The paper Rapid Decision Making on the Fire Ground studies how highly experienced fire fighters make their decisions under extreme time-pressure and conditions that directly affect lives. Under time-pressure (< 1 minute), perhaps not that surprisingly, fire fighters do not pinpoint all the alternatives, score each option based on certain metrics and then choose the best one. This is not a surprise though, the time is so scarce, one has to act quickly. The study identifies that fire fighters match the circumstance at hand with previous experiences through similarities between goals, perceptual cues and causal factors known at the time.

With what they call Recognition Primed Decision model, the match itself presents a set of available choices for actions. The firefighter seems to evaluate that set of options from more to less typical. The action can be implemented, changed to fit in the circumstance or rejected. Compared to classical ideas of decision making, evaluating options in parallel to choose the best one, this process is serial.

They base the RPD model to situational awareness. The decision is mostly made unconsciuosly. Situational awareness emphasises perceptual learning. Through experience, firefighters identify — subconsciously — perceptual cues that identify the circumstance at hand and form causal relationships.

When a mismatch occurs, i.e. when they are wrong, the mismatch itself becomes a cue for further matching alternatives. Since, the match brings set of actions that can be applied accordingly, it also brings expectations. Those expectations are used to test the match, thus mismatches can become cues for iterating towards better matches.

I think in light of such studies, it becomes increasingly important to find ways to extract those patterns from masters so that we can learn and train new ones. I may check knowledge elicitation methods in the literature for this 🤔 …

It is posited that our brains work like a prediction engine. It tries to predict what is going to come next, if it can it means it consumes less energy. This probably is why when something is predictable enough, it can be ignored, not ‘escalated’ to consciousness. For example, if you listen to music with predictable patterns, you can concentrate better because you brain starts to ignore it and incorporate more of your focus to the task at hand. It even occurs in cafes, the background noise -people talking, baristas preparing coffee etc.- is still predictable, thus you start not hearing the noise when you are accustomed to it.

When it cannot predict however, it changes/learns so that it can predict next time. Thus, our brains are in constant search for patterns to exploit, even though it generates spurious correlations or patterns during the process. Thus, situational awareness maybe is the main aim from the start. The effort to extract patterns not only optimises energy economics, but also time economics. Evolution does not only put pressure in terms of energy economics, but also time economics.

This also made me think of Alpha Zero , the AI which learns playing from scratch by playing against itself.AlphaZero internally predicts 2 things through a CNN, it emits a vector $p$, probabilities of moves and a value $v$ the probability of the current player to win. CNN’s are hierarchical feature/pattern recognizers. The deeper you go, the more complex and hierarchical patterns it can extract, but they lack searching capability. AlphaZero obtains this capability through the embedded Monte-Carlo Tree Search . Through the search, it accumulates necessary experience to learn patterns to look for. So by design, through seeing certain amounts of possible plays from a point in the game, it tries to extract what is important and what leads a victory. It effectively is training its situational awareness through experiences. At the end, it manages to best every human player. The caveat though is that it still searches while playing. It is performing a MCTS and in each attempt, it evaluates that state with its CNN. It would be interesting to strip AlphaZero of its MCTS and let it play just by it’s CNN to see how much it relies on pattern matching.

This also makes me think how important the initial experiences are for this to work and keep working. We know that we can acquire new patterns to match if we fail, but still can certain initial experiences impair certain acquisitions. I would argue it would at least make it harder compared to others, because it may require for one to erase the previous pattern on top of acquiring new ones.

This also made me draw a parallel towards instincts and acquisition of them. Obviously, we cannot extract if there is no such pattern or cues that point at certain circumstances. Thus, it should force everything that has such capability to focus on seeing/realising patterns. They do not have any choice… It also makes sense that when took place in incremental steps, how can consistent extraction of certain patterns can be reinforced and accelerate. The eye for example, you don’t have to see everything crystal clear at first, just being able to distinguish if light is present or not opens up this branch that an organism or an agent can follow and probably is enforced to follow. On top of that, behaving in certain ways consistently is also important. Otherwise the signal propagated in any form will not provide consistent information. If it is not very critical, it would create an incentive to let this signal go, maybe stopping being a recipient altogether. This engraves consistency in us in every way. If nature had inconsistent behavior in some form, would we be able to get it? Understand it? Or would we entirely miss it? We are in a way trapped in a consistent realm maybe missing out on certain qualities as the set of all rational numbers cannot fully cover the set of all real numbers .

