Phil,

Any sign of Michael Mann's litigation getting a final Court hearing date any time soon?

I'm sure Phil Clarke will find something useful to do with his time one day.

Phil Clarke, It's up to anyone who has something to say to contribute. I have always received a responce to my posts. Most of the people I enjoyed engaging with have left the battlefield.
This from NoTricksZone:-

https://notrickszone.com/2023/06/12/new-research-only-2-of-every-100000-co2-molecules-radiate-photons-and-this-controls-climate/ ?

Just popped by to see what's new, if anything.

I see it is now the Swan 'n' Tomo show.

Enjoy the conversation, guys!

Philip Koopman had a lot of interesting things to say in that podcast.

I've only listened to 3/4 of it so far, but no child in a dinosaur outfit. We had mention of a child in Halloween gear, and a bloke in a chicken outfit. Perhaps a bit of blending going on in Ken Tindell's memory.

Koopman was a bit cagey on details (due to NDAs), but that didn't matter much because self driving is such a diabolical problem that the Devil is in the generalities as well as the details.

He said neural nets were great for demos, but hard work to get near the 100% mark. That accords with what I was told 30 years ago, but I was very surprised at his (implicit) claim that some car makers are not using neural nets at all, doing self-driving with more conventional programming methods. I thought everybody had climbed onto the neural net bandwagon.

The business of noisy video signals was interesting. One thing that went over without particular comment was that things that the receptor hasn't been trained for "disappear". This seems just about the opposite to what humans do. Spotting a man in a chicken suit, the car sees nothing, where you or I would be taking a second look. (would contend the famous man in the gorilla suit is a somewhat different case.)

I suspect that a conservative alogrithm, in which everything that is not specifically recognised is to be treated as a danger, would lead to so many alarms the car would get nowhere. That's as may be, but if I don't recognise it, it doesn't exist is way too cavalier for my taste. There was once a choice of outlooks in network firewalls: allow everything except what's on the blacklist versus deny everything except what's on the whitelist. The second one won the contest, for obvious reasons.

Dr Koopman mentioned the dangers of "a single point of failure" a few times. I think a little more subtlety is called for. It absolutely is a problem when it comes to hardware, but I maintain that the opposite applies with software. Software doesn't wear out, work harden, or whatever. If you have multiple chunks of code solving the same problem you have unnecessary complexity, added maintenance and bloat. (not terribly relevant, but it's one of my hobbyhorses)

One disappointment with the podcast is that (so far, at least) all focus has been on sensing, situational awareness/modelling. For my money, the *big* problem is that, even if all those technical problems are solved, somebody has to program the top-level decision making. The nice thing is that a computer will be *much* more attentive than your typical driver in humdrum day-to-day driving. It will be constantly assigning scores to "apply more power", "another degree left on steering", etc., and optimising the values. Mileage, tyre and brake wear, etc., will be improved.

The nasty thing is that the computer will also have to assign scores in extraordinary circumstances. As I mentioned the other day, the brakes have failed and it's time to decide whether to kill all occupants by ploughing into the truck ahead, or to take evasive action and choose a path through the pedestrians, targeting this one instead of that one because of perceived value. The car is going to be setting values on these "targets" as soon as they enter the field of possible paths. Who decides these values?

This has always struck me as the killer problem with the view that self-driving cars are just ordinary cars on ordinary roads, but with an automated driver. I think it's eventually going to dawn on the developers that the hard vision/modelling problems they've been solving are not that important. It might end up like Godel's incompleteness theorem nullifying all Bertrand Russell's hard work on mathematical consistency.

We already have autonomous vehicles called lifts (or elevators). How about making them more versatile: building to building, suburb to suburb, and all on their own infrastructure? Still plenty of tricky problems, but far more achievable.

Not so sexy though. It's one thing to have a podcast about cars that are smarter than people, there probably aren't any podcasts about liberated lifts.

tomo,

I like the idea of special headgear for hard-core audiophiles, but maybe a dunce's cap would suffice.

The number of "foundations", "institutes", "councils", etc., associated with a particular topic is probably a fair gauge of amount of free loot available.

The NSW Guardianship people strike me as more evil than the UK Court of Protection. Where the latter might celebrate the death of Booker, who had turned the light on their sloth, the NSW mob can celebrate almost any intestate death, because they get to fill their friends' pockets (who, I'm sure, are willing to share). But then what you say of the UK mob says they're doing that too, so maybe they're all just following best practice.

I do find it depressing. I'm in favour of government for one prime purpose: property rights. More or less everything else stems from protecting each person's property (tangible and intangible). It's no bloody good if the government, which is there to protect me and my things from bullies who want to deprive me of them, turns into the worst of bullies itself. That's exactly what has happened to the unfortunate victims of the guardianship caper. It's not incompetence; those "public servants" go home each night knowing what a piece of work they've done each day.

In 1987, the government gave up on the Australia Card, an ID card. It was not at all a popular idea. It's funny that people are now quite happy to pay for an "ID card" that governments could only have dreamt of back then. You carry it on you all day and it constantly reports your position at least to the nearest couple of km (often to the nearest couple of metres). And, as you say, it's not just the government; the phone companies and the app writers get their share of all this information.

Of course it's all fine as long as you can Instagram and TikTok.

Still following up two strands from that Amp Hour podcast. Thought the Autosar real-time operating system sounded interesting, but have failed to turn up much of substance.: a fair bit of marketing blurb on the joys of portability/hardware independence, and a library from NXP for some of their microcontrollers. It gives a choice between waffle-level and lost-in-the-weeds-level documentation. Not the first time. From what I have read, even Autosar sounds a bit heavyweight for my purposes.

The other strand was the autonomous vehicles thing about the child in the dinosaur outfit. Have downloaded the AutonoCast podcast he referred to. Haven't listened yet. Maybe you can beat me to it.