#bayesian reasoning

[gynoidgearhead]

[Image caption for above addition: two comics, each headlined "Mental Gymnastics". The first is a woman or girl stepping a few steps on a gymnastic mat and then happily extending her arms, with a caption reading "I guess fairies are real."

The second has a man energetically swinging from a bar, first from his hands and then his feet; then swinging from rings; then using a pommel horse; then jumping over a flaming car while in a superhero outfit. The various steps are captioned: "A walrus has escaped a zoo or decided to leave its territory for some reason then traveled hundreds of miles of waterways and rivers without being noticed." / It then decided to leave the water and start traveling on land specifically in my city, traveling across miles of streets without being run into or stopped." / "Then it came to my house in particular, made its way to the door, then knocked on it and politely waited." End caption.]

Or someone decided to pull a very illegal prank on you and deliver a walrus to your doorstep, in the manner of xkcd's "instead of office chair, package contained bobcat. would not buy again".

I've asked this question before and been surprised by the results, now I have access to more weirdos it's your problem:

It is the middle of a Sunday afternoon. You have nothing on, and aren't expecting visitors, deliveries or post.

Unexpectedly, there is a knock at the door.

[catboysooyoung]

Why I Find Soleum & Saheon Fucking Hilarious: a yapfest

I talked about this with my good friend Carrot today, who mentioned how it's a bit silly how Saheon, someone who's supposed to be dangerous, is utterly terrified of this harmless guy. Meanwhile Soleum, despite being a self proclaimed coward who is fully aware of how dangerous this guy could be, has so far, still remained coolly in control.

And I said that it's because they're having the funniest standoff ever, that it's like I'm watching a nature documentary whenever I see them.

Kim Soleum, the clever, benevolent "prey"

Kim Soleum constantly acts like he's a cornered prey animal inside, but he's someone who is naturally quite levelheaded, and is someone who makes it a habit to take several steps back, internalize and rationalize things a lot before acting. Combine this with his near-encyclopedic knowledge of the Darkness Exploration Records and Baek Saheon's character settings, and you basically get a wild animal wrangler (the wild animal being BSH haha).

Kim Soleum fully knows of how dangerous this guy gets (after all, he was called Viper for a reason), has seen it firsthand, and knows that is barely the tip of the iceberg, but he also knows of Baek Saheon's mentality and is familiar with his behavior. He's even mentioned this before, in chapter 92, mentioning how he was starting to "miss" Baek Saheon, because at least the guy was very predictable. Lmao

Thus, BSH here is a danger he could easily "handle" and confidently contain, perhaps even going so far as taming and keeping him on a leash for his own use. And all he needed to do here, is to give him a taste of his own medicine, which is to act even more unhinged than Saheon, that is, to out-freak the guy, convince Baek Saheon that he is, in fact:

A threat (stronger and/or more capable than him)

Someone who cannot be reasoned with (due to his posturing as a "deranged dopamine junkie")

In short, you could say that Soleum is a "prey" with Bayesian Mimicry. He makes himself seem more dangerous than he really is, in order to fool the "predator" to avoid being eaten.

Of course, he naturally succeeded, rendering the "viper" into a simple garden snake, taming and containing him. And this is precisely because of how Baek Saheon operates.

Baek Saheon, the not-so gutsy predator

Baek Saheon, unlike Kim Soleum, seemed to have been reared to be a natural, opportunistic predator. He may not be at the very top, but he's the type to slither around, waiting for the most opportune moment to strike at his prey. He's got incredible survival instincts. (Not just anyone would think of knocking out somebody's eyeball to survive, let alone have the ferocity to do it.)

Saheon is someone who acts on these very instincts. They are what have kept him alive so far, after all. (And had Soleum not intervened, or even shown off that spare eyeball, he would've walked out of that Darkness, assured that he's done the best he could to survive that, and it would've bolstered his confidence enough to surpass everyone else and earned him the top recruit position.)

To Saheon, it was natural to rely on them. This is not to say that he's incapable of thinking more thoroughly, but more of... The fact that Saheon has a more... Hands on approach than Kim Soleum. He relies on his senses and experiences and he's always on edge and in survival mode. What's more, he doesn't have all the background knowledge like KSE.

You know how people say, if you see a snake where it shouldn't be, you should always just assume that it's venomous and maintain proper distance? Because even on the off chance that it wasn't venomous, you'd inevitably react with more caution and would not approach it so recklessly. So in Saheon's POV, it doesn't matter that there's an off chance that KSE isn't really the freak he's made himself out to be. There's still a huge chance that he really was, and as Saheon (subjectively) had not had any evidence to prove otherwise, he would remain thoroughly cautious of Kim Soleum, branding the guy as "someone dangerous" in his mind. He wouldn't risk it. Much. It was natural to submit to stronger people, at least to a certain point, you can't really fight them head on.

