30초 요약
In a 2019 ARK Invest interview, Elon Musk predicted Tesla’s self-driving features would be complete that year and that passengers might safely sleep through a trip toward the end of 2020. He argued that driver interventions from Tesla’s fleet could supply examples of rare road situations, and that cameras and neural networks were a better path than reliance on lidar. He also acknowledged that Autopilot users could become overconfident and said the system told them to pay attention and keep their hands on the wheel. Musk later discussed cryptocurrency’s energy use and possible role in transferring value. Tesla’s current [Full Self-Driving (Supervised) guidance](https://www.tesla.com/support/fsd) requires active driver supervision and says the feature does not make the vehicle autonomous.
- 1Musk predicted unattended trips could be safe by late 2020In the 2019 interview, Musk distinguishes his estimate of technical safety from the separate question of when regulators might agree.
- 2Driver interventions can become examples for the systemMusk says Tesla collects information when drivers take over, allowing engineers to investigate unusual roads, weather and other edge cases.
- 3Musk chose camera perception over dependence on lidarHe argues that a neural network must understand visual scenes and that hand-coded rules and lidar alone would reach a limit.
- 4More computing power could make the software problem easierMusk expects Tesla’s new computer to process full-resolution camera images with less need to economize on computation. His safety estimates for the hardware are forecasts.
- 5Drivers can become overconfident in AutopilotMusk says the system repeatedly tells users to watch the road and keep their hands on the wheel; he identifies overconfidence as a problem even as he argues the system improves safety.
- 6Tesla started with highways and was working on intersectionsMusk says highway driving offered convenience and safety benefits, while traffic lights and unusual junctions presented difficult cases.
- 7Musk sees promise and energy costs in cryptocurrencyHe praises bitcoin’s structure and describes crypto as a way to transfer value, but questions its energy intensity and says selling bitcoin is not Tesla’s plan.
핵심 정리
- In 2019 Musk predicted that Tesla passengers could safely sleep during a trip toward the end of 2020; he separated that prediction from regulatory approval.
- Musk argued that collecting driver interventions could expose an automated system to rare and varied road conditions.
- He favored camera-based perception and neural networks over reliance on lidar, a sensor that measures distance with light.
- Musk said drivers were instructed to pay attention and keep their hands on the wheel when using Autopilot.
- He saw potential in cryptocurrency as a way to transfer value but questioned its energy use and said Tesla was not planning to sell bitcoin.
챕터별로 다듬은 전문 읽기
The timetable and data behind Tesla’s approach
Musk gives a 2019 forecast for self-driving and separates it from regulatory approval. He says driver takeovers can help Tesla study unusual conditions across its fleet.
In 2019 Elon Musk estimates that toward the end of the following year it might be safe enough for a passenger to fall asleep and wake up at the destination. He says he does not know when regulators would agree, and describes regulators in China and at the US federal level as generally thoughtful, while calling Europe more conservative. Tesla’s current [guidance for Full Self-Driving (Supervised)](https://www.tesla.com/support/fsd) requires active driver supervision and says it does not make the vehicle autonomous.
Asked why Tesla could solve self-driving, Musk points to his engineering and software background and the company’s technical team. He says he expects its full self-driving features to be complete that year, a separate milestone from his prediction about unattended travel.
The ARK interviewers ask how customers’ use of Autopilot contributes to development. Musk says Tesla has a large volume of data about driver interventions: when someone takes over, information is saved and uploaded so the team can examine what would be needed to address the situation.
The situations are varied, Musk says: road markings and rules differ by country, while rain, snow, smoke and other conditions create unusual cases. He says the company is getting better at learning from how people drive without always requiring human labeling.
Why Musk favors vision and more computing power
Musk argues that cameras and neural networks are central to automated driving. He expects Tesla’s new computer to make the software problem easier by processing more camera data.
The interviewers compare Tesla’s approach with those of Waymo and Cruise. Musk argues that a system based on hand-coded rules and lidar—a sensor that uses light to measure distance—will reach a limit. In his view, a neural network must solve visual perception, and cameras can provide what is needed.
Musk says the fleet encounters a long tail of rare events: road work may be marked incorrectly, a road may have an unexpected hole, or an unfamiliar object may appear. He attributes Tesla’s development progress in part to the volume of real-world examples and the growth of its fleet.
He also points to Tesla’s forthcoming in-car AI computer, which he says would be provided to customers who ordered full self-driving. Musk argues that better hardware and better software work together: limited computing power forces engineers to crop images and make other compromises.
Musk expects Tesla’s hardware to process all cameras at full resolution. He thinks full autonomy could in principle be pursued with the earlier hardware, but that the newer computer would make software development easier by reducing the need to budget every unit of computing power. His numerical comparisons of future safety are predictions.
Safety, regulators and the rollout
Musk discusses the need for safety data and the risk of driver overconfidence. Tesla began with highways and was developing intersection handling.
Asked about regulators, Musk says evidence such as accident and intervention rates over billions of miles would carry more weight than simply claiming the computer is fast. He argues that Autopilot is safer, while also saying some drivers become overconfident in the system.
Musk says Autopilot repeatedly tells drivers to pay attention to the road and keep their hands on the wheel, and that it warns drivers and slows down if hands are off the wheel too long. He says he would recommend the system, while identifying overconfidence as a recurring problem.
On regulation, Musk says that, aside from a few jurisdictions and minor delays, he does not see regulators as the principal obstacle at the time. The interviewers then ask how Tesla plans to extend the system beyond highways and across countries.
Musk says Tesla started with highways because assistance there can be valuable during long or congested drives, and because high-speed crashes can be serious. He says intersections are the next focus: recognizing a stop sign or traffic light in development is one thing, but determining the relevant light at a complex junction can be ambiguous.
Tesla would first try to make the intersection functionality work in the United States and then extend it elsewhere, Musk says. Different road layouts and edge cases require attention in other countries; he mentions Norway, the Netherlands, Germany and Switzerland.
Sharing technology and the cryptocurrency question
Musk says Tesla would be open to other automakers using its systems without major engineering changes. He also discusses cryptocurrency’s possible utility and energy costs.
Asked whether other carmakers could build on Tesla’s platform, Musk says the company would be open to sharing its charging network or Autopilot system. His condition is that adopting companies use the existing system without demanding many changes that would divert Tesla’s engineering resources.
The discussion moves to cryptocurrency. Musk recalls joking about bitcoin on Twitter and having his account temporarily suspended, which he attributes to an automated anti-scam rule. The interviewers ask whether a cryptocurrency could become native money for the internet.
Musk calls bitcoin’s structure clever and sees similarities in Ethereum and other projects. He questions whether cryptocurrency would be a good use of Tesla’s resources, however, and says creating additional bitcoin is computationally energy intensive.
An interviewer argues that bitcoin could help businesses transfer money between countries where currency conversion is costly. Musk says cryptocurrency can bypass currency controls and could transfer value more effectively than paper money, while having pros and cons. He adds that Tesla is not planning to start selling bitcoin.