Elon Musk has contributed to some of the biggest technological advances, and there is more to come in the future. CleanTechnica talks about the gigs Musk has pulled off these last years, along with the hints he has dropped for what could be The Real-World AI.
Some of the clues can be found in three of the current projects Elon is working on:
- Self Driving Auto-Pilot
- Get to Mars or real-world AI
- Machine builds Machine
It is close to a year since Musk first tweeted: Humans to be 2 cameras (eyes) mounted on a slow-moving gimbal (neck).
He also mentioned how humans are often distracted. Tesla vehicles, in comparison, with radar, sonar, 8 cameras mounted on all sides and programmed never to be distracted, would definitely be superior.
Humans drive using 2 cameras on a slow gimbal & are often distracted. A Tesla with 8 cameras, radar, sonar & always being alert can definitely be superhuman.— Elon Musk (@elonmusk) April 8, 2020
Unfortunately, Artificial Intelligence systems still struggle to understand real-world surroundings. This makes it the biggest problem for AI to create its proper "responses." Musk hints that finding a solution for this could completely change autonomous driving and the rest of the world.
Tesla Creates A Unique Approach
Waymo, a competitor on autonomous driving, teaches their AI by driving it to a single road many times. As the car familiarizes with the street, it learns to do well on its own.
Unfortunately, it might not do well on other untrained streets. Andrej Karpathy, director of artificial intelligence and Autopilot Vision at Tesla, said that the Autopilot team could choose an area to improve on. For example, all streets with STOP signs are present. The data is collected and trained on the Tesla vehicles, giving them signals in advance where they might need to stop. Using this strategy, they teach the AI system new things.
By fully automating these data, two important events happen. Cars on the roads can collect new data, and cars coming down the road process these data. Tesla prioritizes reducing the time needed to train new vehicles. All these data would later be consolidated when Dojo, the supercomputer for AI-training, comes online.
Tesla is developing a NN training computer called Dojo to process truly vast amounts of video data. It’s a beast! Please consider joining our AI or computer/chip teams if this sounds interesting.— Elon Musk (@elonmusk) August 14, 2020
The Real-World AI
Musk released a lot of interesting tweets about concepts and new ideas. This month of March, an exciting conversation about AI in the current world came out, specifically about a real-world AI or a physical world AI.
Musk explained this as a system that goes beyond recognizing a single object or task but gaining awareness and understanding of the surrounding world. A real-world AI would automate the learning process and conclude with its own realistic solution for basic functionality.
The system might run in the same process that Tesla does with their cars. Unfortunately, collecting tons of data and processing it is not easy. What if we break down the process into singular independent tasks?
There are some amazing solutions for automating tasks with AI. Boston Dynamics, for example, has created one with four legs that autonomously understand where it can walk and how to overcome terrain obstacles.
Machine That Builds Machine
Tesla has worked for years aiming to create a highly automated factory where machines build machines. Musk focused that the solution is not in the hardware but the AI that will power the system. All the hints and the projects he has done are solutions to develop the most profitable and realistic way to create Real-World AI.
This article is owned by Tech Times
Written by Czarina Grace Del Valle