• or you will forget that you can, and your brain will go stupid

• make use of vertical space to reduce ground-level clutter¹

  • hanging (e.g. pegboard)
  • stacking (e.g. shelves)

²required reading: https://waitbutwhy.com/2013/10/why-procrastinators-procrastinate.html

¹(which may be signs of brain-stupid, or scheduler-processor imbalance) see digital-gardening

³see cave-carving

• Scheduler

  • the part of an operating system that decides what work goes first (if any)

• Processor

  • the part of a computer that does work

• Interrupt

  • something that snaps you out of a process

• Scheduler:Processor ratio must be delicately maintained

  • Processor-heavy imbalance: hard work in no direction - waste of energy

    • diagnostic example: if the first thing you do when you enter the garden is check emails, you are likely Processor-heavy
  • Scheduler-heavy imbalance: overengineering, no progress

• switch through emptiness

  • example:

    1. (Processor) finish reading enlightening article
    2. [Interrupt] goto empty (e.g. 'blank' screen/wallpaper/desktop)
    3. (Scheduler) decide what is important in your remaining life
    4. ( Sch:Pro ) garden accordingly

  • counter-example:

    1. (Processor) finish reading enlightening article
    2. (Processor) close tab without thinking
    3. (Processor) get sucked into mundane emails
    4. (Processor) live like this forever

• carefully maintain Interrupts and reactions

  • uncontrollable Interrupts

    • can be extremely upsetting, but the world usually does not care.
    • only way to make it less upsetting is to practice getting less upset.

  • controllable Interrupts

    • these are to be carefully and intentionally laid by the Scheduler
    • remove them with somewhat less caution
    • where do you make balance mistakes? Schedule an Interrupt there.
    • where do your important processes fail to complete? Remove Interrupts there.

    • Pro-!>Sch examples:

      • process complete -!> goto empty, become the Scheduler
      • body hurts -!> goto empty, become Scheduler, goto sleep

• choose a wallpaper image/animation so compelling that much of your time in the garden is spent watching it.

  • this will help break you out of the Processor back into the Scheduler when you inevitably lose yourself in the work
  • if you do not feel compelled to stare at your wallpaper/desktop background image frequently, your background is not compelling enough and will lead to poor (Processor-heavy) gardening

  • iOS example:

    • move all apps one screen to the right
    • switch through screen 1 (compelling wallpaper)

    • goto empty before locking device

      • (means you will arrive at empty when opening device)

    • over time, should lead to Scheduler-first mobile-gardening

      • can take years to grow, given depth of necessary dopamine reprogramming

• song change -?> consider goto empty

• Feudalism, but with freedom¹

• different from Federation

• usually requires disembodiment or near-disembodiment in practice, or else coercion reigns.

  • e.g. online 'kingdoms' with pretty good privacy
  • if your physical body can be deanonymized and accessed, you can be coerced through threat of violence etc.
  • therefore you are no longer in an environment of 'Elective' Feudalism - just plain old feudalism :(

• difficulty of creating a new kingdom must be reasonably low

  • otherwise, centralization occurs and corruption follows.

Under such circumstances, Elective Feudalism may better serve its participants than more easily corrupted systems like pop 'federation'²

1. consider the smallest possible digital kingdom: one person (you)

   • you live in a benevolent dicatorship (or else you are insane)⁶
   • note that at such quantum scale, Kof1 is quite similar to benevolent nongovernance, the (theoretical) optimum.

2. your kingdom of 1 elects a benevolent dicator named "Dick"

   • becomes kingdom of 2
   • one king (Dick), one baron (you)

3. Dick becomes non-benevolent

   • "we decided to sell your data after all"

4. You leave (elective feudalism)

   • if you stay in Kingdom Dick against your will, that's just plain old feudalism

5. GOTO 1)

   • become kingdom of 1 (therefore benevolent)
   • keep an eye out for good leaders to follow. Or don't, you're in charge.

¹Usually I write uppercase Freedom^TM with a cheeky trademark superscript, but in this case I actualy mean it. The Pirated version of Freedom is better.

²(not to be confused with the original Iroqois model, which is very cool and frankly should be above criticism by the likes of me even if it weren't)

³the first optimal model of governance is benevolent 'nongovernance' - or everybody just doing the right thing. It is unclear which is more likely - benevolent governance or benevolent nongovernance.

⁴https://pluralistic.net/2022/04/17/revenge-of-the-chickenized-reverse-centaurs/

⁵don't listen to any billionaires that tell you to come live in their utopian crypto village subject to their benevolent rule. And definitely don't bring your kids.

⁶I've certainly spent some time in a non-benevolent kingdom of 1. Find a new kingdom.

• turns out, if you focus on one thing and one thing only, it starts a process that ends with 'tripping balls' as they say. No drugs required.

