Elon Musk promised to drive self-driving cars that would eliminate traffic accidents, connect the human brain with machines, fill the sky with satellites so everyone can access the Internet, and colonize Mars. He doesn’t know exactly how to do these things, but he has enough money to turn every good idea into an effective business. For the mainstream, what the scientific community thinks is usually irrelevant, unless the truth is that AI can’t do the things Musk needs to do for Tesla and Neuralink to make good on their promises.
AI has a serious “mapping” problem that Tesla, Neuralink, Google, Amazon, Facebook, Microsoft, OpenAI, DeepMind and other players in the field cannot solve at present. And Elon’s money won’t help here either. When we talk about the mapping issue, we don’t mean Google Maps. The point is that the map cannot provide a one-to-one view of the area. Each “map” automatically suffers some data loss. On the ground, you can count every blade of grass, every pebble, and every puddle.
You only see a small representation of a huge reality on the map. Maps are useful for navigation, but if you are trying to count how many trees there are in a particular area, they are completely useless. When we train a deep learning system to “understand” something, we have to provide it with data. And when it comes to very complex tasks like driving a car or interpreting brain waves, it’s simply impossible to get all of the data. We’re just sketching a little rough map of the problem and hopefully we can adapt the algorithms to the task. This is the biggest problem of artificial intelligence.
Tesla can use millions, billions, or trillions of iterations to train its algorithms, giving its vehicles more driving experience than all people can muster, and still making unexplained mistakes. Some might argue that autopilot is safer than human driving without this improvement, but the fact remains that humans are safer drivers without autopilot than Tesla’s fully self-driving, manless jobs.
Producing the safest, fastest, and most efficient car in history is a commendable performance, but that doesn’t mean the company is about to solve the problems with self-driving cars or any artificial intelligence problems that have plagued the entire industry. Money cannot impose algorithms on a human level. The exact same problem applies to Neuralink, but on a much larger scale. Experts estimate that there are more than 100 billion neurons in the human brain. Neurologists remain skeptical about the idea that brain activity can be regional.
Recent studies show that different neurons light up in changing patterns, even when the brain accesses the same memories or thoughts more than once. In other words, if you draw a perfect map of what happens when one thinks of ice cream, the old map may not be usable for the next idea of ice cream. We can’t map the brain, which means we don’t have the ability to create a data set until we can train an AI to interpret it. If the money could be solved by self-driving cars, the human brain, or artificial general intelligence, they would have been solved a long time ago.
“Organizer. Pop culture aficionado. Avid zombie scholar. Travel expert. Freelance web guru.”