- The invasiveness of the interface is high. The implant requires neurosurgery and a constant, hardwired link into the brain. This means that it is restricted to people with a life-changing need, and currently to those for whom that life might not be long — as the hardwired link increases the chance of infection directly into the brain.
These are the two problems Neuralink is setting out to solve in the short-term. The company wants to build a high-bandwidth, minimally-invasive BMI that will be FDA approved so it can start to use in real-life patients within a few years, and everyone else soon after. Musk sees this as the only way the human race will survive given the ongoing encroachment of AI.
As Musk sees it, AI advancement is driven by capitalism. Companies like Amazon need to invest millions into developing its AI because if it doesn’t, Google and Microsoft and Facebook will, and so on. The question is not if this will lead to the creation of an artificial intelligence that can leave regular humans in the dust, the question is when. And there is no positive answer for humanity there.
“Even in the [most] benign scenario,” Musk says, “We would be pets.” The worst-case scenario would be the complete end of mankind.
One of Musk’s approaches to this problem is OpenAI — working to make sure we proactively regulate artificial intelligence before it has a chance to turn us into pets.
With Neuralink, he’s coming at AI from a different angle. The goal is to augment the human level of intelligence and preemptively mesh us with the digital world so we can build ourselves up before an AI can surpass us.
Between here and there, Neuralink has the potential to help people suffering from stroke, neurodegeneration, cancer, spinal cord injuries, amputations, and dozens of other healthcare issues. These conditions afflict millions of people every year and costs the healthcare industry millions to treat. And if the Neuralink project is successful, years of expensive treatment and therapy (and in many cases risky surgeries) could be replaced with a simple microscopic brain implant.
Almost a year after launch, the company’s website still only consists of a single page highlighting the roles that need to be filled at the company, including machinists, electrical engineers, and software engineers.
The eclectic team the company is trying to build gives a brief glimpse into this multidisciplinary effort needed to understand the brain and engineer a patch for it.
BMIs are brain implants, usually a chip of electrodes a few millimeters square, that are surgically implanted directly into the brain.
The electrodes pick up the electrical activity from brain cells, neurons, and transmit them to a computer. While the brain activity is being recorded, the animal (or human) performs a task such as moving a joystick to guide a cursor around on the screen. The scientists can then use algorithms to correlate the brain activity to the movement, teaching a computer that when certain neurons fire, the cursor should move left. Then you can turn the joystick off and move the cursor purely through the brain activity. Then you have a BMI.
The driving force behind BMIs in the past decade has been the military. As the use of improvised explosive devices (IEDs) became widespread in Afghanistan and Iraq, limb loss became more common among soldiers. Body armor improved, meaning soldiers were less likely to die in the blast, but extremities weren’t protected. From 2000 to 2015, approximately 1,600 soldiers had amputations.
Helping these soldiers was the goal of DARPA’s Revolutionizing Prosthetics program. Funding was given to research groups around the US with specialties in neuroscience, biomedical engineering, and robotics to develop new implants, new prosthetics, and new understandings of how to control the latter with the former.
Substantial progress was made, with human trials starting and patients capable of both controlling and sensing robotic arms:
The problems Neuralink need to solve include bandwidth and invasiveness. The bandwidth problem can be easily visualized through this graph:
There are about 85,000,000,000 neurons in the human brain. Up to 2013, the record for the most neurons recorded simultaneously from an animal brain was approximately 500. About 2,000 are possible over time from a single implant.
Only a fraction of all possible information is extracted by current BMIs. Millions of neurons are involved in the decision and movement when you move your arm to pick up a cup of coffee while reading this article. To allow an amputee with a prosthetic limb the same degree of control as they had with their original limb requires the ability to record from significantly more neurons at one time.
Once a human is hooked up to a BMI, a learning phase starts. The person learns how to control the robotic arm with the limited bandwidth. The algorithms learn which neurons are signal and which are noise and get better at processing the information. The two symbiotically adjust until the person incorporates their new “arm.”
The second problem has more variables. The brain is usually cocooned away from the world in a sheath of meninges and sterile fluid. It does not like invasion. Non-invasive BMIs exist, but they have even lower bandwidth as they can’t discern the individual neuronal activity needed for close robotic control.
The Neuralink team is looking for ways to minimize the invasiveness of its BMI while still having high bandwidth. Wireless is an obvious choice, but presents its own problems:
- How do you get power to the device? Wireless radios are power-hungry and processing and sending high-bandwidth information will also require significant power.
- How do you dissipate heat from the device? Chips, radios, and batteries all produce heat. The brain can only heat up by a degree of two before damage occurs.
Additionally, the electrodes themselves cause damage as they are inserted. The brain’s natural defenses literally encapsulate them over time, cutting them off from the rest of the brain and rendering them useless.
These are all the issues that BMI researchers have faced over the past two decades. But the team assembled by Musk at Neuralink includes people who have completely novel ideas to overcome these issues. DJ Seo, has developed “neural dust,” tiny silicon sensor nodes that could be spread throughout the cortex. Elsewhere, researchers are developing a “neural mesh” that can be injected into veins and travel up to the brain and record neural activity through blood vessel walls.
Musk himself calls these implants “neural lace” and imagines a mesh sitting over your cortex, acting as a digital layer above your animal limbic system and your human cortical system.
For now, the main beneficiaries of Neuralink could be the 300,000 people in the US living with spinal cord injuries, the 5.5 million Americans living with Alzheimer’s, and the 2.5 million with stroke or traumatic brain injuries. Each could be treated with an implant that restores motor, memory, or other cognitive functions. After that comes the day when you have a quick injection in your arm, and a few moments later you are a living cyborg.
Make it Better
Each of Elon Musk’s companies is formulated on an existential bet on our future:
- Tesla: Fossil fuel powered cars will soon be a relic of the past and electric vehicles will reign supreme — and alternative power will be cheap and accessible
- SpaceX: Being a multi-planetary civilization will be highly preferable to being a single-planet civilization — in case something very bad happens to Earth
- OpenAI: A super-intelligent AI would likely be the end of all life on Earth, and we might not even realize we’re building it until it’s too late — so it’s better that we prevent it now
These are some of the biggest bets that anyone can make, let alone an entrepreneur. That’s important to remember when you look at the various industries that Musk and his companies are disrupting.
These companies represent huge possible disruptions, some sized in the trillions of dollars, because their potential payoff is much more than winning a specific vertical or market — its the future of humanity itself.
And yet behind those high stakes and innovations is a relatively “boring” fundamental strategy: rather than invent something entirely new, take something old and make it better.
Across industries, Musk and his companies aren’t disrupting the state of play by inventing new things out of whole cloth — they’re taking ideas that failed, and bringing them back to life.