  • many terms for this across many traditions:

    • the 'psychedelic', 'transcendental', or 'religious' experience
    • bliss, ecstasy, superconsciousness
    • nonduality, Nirvana, entering the Dao

• Choose whatever term or tradition motivates you - it's all the same goal. For me, I like 'tripping balls.'

• this Focus is most difficult

• three phases of meditation in the literature

  • Dharana: concentration
  • Dhyana: meditative state
  • Samadhi: ???
  • I can't really speak on Dhyana or Samadhi, so these notes will explore Dharana and one guy's half-baked method of reliably getting there.
  • my understanding is that if you get good enough at Dharana, Dhyana and Samadhi will come with continued practice. But that's above my paygrade

• deep, life-improving benefits are manifold, but mostly it's just fun. (fun, but damned difficult and frustrating to get there.)

• this does not please the Ape

  • much as the Human would like to think it's in charge, ultimately the Ape steers the long-term ship, one short-term decision at a time

• therefore, in getting to the fun part (past the part that feels like Work, which no Ape will abide without proper bribery)… one must first negotiate with the monkey.

1. know what you want
2. find what they want
3. reliably offer enough of what they want, link it to what you want
4. behave appropriately (e.g. polite, meek, scary, provocative, etc.) so as not to ruin the deal in the long run

• For me:

  • I've come to terms with the fact that deep down I am just a Junkie

    • once a junkie knows a junk, the full focus of life becomes hitched to attaining that high
    • this is an enormously powerful force, that usually ends in destruction
    • if carefully respected and balanced, however, this force can be harnessed and redirected constructively
  • word on the street (for the past few millennia) is that meditation brings the greatest high
  • I want.

• For you:

  • will take some introspection, I cannot possibly know what motivates you

• in this case (and in almost all cases) what I want as the Human is just a prefrontal convolution of what the Ape has wanted all along: instant and never-ending gratification

  • if the gratification must end, then (as a hack) let me forget the concept of time

  • if you cannot remove all non-gratifying activities, then the Ape will press to bring gratification into previously non-gratifying activities.

    • This can be dangerous (cocaine at work) or deeply special (Zen dish-cleaning)

• all our lofty ideals and Sapien thoughtforms are supposedly just expressions of desires

• desires are procured and managed by the Ape, with scant oversight

• 'motivated and stupid' is, however, the most dangerous of all combinations

• left untrained, the Ape (lacking the faculties for long-term planning) will simply do the exact last thing that got it high - over and over with no vision for finer herb

  • for some that's alcohol or narcotics, for others it's combat sports, many get there through psychedelics, and most of us dope digitally

  • if the Human is inattentive and the Ape is un-gratified, it will sneak into the control room and hijack faculties of higher consciousness

    • example: "I will work my ass off this evening so tomorrow I can wake up and get shit-high without anybody calling me"
    • example: "I only seem to feel comfortable getting black-out drunk with my college friends…I will plan a vacation (using opposable thumbs and capacity for written language) so that I can comfortably get black-out drunk again"
    • example: "My mind is a terrifying horrorscape at all times, except when someone is trying to kill me (and failing). I will dedicate years to the practice of Martial Art so that I can experience minutes of blissful peace in a cage"

• therefore, careful guidance (and control monitoring) are required.

  • Mostly though - if you keep your Simian co-captain chemically balanced it will stay out of the bridge, happily stoned in the engine room.

• Problem is, 'chemically balanced' individuals seem to be our civilization's least common product.

• the Ape will never understand what it has not experienced

  • no use telling a monkey about meditation
  • very hard to do bribery if reward is incomprehensible or unbelievable

• If the Ape does not believe, it will continue to 'smoke bad dope' with predictable outcomes

  • must give the reward first, then can use the reward as a bribe in the future
  • but reward is very difficult to find without sufficient motive: chicken-egg problem

  • most practical way to break this deadlock is through psychedelics

    • psylocibin mushrooms

      • with proper research and dosage, or in-person guidance
      • most reliable way to have meditation 'done to you'
      • not to be trifled with

    • full-spectrum live resin cannabis extract, open source ("good weed")

      • cannot yet reccommend

    • full-spectrum Non-decarboxylated raw hemp extract ("CBD")

      • reccommended under very specific conditions, more on this later

• OF UTMOST IMPORTANCE: the reward is not the drug. the reward is the meditative state, learned with the assistance of medicine.

• Drug vs. Medicine

  • Drug: used for its own purpose, instant gratification, long-term damaging, dose is stable or increasing over time.
  • Medicine: used to help you improve your long-term health, dose eventually reduced to zero.
  • anything can be either. Be a grown up and recognize which you're taking.

