Nostr notes by David J. Atkinson

Honestly, the Waymos seem to be, if not in the clear, then only ...

2025-12-09T01:04:32Z

In reply to nevent1q…8x3m
_________________________

Honestly, the Waymos seem to be, if not in the clear, then only partially to blame. If you read the article, it refers to the street as a “dead end”. Two paragraphs later, the street is “one way”. Now, correct me if I’m wrong but I don’t think you can have both. A dead-end one-way either goes in (and you can never go out), or out (and you can never go back in). I’ve had brouhahas with passengers in my car that started on less than this. Which raises a question: were there any passengers? Who on that street called for more Waymo taxis to come, thus creating a guaranteed newsworthy event?

4/ I’m going to delay elaborating my other concerns about the ...

2025-12-04T22:26:38Z

In reply to nevent1q…89q8
_________________________

4/
I’m going to delay elaborating my other concerns about the maturity of #SelfDriving #autonomous vehicles for now. For now, please consider the following:

1. #AI in general, self-driving cars in particular, are not people. These technology systems do not have our human-lived experiences, they do not think like us, even if you believe that thinking is computational. With a few exceptions, such systems have no common sense ability to reason about the world. They don’t understand human behavior the way we do.

2. They will not make the same mistakes that humans make while driving. That is not only a requirement, it follows from 1. Instead, they will make their own mistakes. We are already seeing plenty of these. Sure, engineers will grind out most of these, but not all.

3. The first two points mean that the behavior of self-driving cars will be difficult to predict in all but the most common vanilla driving situations. People complain about how rigid the current vehicles are at following the law. What? Now you want them to break the law when it is expedient?

4. There are a near infinite number of “edge cases” and those are when safe driving is the most difficult — exactly when we want self-driving vehicles to excel. There are too many to test. The complexity of the real-world, specifically edge cases, cannot be simulated in a laboratory. A decade or more experience on the road is required.

5. Cars are increasingly connected and computerized, and that makes them a new #security threat. Any modern car today can be hacked and remotely controlled. AI systems add multiple new attack vectors. Yes, companies are working on security, but so are the bad guys. #Infosec people will tell you their world is hand-to-hand combat. The more such cars are on the road, the greater the opportunity and attraction for mischief (or worse).

The big question is when will we, as a society, feel safe and convinced by the benefits of self-driving cars? That question is a trap, because most people don’t know the details. It is already happening.

Speaking as an expert and a grandfather, I will not be putting my grandchildren in the back seat of a self-driving car any time soon.

3/ Design issues of #autonomous vehicles: This COULD be a very ...

2025-12-04T21:21:32Z

In reply to nevent1q…4pfv
_________________________

3/
Design issues of #autonomous vehicles:

This COULD be a very long list, but that is not for social media. Instead, I’ll highlight a few major topics. These criticisms are limited by my first hand observations and knowledge.

1. Over-reliance on associative #AI, the branch of machine learning that many people loves to hate. There are some tasks it is really good at, such as low-level perception in multiple modalities, categorizing, and similar tasks dominated by massive amounts of data (including uncertainty).

2. Under-reliance on conceptual/symbolic AI, the “classic” nature technology of automated reasoning, problem solving, situational awareness and recognition, constraint resolution, logic and inference, and more. Don’t let anyone fool you by claiming associative methods are just as good or better.

3. Software/Hardware System architecture. This is both a technology design challenge and a development challenge. Architecture is generally given cursory consideration (e.g., in comparison to #NASA systems) when it should be first priority, dominant and enforced. All kinds of issues, design flaws and more flow from ill-considered system architecture. I saw exactly ONE case where the system was praiseworthy. It takes high skill, training, years of experience, and core discipline to do this right. It is expensive, time-consuming and the talent needed is rare. HOWEVER, there are plenty of examples of good autonomous system architectures to emulate.

4. Naive consideration of the importance of #temporal reasoning, non-monotonic reasoning, guaranteed real-time “good enough” decision-making. (See 2 above)

5. Very little or no explicit knowledge about the world (see 1 above). To illustrate this, I used to give this test to the self-driving technology leadership at companies: Your car is driving along when a ball rolls out into the street from between two parked cars. What does your car do? The typical answer was like this:

Them: yes, the car would see the ball (size of obstacles issue) and it would brake/would not brake (because a car can run over a ball).
Me: what about the kid?
Them: what kid?
Me: The kid chasing the ball into the street?
Them: You didn’t tell us about the kid.
Me: I shouldn’t have to. Kids chase balls into the street.

Arghh! Then the would talk about coding specific scenarios. In other words, the knowledge (in some form) that is CENTRAL to safe driving.

To be continued …

So you are referring to the fossil fuel industry?

2025-10-24T17:20:01Z

In reply to nevent1q…xf3t
_________________________

So you are referring to the fossil fuel industry?

Before my recent retirement, I worked for a global Fortune 500 ...

2024-10-19T18:42:17Z

In reply to nevent1q…ff8q
_________________________

Before my recent retirement, I worked for a global Fortune 500 company as Chief Scientist for #AI. From the inside, and what I saw in other such companies, the primary risks of the AI bubble are not financial, they are reputation and credibility. I’d wager they will suffer both, sooner rather than later. Their crush on AI is mostly marketing hype; there is very little of substance due to lack of talent and investment.