The Uncanny Valley of Automation - Lessons from a Car Crash

A few days ago marked the one year anniversary of when I crashed my car. Like any accident - car or otherwise, there were many contributing factors, but the facts of the event are that I was distracted - I was looking for my phone to confirm an address, looked up and saw my car way too close to a car parked on the side of the road. A road that I would have sworn did not have as much of a curve to it before this trip. I contributed to the damage by hitting the accelerator instead of the break while turning the wheel sharply. Accelerator or no, I still would have hit the trailer behind the car on the road.

After this point, I’m less clear on what happened. After hitting the trailer behind that car, with the added acceleration, my car gripped part of the trailer or SUV that was pulling the trailer, and my car ended up on its side. I and my daughter in the back were buckled in our seats the whole time, her in her car seat and me in the driver’s seat. Before I knew what had happened, the car or my phone, I’m not sure which, had dialed 911. I did my best to explain what happened and where I was. Like many accidents, I was close to home (crazily less than 0.3 miles from my front door). Following the call, passersby stopped to help. I don’t know any of their names, but they opened the rear passenger’s side door to get my daughter out of her seat. I was able to climb out of the window of the front passenger side door and hop down to the ground from the overturned car. Neither of us had any injuries and the paramedics that came to the scene and medical visits the next day confirmed this.

The pictures below show the extent of the damage and the hectic aftermath in the eerie light of police sirens while my car was on its side and after a tow truck flipped it back on its wheels.

What Went Right:

• We both had our seatbelts on and a secure car seat.

• Safety protocols in the car assisted in calling emergency help.

• When the tow truck flipped the car back on its tires, the airbags had deployed. Either from the jolt of being back upright, or during the accident.

• The car was structurally sound throughout the accident - we were able to open all doors and windows and access electrical systems up until days later.

• Insurance covered most of the damage, and I was in a replacement car a few weeks later (probably with higher insurance rates… ).

• The available data from the car’s cameras and sensors makes understanding what happened much clearer.

• My speed was below 30 mph for the entire trip - had it been higher, it could have been a lot worse.

What Contributed to the Accident:

• I was exhausted. Not from anything in particular on this day, but cumulatively. We had finished moving to a new neighborhood a few weeks ago with an almost-two-year-old, and living the life of a parent trying to adjust a child to sleeping in a new place and still working a full day after closing on a house, packing and moving in the weeks leading up to the accident was a recipe for exhaustion.

• I did not know local roads well (hence looking for an address).

• I had been using Tesla’s Full Self Driving (Supervised) feature more and more for the past year. FSD (Supervised) was not enabled during the drive and didn’t contribute to the car’s reaction. The reliance on a driver assistance system did however lead to complacency on my part, and in turn a loss of situational awareness due to an over-reliance on automation.

So What’s the Point?

This is not written as a hit piece against Tesla. (Although some will read it that way. But at least note there was no fire, exploding battery and the generally high safety ratings of Teslas from multiple sources.)

I’m writing this as a caution to understand the limits of automation - read as AI, ADAS or any other upcoming system. Automation and AI are at the uncanny valley - a term often used to describe how computer graphics render human faces where they do not appear completely natural and we can’t explain completely why that is. Similarly, with automation of daily tasks, technology appears helpful, until certain realities happen and a situation goes sideways.

While Tesla does have its approach to automation systems using cameras as its main sensor source, many cars are also adding automated driver assistance systems (ADAS). We are coming to a point where many of these systems are becoming increasingly capable, to the point of fully driving the vehicle. My point for sharing what happened is to reiterate what’s said over and over again - that it’s when we’re in this transition period - when automation is very capable - whether it’s your latest ChatGPT answer, or your car helping you on the highway - when the most attention is needed. Until the capabilities of these tools are well beyond what humans can do, caution is needed.

The graph below shows the increasing amount of mileage I used the ADAS system compared in orange compared to driving on my own in blue. The time scale is for about 15 months, from July 2023 to October 2024. In this time, the software received frequent updates and I become more comfortable relying on it to get me from Point A to Point B.

I was fully away of its capabilities and faults but still was caught of guard in the moment of a strange situation. It’s similar to many high profile aviation accidents where a sequence of events leads to over-correcting or being complacent with automation. A summary of two occurrences where automation led to an unfortunate accident are below:

• Air France Flight 447 - The autopilot disengaged after the pitot tubes iced over, leading to inconsistent airspeed readings. The non-flying pilot, likely suffering from automation surprise and a failure of basic manual flying skills (due to over-reliance on automation), incorrectly applied and maintained a nose-up control input, which contradicted the aircraft's stall protection logic and caused the A330 to enter an unrecoverable aerodynamic stall. The pilots did not recognize the stall condition until it was too late.

