Tesla to be More Intelligent Than the Human Brain by 2033?

In an attempt to look special and controversial, Tesla recently ran a presentation on how their future Tesla's will have more computational power than the human brain by 2033, and yet completely fails to understand how that figure has absolutely no meaning whatsoever. What did Tesla claim regarding their vehicles, why is this completely disingenuous, and why is it likely that computers will not think like humans for another few hundred years?

Tesla claims its vehicles will be more computationally powerful than the brain by 2033

Tesla is famous for its sleek electric cars, the advances made in battery technologies, and undoubtedly the man who runs the company, Elon Musk. At the same time, Tesla is also working hard to develop full driving capabilities, recognising that electric cars can only get them so far and that competitors are quickly catching up. In fact, besides the name, there are few incentives for a driver to choose a Tesla over other EVs. 

Now, for all the brilliance that Tesla has introduced to the market, it hasn't been without its share of faults. Certainly, one of the most frightening is that Tesla vehicles disable the entire self-driving system one second before a collision (most likely to absolve Tesla from liability). Another challenge that Tesla has faced is the numerous mistakes made by the vehicles, confusing the moon for an amber light, not spotting children running across the road, and outright not seeing the side of a truck.

Recently, Tesla ran a presentation on the current development of their D1 microchip that aims to provide Tesla vehicles with intelligent capabilities. The use of a custom SoC allows Tesla to craft devices specific to their application, thereby allowing for lower latency processing and improved energy performance. 

The current D1 has the ability to perform 362 trillion operations per second, and this has been stated as being 36% of what the human brain is capable of. The presentation then goes on to show how its future hardware platforms will improve up until 2033, where the microchip platform will have more computational ability than the human brain. 

Why are such figures disingenuous?

Tesla may be right in saying that their microchips by 2033 could have more computational power than the human brain, but so do most data centres and all supercomputers in existence. In fact, when considering that the human brain has computational capabilities of around 100 TFLOPs, it is likely that even small modern servers will be on par with the human brain. And yet, despite modern supercomputers having more processing power than the human brain, not a single machine in existence can match human intelligence.

Comparing the human brain and human intelligence to computers is like comparing an axe and a spade. They are both excellent tools for what they are designed for, but you can't expect to use a spade as an axe and vice versa. The human brain is a massively paralleled machine that is not only able to learn but also grow new connections between neurons (something that electronic systems currently cannot do). A computer, however, is a specialised counting machine capable of performing a sequence of unrelated operations extremely fast. 

As such, a computer can count to 1 billion much faster than a person, but a person can perform any task while simultaneously thinking about what they will eat for dinner and contemplating their existence. It doesn't matter if the Tesla vehicles in 2033 have 100 times more "computational power" than the human brain; it still won't have creativity or common sense and will likely still misinterpret data while driving. This concept of intelligence is even more dramatic when considering that ants seem to have better object avoidance skills compared to modern self-driving systems.

Why is it likely that human-level AI won't exist for at least 100 years?

In the world of technology, it is almost impossible to guess where it will go as researchers constantly make all kinds of discoveries and developments. Many would have guessed that TVs would become thinner, but no one would have guessed that TVs would become cheap and that they could be integrated into a device that fits into a pocket. The introduction of online banking was already thought about decades ago (as early as the 1960s), but no one would have guessed that criminals would be hacking bank details, stealing money, and holding data to ransom.

But in the case of human-level AI, it is likely that it will not exist for a long time, and it is highly unlikely to be some super machine that sees humanity as a threat and then launches some kind of nuclear war while using the internet as an escape route. The simple reason for this comes down to the fundamental difference between how computers and the human brain are constructed.

Electronic circuits can utilise non-volatile memristors to try and simulate connections between neurons, but the human brain has the ability to selectively grow new connections, something that electronic circuits cannot do. Another option for computers is to utilise programmable logic devices that can change their configuration in real time, but this still presents numerous challenges. 

If a true neural network is to be built using electronics, some kind of semi-fluid matrix would be needed that can utilise electric charges to grow connections between points (in a similar fashion to how lithium-ion batteries grow short circuits if overcharged). Such a design would allow for new connections to be established between individual nano processors and effectively replicate the human brain in a circuit, thus allowing for a 1-to-1 copy of human intelligence.