Granted, this doesn't mean he's been completely tamed/cowed. Like I said, BSH is like a snake in the grass. In contrast to Kim Soleum, he adopts a more aggressive mimicry, where he feigns meekness and weakness in order to let his "prey's" guard down, aiming to strike when they're down. To be clear, he doesn't do that to just KSE, but everybody. In fact, he's done it since episode 2. He's not as meticulous in performing it though, seeing that he's already clocked by Go Yeongeun and Lee Seonghae.

And the moments he tried to "strike" KSE down, were when he was convinced that KSE was in a 'vulnerable' enough of a state to be retaliated against. Baek Saheon is not too reckless, he just acts on instinct and calculates the risks and rewards of each action, then he charges forward immediately.

[triviallytrue]

Other funny takeaways from that ACX post:

- The guy arguing for zoonotic origin was ridiculously hypercompetent, to the point of having photographic memory of virtually every study published on the subject. One of the things the lab leak guys attributed their loss to was that their opponent had basically become one of the world's foremost experts on COVID origins through grindset alone and they couldn't keep up

- This great quote from Scott Alexander:

While everyone else tries “pop Bayesianism” and “Bayes-inspired toolboxes”, Rootclaim asks: what if you just directly apply Bayes to the world’s hardest problems? There’s something pure about that, in a way nobody else is trying.

Unfortunately, the reason nobody else is trying this is because it doesn’t work.

[max1461]

Regarding the culture on here (and twitter) being overly concept-centric; people having a tendency to view particular things as being mere instantiations of concepts as opposed to concepts emerging from the particulars of life:

It's like, whether it's therapy jargon or Bayesian reasoning or Marxist analysis, people on here really seem to love analyzing everything through some kind of established frame. Often the frame is constantly rotating. When social justice was tumblr's épistémè it used to constantly pick up new frameworks, and still the rationalists are always picking up on some new framework (or rolling their own). There's always a new framework du jour.

Nobody just looks at shit and decides independently what they think of it. Nobody just has specific, isolated, personal opinions about this or that particular topic. It's all tied into some established body of theory. Or at least, when people do just look at shit and come to an independent opinion, they feel the need to articulate it in reference to some established body of theory.

I hate this! I can't express how wrong I think this is in terms of like, being a productive way to engage with the world. Like, this kind of social theorizing... whether Marxist, rationalist, or psychiatric... it should be secondary. It should be something you put in an essay or research paper, and not something you live your life by. Why? Well there are a lot of reasons. Models, even accurate models, are necessarily simplifications of reality. In your actual lived experience you have the kind of direct knowledge of a situation that makes this sort of theorizing superfluous. And more humanistically, subordinating the actuality of your life, your human interactions, etc. to like... static, lifeless ideas... it sucks! Why would you do this!

There are epistemic problems with social theorizing in general too, but that's not even the point, that's beside the point.

Before anyone is a proletarian, an autistic narcissist, a Bayesian agent... they are an individual, they are the person in front of me, right? Before any of my actions are elements of any system or pattern, they are my specific actions, taken in some moment to some specific end. All the concepts are secondary. The real world is actual.

Right? Right?

This is not really a fact claim so much as an... emotional claim? I am claiming the correctness of some specific outlook or way of relating to the world, not some specific set of facts. And actually I think there's a lot of value in the concept-first way of relating. I certainly don't want to embody it, to me it seems awful, sterile, frightening and lonely. But some people don't feel that way, they love it, and that's great on its own, and beyond that: through their way of engaging with the world they clearly produce all sorts of concepts that I and others like me would never think to produce, and some of those concepts are bound to turn out to be nonsense but some will turn out to be worthwhile, and there you go. There is value in being a uh... a conceptualist, I'm going to call this worldsense "conceptualism". But I don't want to be one. Or be expected to act like one, as is the implicit expectation on here. I want to be an actualist. I am an actualist.

See look. Even here I'm framing things in terms of concepts. I'm behaving now as a conceptualist. I don't do this naturally. But it's impossible not to write for your audience. It's impossible not to speak to your audience. Some people can do it but not me. Speech, for me, inherently involves a model of the other, and I cannot speak but specifically to the other I model. And on here, the other is conceptualist, so I behave as a conceptualist.