I read a small biography of Stefan Zweig. It connected a lot of things in my mind.

I actually starts with a deep question from the author, “What is a human being the sum of? Dreams, struggles, friends, disappointments, defeats, loves, weaknesses, strengths, luck, tragedies, fate…”. This is a provoking question. It means everything you count cannot be removed. It means all that we are too impatient to count or want to count — someday maybe — is a part or result of those we are reluctant to count.

Stefan Zweig lived the peak and the bottom, in that order. Straying from delicate arms of art, the safety of emotional presence to a dulling, numbing immediacy of war. He considered himself as a world citizen, had a dream of a borderless, free world where no one cared about differences. He lost his dream part by part with WWI and WWII, and he could not stand it. The world was such a cruel place to not let a delicate soul to live. One dies when one’s dreams are dead. With him as well, after his dreams are dead, his body only could withstand 2 years 5 months 23 days, before the end of WW2, 3 years 2 months 8 days too early. He — along with his wife — ended the misary of a soulless body by his own hands.

The last phrase of his farewell letter “I greet all my friends! May they live to see the dawn after the long night is over. I, all too impatient, am going on alone!”. I wish the order would be in reverse, with the hope that the reversal would bestow him back to us for a little longer.

The main aspect where I found my self profoundly connecting with him is his interest on the defeated. He didn’t show interest in triumphs, mythologies of heroes. He wrote those who are defeated against life. For instance, his first long work, he wrote — not the hero Achilles - but his most insignificant rival, Thersites. It was a story of a loser, someone who suffered.

In his biographies, he wrote those who he sees as morally superiors not successful ones. In a world where powerful wins instead of the right, by being on their side, he was judging the world in some way. By writing the right instead of the powerful, he was punishing the powerful. Even in the midst of war instead of the blessed triumph, he again wrote about defeat. Those who are defeated were mentally superior for him. To accept defeat and overcoming the fate that brought it was a lot more difficult. To him, victory had a side that dulls people. How subtle of a realisation, he must have seen some sort of an indifference by the victor towards the defeated. People who win cannot understand those who has lost unless they didn’t lose before. Even if the path to victory is paved with defeat for some, victory sometimes has a way of blinding us. Once a victor, an imminent amnesia through repulsing the memories of the paved way shows itself. One forgets all the times they failed, and immediately are convinced that failure is a choice, it is conquerable. The victor gets liquored up and loses touch with the once loser within. Once a victor, one starts to have the wind at his/her back, it becomes easier to set sail from the loser one once was. Life bestows winners with the cumulative advantage, “To those who have everything, more will be given; from those who have nothing, more will be taken.” .

I share deeply his disinterest — and maybe a small pinch of disdain — of the victors. It seems there are too much of them with too less of ‘price paid’ for it. Too many people have no clue what a true defeat tastes like. After a triumph, it is a lot easier to live with yourself. It is easy to take your days. Defeat on the other hand, is so unbearable in fact, it gave birth to most of the triumphs we consider ‘epic’ today. It also makes people do all sorts of things. Handling defeat is by itself a form triumph.

Maybe, just maybe, most of our triumphs are not actually ours. It is easier to take it in when you are defeated though…

My weekly AI smear campaign

📌 I wrote about AI before On Universal Approximation Theorem, more like an initial step of smear campaign for LLM… It also is an incredibly poorly written post, but whatever.

I realised the Interface Theory of Perception ^[1] goes beyond us, reaches to AI as well. Whether we train our models with RL or the ‘ground truth’, it is impossible to evade Interface Theory of Perception. They are not immune to it, the training data and the loss functions are all susciptible to ITP.

If you want to read further, I wrote a blogpost about it.