• Turkish Airlines Flight 1951 - A faulty radar altimeter on the captain's side caused the autothrottle to retard thrust to the idle setting prematurely during the approach to Amsterdam. Crucially, the crew was not monitoring the airspeed, assuming the autothrottle was maintaining the correct speed. When the speed dropped dangerously low, the pilots were late to react and failed to increase thrust manually, leading to a stall and crash just short of the runway. This is a classic case of complacency and the loss of situational awareness due to over-reliance on automation.

For those who are curious, I share the data for what happened on the day of the accident below. I was at fault, clearly. I am lucky? blessed? that things did not turn out another way, especially for my daughter.

So the least I can do is share my story in the hopes that others can be vigilant when needed. Moreover, I’m writing this as an anniversary reminder to myself that life is short—we’re not guaranteed to walk away from any event.

Build up your fitness so you can help yourself and others when misfortune occurs, read your history so you can prevent repeating the same mistakes over again, and get some sleep. We could all use some more rest.

On DOGE's the Way of the Knife: "Elevate" - Part 14 of ...

I started these posts over five years ago, breaking apart the lyrics in the song Elevate - it seemed like it touched on every aspect of life and then jumped to the next thing.

The song has a lyric -

What don't kill me, gon' make me (gon' make)
Shoot for the stars, no safety (no safety)

We are now in the time of the Department of Government Efficiency (DOGE). A temporary “government” agency with the self-proclaimed mission of “major reform”. DOGE is led by Elon Musk, a person who I have admired in different contexts before, whether it’s SpaceX, Tesla or the latest master plan.

Coming back to the title of this post, “The Way of the Knife” is a concept from the science fiction novel Dune, that professes - in order to go forward, you have to cut off your dependency on things of the past. Dune has had its own resurgence thanks to recent movies, and just like Frank Herbert, Dune’s author, Elon’s perspective on the real world is based on strongly held beliefs. A belief that the world’s resources are finite, a belief that labor is the ultimate scarce resource, a belief that underpopulation is a signal that the worst is about to come. Musk wants to remove the “safety” of layered regulations that in his view will only get more and more complex as time goes on without drastic efforts.

I understand where the impetus for DOGE comes from - the U.S. is in debt, the interest expense on this debt is one of the U.S.’s top five budget expenses, jumping from single digit to double digit percentages of the national budget. Every budget cycle, the question of - can we keep increasing the national debt is just as loud as all of the accompanying expenditures. In the view of DOGE, getting this debt down, and even making a deficit-free budget is the goal.

But where this all breaks down is the method of achieving these cuts. Congress controls the budget and should be the vehicle where these appropriations, cuts and changes are studied, recommended and executed. By having a temporary group of people make these decisions, Congressmen and Senators can feign ignorance and then if things don’t go well, it’s not their fault, it was those DOGE-guys who four months from now will not be around to provide answers.

On the backs of anyone else - a woman, a person of color, a person who does not have their own media company - something like DOGE would fall apart on its first real steps. From my view, both Trump and Musk believe budget disussions are a zero-sum game. It’s either now or never. We will either figure this out in this brief two year Republican-MAGA controlled everything, or the U.S. will be lost.

I do not agree. The damages these cuts will do will not be felt for years or decades. The loss of a regulation here, an enforcer there, a lack of preparedness for a foreseeable disaster will point back to the first half of 2025 when things changed.

For DOGE to be successful, it should have agreed-upon boundaries for what constitutes waste and fraud. It should approach the problem from the mindset that constructive government projects are indeed possible and then remove the steps that create conflict in getting there.

I write all this as a Tesla shareholder (well before Elon’s politics and Twitter era) and car owner, which I know has its own set of conflicts for the implicit support it gives to everything I wrote about above. This doesn’t touch on Musk’s position as a far right symbol, misguided views on DEI and too many other issues to get into here.

Living on the outskirts of DC, these discussions are more than abstract, they include friends who work for various agencies, friends who studied together, friends who have lost their jobs or careers are uncertain regardless of budget, it’s a question of who holds the power. We have elections to give our voice to our represented leaders. Musk has a talking point that says bureaucracy is not democracy, and yet he is acting as the ultimate bureaucrat, walking through countless agencies and expenditures making arbitrary decisions on what stays and what goes - making important decisions as a makeshift official rather than as an elected representative. He has taken this mandate from his adjacency to Trump, who is not seeking the spotlight in this area because cutting specific things is unpopular. Better to take credit for the effort as a whole and leave the details to someone who he will likely disavow before his term is over.