[kaurwreck]

mori speaks sincerely only around chuuya, because chuuya fought the port mafia to protect others from the old boss's reign of terror, and because when the sheep fell apart, chuuya did not blame them for being scared children, but asked mori to teach him how to lead.

kouyou speaks warmly to chuuya, likely in no small part because when she met him and told him she would take him to a negotiation, he fretted over whether he could reach others.

verlaine survived enough to be given a second opportunity to be a person by rimbaud because despite all that verlaine did to him, chuuya recognized his loneliness was ferocious love yearning, and he honored that love in verlaine.

chuuya is so special and so powerful, not because he has a storm inside of him, but because he's disciplined and self possessed enough to tame it within himself, uncertain enough to question himself, and discerning enough to know when to act and when to restrain himself.

the port mafia's leadership is wrapped around chuuya's finger; it's deranged. mori canonically unmasks around chuuya, kouyou mentored chuuya, verlaine would kill and die (and has done both) for chuuya. ace is dead.

[vague-humanoid]

cool how anyone who does a financial crime against the poor gets a government bailout, but if you defraud the rich you die for no reason within a week.

[canmom]

i probably would call myself a consequentialist, but not a utilitarian. my objection to utilitarianism is similar to my objection to the absolutist Bayesianism practiced in That Subculture: it's a philosophy that claims to be based around a certain computation, but actually performing that computation is completely intractable. there's no way to actually update your probability assignments of all possible statements in response to new information, any more than it's possible to aggregate the total happiness/suffering/whatever across the entire future for each imaginable course of action.

so this calculation is entirely notional. what you're actually doing is coming up with verbal arguments and vague heuristics for how you think this notional calculation would work. perhaps it's as good an entry point as any. but the supposed mathematical rigour is just rhetoric! you can talk about utilons this and QALYs that, but there is no way to calculate this shit, it's just a mathematical coat of paint.

the second objection is the 'seeing like a state' objection (or seeing like a company/NGO): the 'utility function' is a construct used to make economic models. it doesn't model humans particularly well, who have a variety of competing impulses that don't lend themselves to nice formalisms. and to demand that you should live according to a utility function is accordingly to strip the world of its complexity to make it more tractable. instead of specific people with specific desires and needs and relationships into which you fit, which aren't necessarily commensurable, you have abstract fungible units of pleasure or suffering or whatever else you're trying to optimise.

this worldview appealed to me as a teenager. I imagined that you could model an agent as a some kind of surface between it and the world - a sphere, perhaps, inside your head; the course of your life would be the movement of particles in and out of this sphere, and theoretically there would be a pattern for every instant of time that would lead to the best possible impact on the world, solving 'life' much like a tool assisted speedrun solves a game. the goal would be then to approximate this optimal run as much as possible. then I'd think of problems with this model: couldn't you just spawn high energy photons on the sphere to melt shit like a laser? we'd have to put some restrictions on it, obviously. what if the optimal run was really close to a harmful run, so a small mistake would lead to disaster? perhaps you'd be better to find a stable local maximum instead. and so on.

I'm not sure what good it did me to imagine this funny (or if you prefer, terminally STEM-brained) thought experiment, but it was very nice and mathematical-looking, and back then I really wanted my philosophy to be impossibly demanding for some reason. some weird combo of depression and autism and a self image very much dependent on being told i was good?

these days my feeling is that the pretense of mathematical rigour where it doesn't exist is untrustworthy, and particularly where people are concerned, abstracting too much loses important information. I'm not a court of law where strict consistency matters for the sake of stability or whatever, nor a government trying to figure out which levers to pull to create the ideal society - I'm an organism embedded in a bewilderingly complex system, and I can take each situation as it comes. treating the people I interact with well is important to me. I still sometimes think along utilitarianish lines sometimes - particularly 'this person could use this money more than me' - but I make no pretense to rigour or optimisation with it.

[skitter-queen]

oh youre a "thinker 5?" cute. i know bayesian reasoning

[blubberquark]

Things That Are Hard

Some things are harder than they look. Some things are exactly as hard as they look.

Game AI, Intelligent Opponents, Intelligent NPCs

As you already know, "Game AI" is a misnomer. It's NPC behaviour, escort missions, "director" systems that dynamically manage the level of action in a game, pathfinding, AI opponents in multiplayer games, and possibly friendly AI players to fill out your team if there aren't enough humans.

Still, you are able to implement minimax with alpha-beta pruning for board games, pathfinding algorithms like A* or simple planning/reasoning systems with relative ease. Even easier: You could just take an MIT licensed library that implements a cool AI technique and put it in your game.