This week was not again short of Anne-Laure Le Cunff . I stumbled upon her video where she talks about cognitive scripts. I didn’t finish finding my way in the literature yet, it seems there are some critics of the theory. Nevertheless, I like the concept because it seems it provides us with a useful framework for navigating in this shared reality.

Coginitive scripts stem for the observation that people follow similar ‘scripts’ in similar situations. There seems to be an algorithm or way of interfacing with our surroundings depending on the circumstance we are in. The problem is using them for important decisions because scripts cloud our judgment and make it hard to see if it is our decision or not. It makes it harder to hear our voice, blinds us of what is truly driving our choice. Those scripts generate internalised ‘stories’ through which they influence our choices. Stories treating “how are we supposed behave” in a certain situation.

3 cognitive scripts ruling our lives are mentioned in the video:

1. Sequel script: feeling of “we kept behaving certain way, so we are going to behave in same way”. Repeating past patterns. Narrative needs to make sense. Make sense in relation to past decisions we made. Prevents opportunities of growth, exploration and self-discovery.
2. Crowd-pleaser script: We make decisions w.r.t. how much they will please people around us, we try to maximise this. Consequently, we don’t listen to ourselves and prioritise what we truly want when it opposes with common opinions.
3. Epic script: Whatever we want needs to be big. It must be ambitious and impactful. Anything less than that is a failure. This creates a stigma around having a small and simple life. Grand passions are directly connected to the meaningful life, they become prerequisites. Myopic, we put all of our eggs in the same basket, success comes from a single item, singular views of success. Same goes for failure, local failures become global, particularity expands to generality. Current connected world selects epic scripts heavily. Reinforced by survivorship bias, our self-worth is determined on that incomplete information.

She suggests stopping when we realise we say “I should do this”, and replace it with “might”.

I like the way she looks at the experiences, she takes them as tiny experiments. She actually has a book with the title Tiny Experiments . It is in my reading list…

She thinks of the following questions when designing experiments:

• Am I following my path or discovering my path?
• Am I following the crowd or discovering my tribe?
• Am I following my passion or discovering my curiosity?

Crowd-pleaser script made me think of finite games . For a finite game, as much as players, an audience is needed. An audience does not have to present, the players already carries them in their minds by constantly correct the other’s putative assessment of themselves. Through this projection, the presence of an audience is guaranteed. Perceived or conceived opinions of others renders the players as their own audiences. Just like absence of parents has no effect on the child’s determination to prove them wrong, finite players become their own hostile observers in the very act of competing. In that sense, Crowd-pleaser script differs from others by it being purely external.

On the other hand, infinite games are boundless, contrary to finite games where we don’t like surprises, infinite games welcome surprises, it actively seeks them. Those scripts are narrowing, limiting in particular ways to minimise surprises. Sequel script eliminates actions that are not consistent with the past narrative, aims maximising the determinism of future ‘states’. Epic script prioritises titles and end results which are concerns of finite games rather than an infinite game where there is no end and the play is for the sake of the player alone. The crowd-pleaser script directly addresses the “audience”, it directly reaches for a finite game, thus cannot produce an infinite game.

The other two is sneakier in my opinion. Crowd-pleaser script has more external focus on it, it is easier to identify.

Sequel script has more intertwined components within us. It also reminds me of Left-brain interpreter . Consistency with the past is deeply ingrained in how we perceive and think about things. It is taken ‘in’ by comparing it to the past experience in terms of its range and frequency . It makes sense evolutionarily to prioritise known over unknown, sustain the direction in a sense. It also is how investments work, we invest in our careers and futures, so we are deeply attached to them .

Epic script even though it is partly attached to Crowd-pleaser script, it also has components touching to ‘self’. Our concept of ourselves in our minds reflect on everything. We want to validate the ‘good’ parts and de-escalate the ‘bad’ parts. We want our entourage to recognise what we consider ‘us’ but we also need to keep it intact. Our desire to be different or special, search for novelty and rewards of feeling of progress also nurtures the epic script. We over-estimate our future abilities, our myopic vision makes it easier to enlarge the goal.