I would ask Musk to put on his sustainability hat for a moment - we agree that battery powered vehicles are the ultimate path forward for sustainable transportation. Did that begin by destroying a chunk of gas stations and then building better alternatives? No, we need to devote more resources to what works and give the regulations, agencies and contracts which have ceased to serve their purpose a chance to end gracefully and without undoing the good they were initially created to achieve.

What will I do with my own Musk-adjacent purchases and what other actions will I take? I’ll take my own advice and stop further contributions to those areas and find the maximum good I can do. I have written my representative and will continue to execute democracy the way we can, by voting and by actions.

“Sample letter to your congress person:

To the office of [Rep. Name] -

I am writing to share my concern about the role the newly created DOGE effort is having on federal regulations and employees. Although DOGE writings use the Presidential executive statement as a reason for their existence I find that the effort is infringing on Congress’ duties to pass a budget and have it enacted appropriately. There is no plan for what effects the cutting efforts will have after the temporary effort’s disbanding and what the transparency looks like in this timeframe. Congresspeople are abdicating their responsibility to this temporary effort and there is no accountability in place for their decisions.

Please do whatever is possible to stop the DOGE effort.

I understand where the impetus for DOGE comes from - the U.S. is in debt, the interest expense on this debt is one of the U.S.’s top five budget expenses, jumping from single digit to double digit percentages of the national budget. Every budget cycle, the question of - can we keep increasing the national debt is just as loud as all of the accompanying expenditures. In the view of DOGE, getting this debt down, and even making a deficit-free budget is the goal.

However, the method of addressing this budget disparities needs to be from Congress and a budget, not from a temporary group of unelected and unaccountable people.

Thank you.”

— Quote Source

Cybertruck: The Next Honda Element or the Next Pontiac Aztek?

On the eve of Tesla’s Cybertruck launch, there are endless posts and articles about what is expected or what is disappointing or what Tesla should have built. On my end, I have no doubt that this will be an impressive vehicle and those that can understand its utility will love their purchase. Although it’s billed as a truck, I think comparing it to a versatile large SUV will more closely match its customer base. What I’m not so sure about is if in ten years, the design will seem completely ridiculous because no other vehicle adopted a similar aesthetic (Pontiac Aztek), or if its owners will hold onto every build year with imitation after imitation (Honda Element).

If you’re not familiar, the Pontiac Aztek was built from 2001 - 2005 and has largely been hailed as one of the ugliest vehicles ever made. Its selling points tried to be its versatility as a camping vehicle, large cargo space, configurable cargo net system and built-in air compressor. I rode in it a few times and mostly think that it was ahead of its time, since basically every vehicle these days is a similar sort of crossover. The biggest difference between it and the Element is that much fewer were built, Pontiac as a GM sub-brand no longer exists, and the car did not age well.

2001 - 2005 Pontiac Aztek

The Honda Element was a small SUV built from 2007 - 2011 that was first introduced appealing to a younger audience but was eventually found its niche for anyone (including the elderly) that wanted a vehicle that was easy to get into and out of, was easy to clean, had numerous attachment points and accessories, and plenty of storage. Contrasted to the Aztek, you still see many Honda Elements out and about, and their owners are a fan-filled community. The orange Element below is driving around 12 years after its last production year, and you will see many for sale on used car lists as a heavily sought-after vehicle.

2008 Honda Element

The Cybertruck fits feature lists from both Aztek and Element:

• Emphasizes function over form

• Emphasizes durability over aesthetic

• Emphasizes ability to be configured for multiple different uses

• Emphasize high amounts of storage space

• Emphasizes atypical vehicle options (built-in air compressor, vehicle to load charging, mutliple tie-down points, ability to drive in large amounts of water)

Honda Element in the wild in 2023 - 12 years after it’s last production year

So, what’s my point? Like the Honda Element, the Cybertruck will find its fans among those who use it in unique ways, and probably won’t fit the use cases Tesla had in mind. There will eventually be a lot of copycats, but its current large truck size will not be duplicated by many. In the future, I can see there being more adventure/car companies who can lean into the boxy, angular design in a smaller form factor. Likely these will trend more toward commercial uses (Boring Company van, last-mile delivery, cargo transport, etc.) rather than personal use.

Tomorrow’s delivery event will be a marker in time - the Pontiac Aztek reborn? Not exactly. This Cybertruck will be around for a long time to come. If you hate it now, you will still be hating it 12 years from now when they will continue to be driven, sold and re-sold.