So why is it so hard to add AI to games, or more AI to games? The first problem is integration of cool AI algorithms with game systems. Although games do not need any "perception" for planning algorithms to work, no computer vision, sensor fusion, or data cleanup, and no Bayesian filtering for mapping and localisation, AI in games still needs information in a machine-readable format. Suddenly you go from free-form level geometry to a uniform grid, and from "every frame, do this or that" to planning and execution phases and checking every frame if the plan is still succeeding or has succeeded or if the assumptions of the original plan no longer hold and a new plan is on order. Intelligent behaviour is orders of magnitude more code than simple behaviours, and every time you add a mechanic to the game, you need to ask yourself "how do I make this mechanic accessible to the AI?"

Some design decisions will just be ruled out because they would be difficult to get to work in a certain AI paradigm.

Even in a game that is perfectly suited for AI techniques, like a turn-based, grid-based rogue-like, with line-of-sight already implemented, can struggle to make use of learning or planning AI for NPC behaviour.

What makes advanced AI "fun" in a game is usually when the behaviour is at least a little predictable, or when the AI explains how it works or why it did what it did. What makes AI "fun" is when it sometimes or usually plays really well, but then makes little mistakes that the player must learn to exploit. What makes AI "fun" is interesting behaviour. What makes AI "fun" is game balance.

You can have all of those with simple, almost hard-coded agent behaviour.

Video Playback

If your engine does not have video playback, you might think that it's easy enough to add it by yourself. After all, there are libraries out there that help you decode and decompress video files, so you can stream them from disk, and get streams of video frames and audio.

You can just use those libraries, and play the sounds and display the pictures with the tools your engine already provides, right?

Unfortunately, no. The video is probably at a different frame rate from your game's frame rate, and the music and sound effect playback in your game engine are probably not designed with syncing audio playback to a video stream.

I'm not saying it can't be done. I'm saying that it's surprisingly tricky, and even worse, it might be something that can't be built on top of your engine, but something that requires you to modify your engine to make it work.

Stealth Games

Stealth games succeed and fail on NPC behaviour/AI, predictability, variety, and level design. Stealth games need sophisticated and legible systems for line of sight, detailed modelling of the knowledge-state of NPCs, communication between NPCs, and good movement/ controls/game feel.

Making a stealth game is probably five times as difficult as a platformer or a puzzle platformer.

In a puzzle platformer, you can develop puzzle elements and then build levels. In a stealth game, your NPC behaviour and level design must work in tandem, and be developed together. Movement must be fluid enough that it doesn't become a challenge in itself, without stealth. NPC behaviour must be interesting and legible.

Rhythm Games

These are hard for the same reason that video playback is hard. You have to sync up your audio with your gameplay. You need some kind of feedback for when which audio is played. You need to know how large the audio lag, screen lag, and input lag are, both in frames, and in milliseconds.

You could try to counteract this by using certain real-time OS functionality directly, instead of using the machinery your engine gives you for sound effects and background music. You could try building your own sequencer that plays the beats at the right time.

Now you have to build good gameplay on top of that, and you have to write music. Rhythm games are the genre that experienced programmers are most likely to get wrong in game jams. They produce a finished and playable game, because they wanted to write a rhythm game for a change, but they get the BPM of their music slightly wrong, and everything feels off, more and more so as each song progresses.

Online Multi-Player Netcode

Everybody knows this is hard, but still underestimates the effort it takes. Sure, back in the day you could use the now-discontinued ready-made solution for Unity 5.0 to synchronise the state of your GameObjects. Sure, you can use a library that lets you send messages and streams on top of UDP. Sure, you can just use TCP and server-authoritative networking.

It can all work out, or it might not. Your netcode will have to deal with pings of 300 milliseconds, lag spikes, package loss, and maybe recover from five seconds of lost WiFi connections. If your game can't, because it absolutely needs the low latency or high bandwidth or consistency between players, you will at least have to detect these conditions and handle them, for example by showing text on the screen informing the player he has lost the match.

It is deceptively easy to build certain kinds of multiplayer games, and test them on your local network with pings in the single digit milliseconds. It is deceptively easy to write your own RPC system that works over TCP and sends out method names and arguments encoded as JSON. This is not the hard part of netcode. It is easy to write a racing game where players don't interact much, but just see each other's ghosts. The hard part is to make a fighting game where both players see the punches connect with the hit boxes in the same place, and where all players see the same finish line. Or maybe it's by design if every player sees his own car go over the finish line first.

[compneuropapers]

Interesting Papers for Week 17, 2024

Computational mechanisms underlying latent value updating of unchosen actions. Ben-Artzi, I., Kessler, Y., Nicenboim, B., & Shahar, N. (2023). Science Advances, 9(42).

Associative learning or Bayesian inference? Revisiting backwards blocking reasoning in adults. Benton, D. T., & Rakison, D. H. (2023). Cognition, 241, 105626.

