文章探讨了诸如平行停车这类看似简单的日常驾驶技能，为何至今仍存在多种不同的教学方法，并且普遍令驾驶者感到焦虑。作者对比了自然界和科技发展中的“趋同演化”现象，以及“工具收敛”的理论，质疑为何在近百年的汽车发展和数亿驾驶者实践中，平行停车这一普遍技能未能收敛出唯一且最优的解决方案。文章引用了多方数据和理论，试图解释这种现象背后的原因，并引申出对人工智能发展是否会达到“收敛点”的思考。

🚗 **技能多样性与普遍焦虑：** 文章指出，尽管平行停车是一项非常普遍的驾驶技能，但存在多种教学方法，并且调查显示超过半数（57%）的汽车车主在平行停车时感到压力或焦虑，尤其是女性和年轻驾驶者。

💡 **趋同演化与工具收敛理论：** 作者引用“工具收敛”假说，即具有相似子目标且足够智能的代理会收敛于实现目标的特定方法。同时，也提及自然界的趋同演化和科学史上的“多重发现”现象，以及汽车设计（如方向盘、踏板）从多样化到标准化的过程，以此类比平行停车技能的收敛问题。

⏳ **时间与经验的累积：** 文章追溯了平行停车至少有90年的历史，并估算了全球近百年来驾驶者和车辆的总行驶时间。作者质疑，如此庞大的集体经验是否足以促使平行停车方法走向统一，并思考为何这种普遍的技能仍未形成单一、高效且无压力的教学模式。

🤔 **对AI发展的启示：** 文章将平行停车的“不收敛”现象引申到人工智能研究领域，提出AI研究是否也尚未达到“收敛点”，仍然存在许多值得探索的新技术和方法。同时，也对Waymo等自动驾驶汽车如何进行平行停车进行了疑问，暗示技术进步可能带来解决方案的改变。

Published on July 20, 2025 10:37 AM GMT

I was taught two different ways to parallel park. I am sure there are more than just two methods. This strikes me as odd since it's a very common part of driving. Nevertheless, 
I hate parallel parking and it was a constant source of anxiety when learning to drive. Apparently I'm not alone in this:

Over half (57%) of car owners surveyed report feeling stressed or anxious about parallel parking, with women and younger drivers reporting higher levels of anxiety.

Instrumental Convergence hypothesizes that sufficiently intelligent agents with similar to the same sub-goals will converge on those methods or instruments of achieving that. Parking your car parallel to the curb would appear to be one of those. Yet the methods it is done, and the methods of teaching it remain diverse despite possibly a century of hundreds of millions of drivers doing it. With such brute-force exploration strategies why have car drivers as a collective not converged on a single stress-free way to parallel park?

We can observe in nature convergent evolution. And there are many cases of multiple discovery: Both Leibniz and Newton independently arrived at Calculus is a common, but potentially controversial example of multiple independent discovery. 

I have long been under impression that as a domain matures, through a combination of Mathew Effects and adopting the easier or better solution - they converge. Academy Award winner Walter Murch noted that in the early days of the car they didn't all have steering wheels, there were many different approaches to steering, braking, and acceleration until the "UI" eventually converged on three or two pedals, and a steering wheel. He made the analogy that NLE suites had converged on a single UI. Whether they converge on the best method or best UI or not is not as important as they converge on a single thing- much modern technology uses roads and wheels. A.I. research is almost certainly not at the point of convergence yet, there's still a lot of exciting techniques and methods to discover.

In microcosm learning is often the same for an individual as a whole discipline, you try out a whole lot of different approaches to something until you hit upon techniques that provide consistently acceptable results. When I was learning parallel parking I struggled to find different means of representing space, looking for different markers, turning the wheel to different degrees. And I assume through sheer trial and error I accidentally hit upon something which was more consistent.
Ya know, Babble and Prune.

So again I wonder: why was I taught two different ways to do the one - very common - thing?

Maybe not enough time has elapsed. As Orgel's Second Rule goes:

"Evolution is cleverer than you are."

So how long have people been parallel parking for?

I don't know when parallel parking was first invented. "Ranking" cars - where cars lined up in single file along the curb since the early 1900s. The first curb-side parking meters were introduced as late as 1935 - it was patented by Carl Magee and introduced to Oklahoma while he was a newspaper editor there^[1]. So parallel parking is at least 90 years old.

Is 90+ years long enough for drivers to have converged on a single method of parallel parking that consistently works (and how long will it take to converge on a single approach to teaching that best approach?). It's not just 90 years though, there's also the number of drivers and man hours spent driving.

In 2021 there were 232.8 million licensed drivers in the United States. In 2022 there were 463 million licensed car drivers in China. I was unable to find in a timely manner number of licensed drivers in the E.U. but there are 49 million "passenger cars" in Germany. One estimate puts the total "vehicle fleet" of the E.U., U.K. and EFTA at 335 million. Obviously not all these drivers and cars are being driven every day. But it does give us the ability to do a Fermi Estimate on how many man hours have gone into driving, and maybe even how much collective experience there in in Parallel Parking.

Over the course of almost 100 years over a billion people have driven a car, it is not outrageous to assume that at least 500 million of them have learned to parallel park. So why is it so hard? Why does it continue to provide such a source of anxiety for us? So much so that it is a television trope?

If one approach is simply better - why isn't everybody doing it?

Which begs the question: how do Waymos do it? And do they do it the same/differently to other self-driving cars.

It is easy to make misanthropic jokes that humans aren't intelligent agents. But even if Parallel Parking isn't a sub-goal to which agents like humans can converge, there remain numerous examples - within the automobile itself ^[2]- of convergence as domains of knowledge mature or interoperability needs force convergence.

Why can't this everyday annoyance get sorted?
 

1. ^{^}

   Having left Albuquerque after he was acquitted of manslaughter! His original target was a judge.

2. ^{^}

   The wheel can be compared to a killer app as it rolled over the Central Asian steps, becoming a viral hit. As Walter Murch observed, some cars were steered with levers not the steering wheel. The Internal Combustion Engine reigned supreme - albeit with many variants in numbers of cylinders and special variations like Rotary engines.

Discuss