Neuralink and Tesla have an artificial intelligence problem that Elon’s money can’t solve


Elon Musk’s problems are bigger and more important than yours. While most of us are consumed with our daily activities, Musk has been anointed by a higher power to save us all from ourselves.

He is here to make sure eliminate car accidents, make traffic a thing of the past, sort out autism (your words, not mine), connecting human brains to machines, filling the night sky with satellites so everyone can access the internet, and colonizing Mars.

He doesn’t know exactly excuse me we’re going to accomplish all of those things, but he has more than enough money to turn any and all good ideas he’s had into a working industry.

Who cares if Tesla is 10, 20, or 100 years away from solving the driverless car problem? Financial experts agree almost unanimously that $ TSLA is fine with its current progress.

What the scientific and machine learning communities think is often irrelevant to the mainstream when it comes to Musk. Opinions from across the field on self-driving cars are generally relegated to a muffled sentence in the penultimate paragraph of articles on Tesla’s efforts.

Usually it goes something like this: “some experts believe that these technologies may take longer to mature.

And who cares if Neuralink no longer lives up to expectations and is currently on the verge of “inventing” a brain-computer interface that is exactly as sophisticated as the one Eberhard Fetz built in 1969??

People who have the opportunity to invest in Neuralink will earn money as long as Elon keeps up. Never mind that the distant technology that claims it will one day take the common BCI your company is making today and turn it into a magic telepathy machine is purely hypothetical in 2021.

The reality is that AI can’t do the things Musk needs to make Tesla and Neuralink deliver on their promises.

This is why:

  1. AI has a serious “mapping” problem that Tesla, Neuralink, Google, Amazon, Facebook, Microsoft, OpenAI, DeepMind, and the rest of the players in the field currently have no idea how to solve.
  2. Elon’s money is useless here.

The AI ​​”mapping” problem

When we talk about a mapping problem We are not referring to Google Maps. We refer to the idea that maps themselves cannot be made possible by one-to-one representations of a given area.

Any “map” automatically suffers a great loss of data. In “real” territory, you can count every blade of grass, every pebble, and every puddle of mud. On a map, you only see a small representation of the vast reality. Maps are useful for getting directions, but if you’re trying to count the number of trees on your property or determine exactly how many wolverines are hiding in a nearby bush, they are pretty useless.

When we train a deep learning system to “understand” something, we have to feed it with data. And when it comes to hugely complex tasks like driving a car or interpreting brain waves, it is simply impossible to have all data. We simply draw a small-scale approximation of the problem and hope that we can scale the algorithms to the task.

This is the biggest problem with AI. That’s why Tesla can use the Dojo to train its algorithms in millions, billions, or trillions of iterations, giving its vehicles more driving experience than any human ever combined, and yet it still makes mistakes. inexplicable..

We can all point to the stats and shout, “Autopilot is safer than unaugmented human driving!” Much like Musk, but the fact remains that humans are much safer drivers without autopilot than Tesla’s full autonomous driving capabilities without a human.

Building the safest, fastest and most efficient production car ever is an incredible feat for which Musk and Tesla should be commended. But that doesn’t mean the company is close to solving driverless cars or any of the artificial intelligence problems plaguing the entire industry.

No amount of money will go to human-level brute force algorithms, and Elon Musk may be the only artificial intelligence “expert” who still believes that deep learning-based computer vision alone is the key to vehicles. autonomous.

And the exact same problem applies to Neuralink, but on a much larger scale.

Experts believe that there are more than 100 billion neurons in the human brain. Despite what Elon Musk tweeted recently, we don’t even have a basic map of those neurons.

In fact, neuroscientists are still challenging the idea of ​​regionalized brain activity. Recent studies indicate that different neurons light up in changing patterns even when brains access the same memories or thoughts more than once. In other words: if you draw a perfect map of what happens when a person thinks of ice cream, the next time they think of ice cream, the old map could be completely useless.

We don’t know how to map the brain, which means we have no way of building a data set to train the AI ​​to interpret it.

So how is an AI trained to model brain activity? You fake it You teach a monkey to push a button to summon food, and then you teach it how to use a brain computer interface to push the button, like Fetz did in 1969.

Then you teach an AI to interpret all the monkey’s brain activity in such a way that it can tell if the monkey was trying to press the button or not.

Note that the AI no interprets what the monkey wants to do, it simply interprets whether the button should be pressed or not.

So, you would need one button for everything. You would need enough test subjects using BCI to generate enough generalized brain wave data to train the AI ​​to perform all the functions you want.

The equivalent of this would be if Spotify had to build robots and teach them to play the actual instruments that are used to make each song on the platform.

Every time you wanted to hear “Beat It” by Michael Jackson, you would have to make a request for training with the robots. They would pick up the instruments and make absolutely random noises during thousands or millions of hours of training until they “freaked out” something akin to “Beat It.”

As the AI ​​changed its version of the song, its human developers gave it feedback to indicate whether it was getting closer to the original tune or further.

Meanwhile, a semi-talented human musician could play the entire composition of almost any Michael Jackson song after just a couple of listens.

Elon’s money is no good here

Robots don’t care how rich you are. In fact, the AI ​​doesn’t care about anything because it’s just a bunch of algorithms that combine with data to produce tailor-made results.

People tend to assume that Tesla and Neuralink will solve the AI ​​problem because they essentially have unlimited backing.

But, as Ian Goodfellow at Apple, Yann LeCun on Facebook, and Jeff Dean at Google can tell you: If you could solve autonomous cars, the human brain, or AGI with money, it would have been solved by now.

Musk may be the richest man in the world, but even his wealth doesn’t dwarf the combined worth of the biggest tech companies.

And what the general public doesn’t seem to understand is this: Facebook, Google and Tesla, and all the other AI companies, are working on the same AI problems.

When DeepMind was founded, its purpose was not to win chess or games of Go. Its purpose was to create an AGI. The same is the case with GPT-3 and almost every other multimodal AI system currently in development.

When Ian Goodfellow reinvigorated the field of deep learning with his take on neural networks in 2014, he and others working on similar challenges lit a fire in the tech world.

Since then, we have seen the development of multi-million dollar neural networks that push the limits of computing and hardware. And, even with all that, we could still be decades or more away from self-driving cars or algorithms that can interpret human neural activity.

Money can’t buy a technological breakthrough (it doesn’t hurt, of course, but scientific miracles require more than funding). And, unfortunately for Tesla and Neuralink, many of the world’s smartest and most talented AI researchers know that delivering on Musk’s huge promises can be a wasted effort.

Perhaps that is why Musk has expanded his recruiting efforts beyond experienced AI researchers and is now trying to attract whatever computer talent he can find.

The good news is that no amount of serious evaluation can dampen the spirits of Musk’s indefatigable fans. Whether you can accomplish the good or not has no impact on the amount of worship you receive.

And that’s as likely to change as a Tesla’s ability to produce an autonomous car or Neuralink’s ability to interpret neural activity in the human brain.