The value of mere completion. Converse, B. A., Tsang, S., & Hennecke, M. (2023). Journal of Experimental Psychology: General, 152(11), 3021–3036.

Stable sound decoding despite modulated sound representation in the auditory cortex. Funamizu, A., Marbach, F., & Zador, A. M. (2023). Current Biology, 33(20), 4470-4483.e7.

Differential attentional costs of encoding specific and gist episodic memory representations. Greene, N. R., & Naveh-Benjamin, M. (2023). Journal of Experimental Psychology: General, 152(11), 3292–3299.

The scaling of mental computation in a sorting task. Haridi, S., Wu, C. M., Dasgupta, I., & Schulz, E. (2023). Cognition, 241, 105605.

Simulations predict differing phase responses to excitation vs. inhibition in theta-resonant pyramidal neurons. Kelley, C., Antic, S. D., Carnevale, N. T., Kubie, J. L., & Lytton, W. W. (2023). Journal of Neurophysiology, 130(4), 910–924.

Attention preserves the selectivity of feature-tuned normalization. Klímová, M., Bloem, I. M., & Ling, S. (2023). Journal of Neurophysiology, 130(4), 990–998.

An approximate representation of objects underlies physical reasoning. Li, Y., Wang, Y., Boger, T., Smith, K. A., Gershman, S. J., & Ullman, T. D. (2023). Journal of Experimental Psychology: General, 152(11), 3074–3086.

Associative and predictive hippocampal codes support memory-guided behaviors. Liu, C., Todorova, R., Tang, W., Oliva, A., & Fernandez-Ruiz, A. (2023). Science, 382(6668).

Same but different: The latency of a shared expectation signal interacts with stimulus attributes. Lowe, B. G., Robinson, J. E., Yamamoto, N., Hogendoorn, H., & Johnston, P. (2023). Cortex, 168, 143–156.

Model-free decision making resists improved instructions and is enhanced by stimulus-response associations. Luna, R., Vadillo, M. A., & Luque, D. (2023). Cortex, 168, 102–113.

Infants’ sex affects neural responses to affective touch in early infancy. Mariani Wigley, I. L. C., Björnsdotter, M., Scheinin, N. M., Merisaari, H., Saunavaara, J., Parkkola, R., … Tuulari, J. J. (2023). Developmental Psychobiology, 65(7), e22419.

Action planning and execution cues influence economic partner choice. McEllin, L., Fiedler, S., & Sebanz, N. (2023). Cognition, 241, 105632.

Parallel processing of value-related information during multi-attribute decisions. Nakahashi, A., & Cisek, P. (2023). Journal of Neurophysiology, 130(4), 967–979.

Extended trajectory of spatial memory errors in typical and atypical development: The role of binding and precision. Peng, M., Lovos, A., Bottrill, K., Hughes, K., Sampsel, M., Lee, N. R., … Edgin, J. (2023). Hippocampus, 33(11), 1171–1188.

Mediodorsal thalamus-projecting anterior cingulate cortex neurons modulate helping behavior in mice. Song, D., Wang, C., Jin, Y., Deng, Y., Yan, Y., Wang, D., … Quan, Z. (2023). Current Biology, 33(20), 4330-4342.e5.

Metacognition and sense of agency. Wen, W., Charles, L., & Haggard, P. (2023). Cognition, 241, 105622.

Sensory deprivation arrests cellular and synaptic development of the night-vision circuitry in the retina. Wisner, S. R., Saha, A., Grimes, W. N., Mizerska, K., Kolarik, H. J., Wallin, J., … Hoon, M. (2023). Current Biology, 33(20), 4415-4429.e3.

The timing of confidence computations in human prefrontal cortex. Xue, K., Zheng, Y., Rafiei, F., & Rahnev, D. (2023). Cortex, 168, 167–175.

[frank-olivier]

Bayesian Active Exploration: A New Frontier in Artificial Intelligence

The field of artificial intelligence has seen tremendous growth and advancements in recent years, with various techniques and paradigms emerging to tackle complex problems in the field of machine learning, computer vision, and natural language processing. Two of these concepts that have attracted a lot of attention are active inference and Bayesian mechanics. Although both techniques have been researched separately, their synergy has the potential to revolutionize AI by creating more efficient, accurate, and effective systems.

Traditional machine learning algorithms rely on a passive approach, where the system receives data and updates its parameters without actively influencing the data collection process. However, this approach can have limitations, especially in complex and dynamic environments. Active interference, on the other hand, allows AI systems to take an active role in selecting the most informative data points or actions to collect more relevant information. In this way, active inference allows systems to adapt to changing environments, reducing the need for labeled data and improving the efficiency of learning and decision-making.

