30 秒でわかる
In a 2019 ARK Invest conversation, Elon Musk discusses Tesla production, data from driver interventions, autonomous-driving hardware and the potential uses of cryptocurrency. He offers ambitious production and autonomy forecasts while distinguishing a feature-complete system from one proven safe enough to operate without supervision. These are historical predictions; Tesla’s current FSD (Supervised) guidance requires an attentive driver ready to take over.
- 1ARK and Musk discuss ambitious electric-car growth estimatesThe interviewers outline a 2023 market forecast; Musk emphasizes that his own production estimates are guesses sensitive to timing.
- 2Feature completeness is separate from proven reliabilityMusk predicts a complete driving feature set in 2019 but says human oversight and regulatory approval are separate questions.
- 3Driver interventions provide examples for system developmentMusk describes using takeover events and rare road situations to improve Autopilot.
- 4Tesla was developing its own driving computerHe describes the relationship between computing capacity and the difficulty of the software problem.
- 5Tesla would consider sharing its platformMusk says another automaker could use Tesla’s hardware and software if integration did not overwhelm its engineers.
- 6Musk calls Bitcoin’s structure brilliant but questions its energy useThe discussion distinguishes his interest in cryptocurrency from Tesla’s business focus at the time.
ポイント
- Musk says small delays in an exponential production ramp can produce large differences in the number of cars delivered by a particular date.
- He distinguishes having a complete set of driving features from validating reliability and receiving regulatory approval.
- Tesla’s claimed data advantage came from driver interventions and unusual road situations, Musk says.
- His 2019 forecast that unsupervised driving could soon be safe is not a current operating instruction; Tesla requires supervision for FSD (Supervised).
- Musk praises Bitcoin’s design but raises energy use and says Tesla’s focus is sustainable energy.
チャプターごとに整えた全文を読む
Innovation and production forecasts
ARK’s interviewers discuss electric-vehicle growth and Musk explains his production estimates.
ARK Invest’s Cathie Wood and analyst Tasha Keeney meet Elon Musk in Fremont for the FYI podcast in 2019. Wood says public markets undervalue innovation and thanks Musk for keeping Tesla public.
Wood presents an ARK forecast that global electric-vehicle sales could rise from 1.3 million to 26 million by 2023 if battery and vehicle production capacity were available. Musk says the estimate sounds plausible, while acknowledging the timing could differ.
Wood also discusses ARK’s Tesla valuation range and a sales scenario. These are the interviewers’ historical investment estimates, not guarantees or current price targets.
Musk cautions that estimates for an exponential production ramp involve considerable guesswork. A relatively small change in timing can make a very large percentage difference to the total number of vehicles delivered by a particular date.
He cites the roughly six-month delay in reaching a peak of 5,000 cars per week. He argues that a calendar delay during a steep ramp can look like a much larger shortfall when expressed as a percentage of expected production.
Asked for rough forecasts, Musk guesses Tesla might produce about 1.5 million cars in 2021 and three million in 2023. He explicitly frames these numbers as guesses, sensitive to timing and execution.
Autonomy, data and hardware
They discuss feature completeness, regulatory approval, driver interventions and the custom driving computer.
The interviewers turn to autonomous driving and the potential value of software and taxi networks. Musk says he sees cars over the long term as vehicles for autonomy software as well as transportation.
Keeney asks what changed since Musk had said the final part of autonomy was extremely difficult. He says the last fraction remains hard, and distinguishes a feature-complete system from one whose reliability has been established.
In 2019 Musk predicts that Tesla will reach a feature-complete driving system within the year: a car could find its owner, pick them up and drive to a destination. He says that would not mean perfect operation or remove the need for observation.
He separates that forecast from two further questions: when regulators permit particular features with human oversight, and when they permit operation without oversight. He speculates that driving without attention might become safe toward the end of the following year, while acknowledging regulatory uncertainty.
Wood asks about China and other regulators. Musk describes his experiences with Chinese and U.S. federal regulators as positive and characterizes Europe as more conservative. These are his assessments from the time of the conversation.
Asked why he is confident, Musk says he works directly with Tesla’s engineering team and believes the technical problems can be solved. Keeney then asks how customers’ use of Autopilot contributes to development.
Musk says Tesla has many examples of drivers taking over and uses these events to identify situations requiring improvement. He describes rare road layouts, weather, construction and unexpected objects as part of the long tail of driving cases.
He says intervention data are uploaded without individual attribution and describes using examples of human driving to train the system. The discussion reflects his account of Tesla’s data practices at that time.
Wood contrasts Tesla’s approach with other autonomous-driving developers. Musk argues that a sophisticated neural network and visual perception are necessary to solve the problem, and criticizes reliance on lidar and hand-written rules.
Musk again emphasizes the variety of unusual events on roads. He credits both the growing fleet and Tesla’s engineering team for progress, while describing the company’s custom Autopilot computer as an important forthcoming change.
He explains the tradeoff between hardware capacity and software complexity: weaker hardware requires more careful management of available computation. He says the new computer can process camera images at greater resolution and frame rate.
Musk makes optimistic comparisons between future system safety and human drivers. Asked about regulators, he says large quantities of real-world mileage and intervention data would be more persuasive than claims about a fast computer alone.
Safety and rollout
Musk discusses regulators, highway features and more complex intersections.
The conversation returns to safety. Musk notes that drivers can become overconfident and says Autopilot repeatedly warns them to pay attention and keep their hands on the wheel. His broader safety comparisons are claims made in the interview, not a substitute for a controlled comparison.
Musk says that, with limited exceptions, regulators were not then the main constraint on rollout. He describes beginning with highways, where stop-and-go traffic is burdensome and high-speed crashes can be serious.
The next challenge, he says, is intersections, where signs and lights can be ambiguous. He describes working on U.S. roads before extending functionality to countries with different road conditions.
Platform sharing and cryptocurrency
The interview turns to other automakers, Bitcoin and the energy cost of crypto.
Asked whether other automakers could build on Tesla’s platform, Musk says Tesla would be open to sharing its charging network or Autopilot system, provided integration did not require extensive changes and consume its engineering capacity.
Wood changes the subject to cryptocurrency. Musk says Bitcoin’s structure is brilliant and mentions Ethereum, but questions whether cryptocurrency work would be a good use of Tesla’s resources given its focus on sustainable energy.
He notes the energy required to create cryptocurrency. Wood discusses cross-border payments and currency conversion; Musk says digital assets can transfer value without physical cash while also acknowledging tradeoffs.
The interview closes with thanks and a discussion of technological innovation. For current use, Tesla’s [FSD (Supervised) guidance](https://www.tesla.com/support/fsd) says drivers must remain attentive and ready to take over; the enabled features do not make the car autonomous.