One of the first milestones in active inference was the development of the "query by committee" algorithm by Freund et al. in 1997. This algorithm used a committee of models to determine the most meaningful data points to capture, laying the foundation for future active learning techniques. Another important milestone was the introduction of "uncertainty sampling" by Lewis and Gale in 1994, which selected data points with the highest uncertainty or ambiguity to capture more information.

Bayesian mechanics, on the other hand, provides a probabilistic framework for reasoning and decision-making under uncertainty. By modeling complex systems using probability distributions, Bayesian mechanics enables AI systems to quantify uncertainty and ambiguity, thereby making more informed decisions when faced with incomplete or noisy data. Bayesian inference, the process of updating the prior distribution using new data, is a powerful tool for learning and decision-making.

One of the first milestones in Bayesian mechanics was the development of Bayes' theorem by Thomas Bayes in 1763. This theorem provided a mathematical framework for updating the probability of a hypothesis based on new evidence. Another important milestone was the introduction of Bayesian networks by Pearl in 1988, which provided a structured approach to modeling complex systems using probability distributions.

While active inference and Bayesian mechanics each have their strengths, combining them has the potential to create a new generation of AI systems that can actively collect informative data and update their probabilistic models to make more informed decisions. The combination of active inference and Bayesian mechanics has numerous applications in AI, including robotics, computer vision, and natural language processing. In robotics, for example, active inference can be used to actively explore the environment, collect more informative data, and improve navigation and decision-making. In computer vision, active inference can be used to actively select the most informative images or viewpoints, improving object recognition or scene understanding.

Timeline:

1763: Bayes' theorem

1988: Bayesian networks

1994: Uncertainty Sampling

1997: Query by Committee algorithm

2017: Deep Bayesian Active Learning

2019: Bayesian Active Exploration

2020: Active Bayesian Inference for Deep Learning

2020: Bayesian Active Learning for Computer Vision

The synergy of active inference and Bayesian mechanics is expected to play a crucial role in shaping the next generation of AI systems. Some possible future developments in this area include:

- Combining active inference and Bayesian mechanics with other AI techniques, such as reinforcement learning and transfer learning, to create more powerful and flexible AI systems.

- Applying the synergy of active inference and Bayesian mechanics to new areas, such as healthcare, finance, and education, to improve decision-making and outcomes.

- Developing new algorithms and techniques that integrate active inference and Bayesian mechanics, such as Bayesian active learning for deep learning and Bayesian active exploration for robotics.

Dr. Sanjeev Namjosh: The Hidden Math Behind All Living Systems - On Active Inference, the Free Energy Principle, and Bayesian Mechanics (Machine Learning Street Talk, October 2024)

Saturday, October 26, 2024

[malbecmusings]

Even before al the bodies have been recovered, the blame game is in full swing over Perini Navi's 55m Bayesian that went down in a storm off Porticello, Sicily. I read this morning the chief of the group that owns Perini Navi came out swinging, putting the blame squarely on the crew.

There's a good technical analysis from designer Chris Feers below the cut.

TL;Don't want to read: Monohulls of all stripes are designed to survive a knockdown; Perini's engineering is some of he best in the world. Bayesian had a retractable keel which could shorten her 32' draft to 12'. With the keel extended, meaning she had full use her counterweight, she would have been able to right herself in a full knockdown. Even if her mast were below the waterline, even if she lost her mast, maybe more so if she lost her mast, assuming she was watertight, her design would allow her to self recover. With the keel retracted tho, her righting moment would be dramatically less. If she wasn't watertight, a knockdown would be unrecoverable and she would, as she did, quickly end up on the bottom of the ocean.

I have so many questions. We won't know the full story for a bit but seven people losing their lives should have been an avoidable outcome.

This is a one in a million tragedy but we should examine the facts and learn from them. Bayesian boasts the second highest mast in the world at 75 metres on a length of 56m. She has a lifting keel to enable her to get into shallow areas. Fully down it gives a draft of 9.83m and raised a draft of 4m. A sailing yacht has a keel to counter the heeling moment generated by the power of her sail plan. I’m sorry to say that size matters to a superyacht owner and naval architects are seduced into providing solutions. As yacht size increases the resistance of the hull reduces in proportion, so less sail area is required to adequately power longer yachts. But these floating fashion items are driven by appearance and bragging rights - and you lose prestige if someone has a bigger mast than you. Always the status pecking order questions are – how big – how fast – what cost – and is it black? If you designed Bayesian with a reasonable sail area and a ‘normal’ mast she would not look impressiive – which is what superyachts have to be.

The stability of a yacht has to be sufficient to counter the power of the rig but, as mast heights increase, the keels can often become so deep that the places of interest are restricted hence the lifting keel solution. Stability comes from two factors – the hull form and the ballast keel which acts like a pendulum. As the yacht heels the volume of the immersed hull section produce a buoyancy force which resists heeling. Initially the keel gives little force but as the angle of heel increases ‘physics’ makes the keel contribution significant (leverage). The greater the keel length, the greater the effect. The combination of the hull buoyancy on the heeled side and the keel on the ‘windward’ side produces the force necessary to keep the yacht from capsize. If the keel of Bayesian was retracted it would lose a significant six meters of moment arm or leverage from its probable 200 tons of keel bulb. When we design yachts we calculate the stability, or righting lever, as a function of heeled ‘bouyancy’ force and the ballast moment arm combined. (the GZ) This can be plotted on a graph to show the stability at any heel angle and identifies the angle at which stability becomes negative causing the yacht to capsize.

Normally an ocean yacht will experience a negative point at about 120 degrees of heel. With a lifting keel this point is greatly reduced maybe to less than 90 degrees. If Bayesian was at anchor with the keel raised and no sail up the crew would have every confidence that she could remain safe in most normal wind conditions. Every captain at this level has passed an exam on stability and would be aware of his vessels stability graph.

Many years ago I sat at Cremorne and watched a spiralling williwaw race across Sydney harbour and pass through Mosman. This twister was only about 30 metres wide but it destroyed houses and overturned cars in its path. A few feet away nothing was harmed.

The power of a twister is intense and powerful with the wind is coming from every direction. This was what hit Bayesian. The problem of large rigs is windage, even with no sails. But this yacht had three furling sails forward and a big boom with the weight of a furled mainsail inside all above the centre of gravity. Also there were a few communication domes on the spreaders.

We use a wind pressure coefficient to measure the force of the wind on the rig and sails. Even without sails the WPC for Bayesian must have been pretty large when hit by a wind force of varying direction with a local velocity way above the norm. Once she was knocked down beyond her stability limit with the keel up she stood no chance and, laying flat to the water, her deck openings would have allowed a flood of water aboard and she would founder. This would happen in a couple of minutes.

The observation of a lightning strike can be discounted because these vessels are grounded and any damage from a strike would have caused a slow sinking at worst – not a capsize and founder.

The individuals within a professional crew with sailing experience may have sensed the wind and motion of the vessel and quickly reacted to instinctively save themselves in the seconds they had. My guess is that some were already on deck alarmed by the general conditions.The guests would have found themselves totally disoriented in flooding cabins, in darkness with the walls, doors and passageways at ninety degrees to the norm. They had practically no chance because it would be completely beyond their experience. The crew would have been unable to be of any help due to the speed of the unexpected event.

I have been a professional yacht designer and builder for fifty years specialising in lifting keel yachts. My son, a professional navigator, was Third Officer on a ketch superyacht with masts 100m tall; a yacht so big, at 88metres, that it was almost beyond human handling even with the machinery on board. But of course it is the biggest and most expensive’ etc etc. What we have here is a one off accident which is a wake up call to an industry where common sense has departed as yachts get more silly in size and design.

In summary Bayesian was caught in the wrong place at the wrong time. A freak accident which the designers and crew would have little chance to predict.

If the keel had been down she would have probably survived the knock down. But without sails up the crew would have experience of her basic stability for normal conditions which would have felt adequate. Any enquiry must examine the design factors such as the stability vanishing point in the condition she was at the time of the accident; keel up, tank loadings and rig factors for windage (WPC) and centre of gravity etc. And a calculation of the wind force required to heel the boat to 90 degrees in the condition at the time of the accident.

All forms of transport have had these unpredictable one off events leading to changes of regulations and professional practice. Titanic, Boeing, 1955 Le Mans, the 1952 Farnborough crash, the 1979 Fastnet – all have made a difference and these events all came unpredicted and out of the blue often at a time of complacency.

Chris Freer – yacht designer – August 2024

[mostlysignssomeportents]

Object permanence

I'm on a 20+ city book tour for my new novel PICKS AND SHOVELS. Catch me at NEW ZEALAND'S UNITY BOOKS in AUCKLAND TOMORROW (May 2), and in WELLINGTON on SATURDAY (May 3). More tour dates (Pittsburgh, PDX, London, Manchester) here.

#20yrsago Danny O’Brien goes to work at EFF! https://web.archive.org/web/20050507123924/https://www.oblomovka.com/entries/2005/04/29#1114782180

#15yrsago 1939 World’s Fair: the future’s cradle, in pictures https://web.archive.org/web/20100501170616/http://www.wired.com/thisdayintech/2010/04/gallery-1939-worlds-fair/

#10yrsago British austerity: a failed experiment abandoned by the rest of the world https://www.theguardian.com/business/ng-interactive/2015/apr/29/the-austerity-delusion

#10yrsago Translation: once they learn the truth, techies hate and fear us https://www.wired.com/2015/04/us-defense-secretary-snowden-caused-tensions-techies/

#10yrsago FBI’s crypto backdoor plans require them to win the war on general purpose computing http://webpolicy.org/2015/04/28/you-cant-backdoor-a-platform/

#10yrsago Anyone can open a Master Lock padlock in under two minutes https://www.youtube.com/watch?v=09UgmwtL12c

#10yrsago Couples counsellor who assigns Ikea furniture assembly calls Liatorp “The Divorcemaker” https://web.archive.org/web/20150430183654/https://laist.com/2015/04/28/santa_monica_therapist_uses_ikea_as.php

#10yrsago UK Tories forged letter of support in the Telegraph from “5,000 small businesses” https://sturdyblog.wordpress.com/2015/04/27/small-business-letter-to-the-telegraph-an-attempt-to-defraud-the-electorate/

#5yrsago How monopolism crashed the US food supply https://pluralistic.net/2020/04/29/banjo-nazis/#butchery

#5yrsago Legendary troubleshooting stories https://pluralistic.net/2020/04/29/banjo-nazis/#cuckoos-egg

#5yrsago Medical debt collection during the pandemic https://pluralistic.net/2020/04/29/banjo-nazis/#armbreakers

#5yrsago British Library releases 1.9m images https://pluralistic.net/2020/04/29/banjo-nazis/#it-belongs-in-a-museum

#5yrsago NSO Group employee used Pegasus cyberweapon to stalk a woman https://pluralistic.net/2020/04/29/banjo-nazis/#loveint

#5yrsago Founder of AI surveillance company was a Nazi who helped shoot up a synagogue https://pluralistic.net/2020/04/29/banjo-nazis/#damien-patton-nazi

#5yrsago Talking Radicalized with the CBC https://pluralistic.net/2020/04/29/banjo-nazis/#zuckervegans

#5yrsago Bayesian reasoning and covid-19 https://pluralistic.net/2020/04/29/banjo-nazis/#uncertainty

#5yrsago Cigna claims to be rolling in dough and on the verge of bankruptcy https://pluralistic.net/2020/04/29/banjo-nazis/#someones-lying

#1yrago Cigna's nopeinator https://pluralistic.net/2024/04/29/what-part-of-no/#dont-you-understand

[sybaritick]

One thing about me is that I am so committed to evidence-based, Bayesian reasoning about risk that I think it's a bad idea to spend disproportionate time warning about dangers that are very uncommon. People more concerned about serial killers than car accidents make bad decisions. Another fact about me is that a few years ago I had a statistically uncommon tick-borne illness, got sepsis, and quite literally could have died. Realistically, warning people about the serious dangers of tick bites would not be a good use of my time. There are far more neglected common risks. It's true that a tick bite could kill you, but very few people die that way. But my monkey brain, knowing my own experiences, prompts me to discuss it as a grave risk. If I do mention it, I always clarify it's very unlikely though... my commitment to the math runs deep 🙂‍↕️

[triviallytrue]

What is rationalism and why are some people adjacent to it or adjacent adjacent to it?

extremely abbreviated version: rationalism is a name given to an online group who are interested in both trying to think more rationally with bayesian reasoning and the potential impacts of artificial intelligence.

over time this group became something of a beacon for STEMy autistic types and people who like the social norms created by those types, but many people socially connected to rationalists were not particularly interested in bayesianism or AI, so the term "rationalist adjacent" came out as a way of saying "well i like the people but i'm not sold on the ideology." and then we entered on a treadmill where people kept adding more -adjs to rat-adj. and here we are

[max1461]

In most applications I think Bayesian reasoning is not that useful in everyday life, but I do think it would be pretty effective a strategy for like, planning the perfect murder. When people talk about committing the perfect murder they always come up with all this elaborate stuff like "oh we're gonna dissolve the body, blah blah", but the thing is, a ton of murderers don't get caught, and there are widely available statistics on every conceivable type of murder and just how likely it is to get solved. By going through all these statistics, you could not only plot a murder with a very low probability of getting caught, but get a quantitative estimate of just how likely you are to get caught and rationally weigh that against the expected pay off.

Someone should write a ratfic whodunit about this.