• Be realistic: will your body and your mind agree to just sit still and Focus on your diaphragm indefinitely without any guaranteed reward?

  • if you have not done this regularly and recently, the answer appears to be no.
  • you can fix this by guaranteeing the minimum reward necessary to produce the desired behaviour

• suggested minimal reward: 25mg BlueBird Botanicals 'Complete' CBD, taken exclusively for the purpose of producing and maintaining Dharana.

  • After two years of meticulous dosage tracking and recording of self-experimental results, I cannot confidently reccommend any other use case for CBD. I also cannot confidently reccommend dosage, but much of the literature surrounding autistic CBD dosage seems to fall in the range of several hundred milligrams.¹ Intuitively, I expect the minimum reward for other people to fall somewhere well above the colloquial 10mg standard, but only you know your monkey.

• start small (but not too small) and if the minimum reward does not produce desired behaviour (Dharana), it is implicitly not the minimum reward and can be doubled as part of the calibration process. Remember, one must negotiate with the monkey. Your behaviour (in retrospect) is the ultimate measuring stick.

• carefully harness habit-forming properties of cannabis

  • when motivation to meditate is lacking, motivation to 'meditate on drugs' will likely remain (once it has been experienced).
  • use the substance, carefully linked, to help form the habit
  • slowly phase out the substance as success comes with smaller Ape-bribes (promised dosage)
  • Dharana will become easier to attain without chemical assistance
  • Ape buys into the plan, fires up the engine room in pursuit of higher gratification
  • only the habit remains
  • substance proves to be medicinal

¹https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6333745/

Armies that decentralize the task of intra-war learning and centralize the execution of intra-war organizational change can be observed to adapt more effectively to counterinsurgency

– James Ondaatje, "WAGING POLITICS: COMMAND AND CONTROL IN INTRA-WAR ADAPTATION TO COUNTERINSURGENCY"

1. Decentralize Learning
2. Centralize Execution

Principle: The "Mythical Man-Month"

What one programmer can do in one month, two programmers can do in two months.

– Fred Brooks

1. Collaboration is incredibly difficult to get right (see quote)
2. Collaboration is worth getting right (see most of human progress)

The main goal of Org Learning. Engenders Deep Focus and Executive Function

The pursuit of Artificial Intelligence is perhaps our greatest undertaking. Creating a sentient being would be to unlock the secret of life itself - but how can you know if someone (or some thing) is conscious? This is the question that Alan Turing examined in his seminal "Imitation Game" Gedankenexperiment. Since re-branded as "the Turing Test", the game consists of an Inquisitor and a Subject. If the Subject can successfully convince his interrogator that he is human, he wins. It is Turing's claim that if a machine were to pass such a test, it would be sentient.

This claim has predictably spawned intense debate over whether a Turing Test-passing machine would truly be "thinking," "intelligent," or "conscious," as well as dissent over what those words really mean. Popular arguments for and against the Test as a sufficient identifier of intelligence include Ned Block's "Blockhead" argument (against), and Shieber's "Accessible Universe" counter. I will expound on these more in the following section, but for now it is enough to say that Block believes a "dumb" program with only the capability to perform dictionary lookups could pass the Turing test. All it would need is a dictionary with a human answer to any question. Shieber claims that such a data structure would be impossibly large.

There is truth in both arguments. However, I also believe that each has its faults. In this project, I hope to present a compromise between the two - by showing how a "Blockhead" may be able pass the test without a prohibitively large dictionary, while still keeping the hope alive that the Turing Test is some measure of intelligence.

It is my belief that a machine need not know the answer to every possible question in a Turing Test - just the ones that humans would ask. But how to know what questions humans would ask, and how they should be answered? Simply play the Imitation Game with them, and learn as you go. I don't think we humans are so uniquely creative as to never repeat questions (or at least similar questions) between us - which means that the set would eventually converge. I will go into the details of the "learning" that takes place, and the methods through which that knowledge is acquired in a later section. What's important is the high-level facts: such a machine will be building a "Blockhead" dictionary from the combined learning of many human interactions. Eventually, that dictionary may be able to pass the Turing Test on its own, without the learning intermediary. To that end, my project is a generator for a finite Blockhead. And if it's learned a bit of what it means to be human from each of us (it takes a village) then we can perhaps reconsider both Shieber and Block's arguments.

The core problem I am solving is the one presented by the intersection of Block's and Shieber's arguments. If I can show that a finite blockhead could potentially pass the test, then that could be a good counter to Shieber's "Accessible Universe" theory. Likewise, if the way that dictionary is built is through large amounts of human interaction, perhaps that can call into question whether the dictionary is totally "dumb," or if it represents some mapping of collective human thought.

More formally, I've set out to prove by induction that such a dictionary is possible. If I can create an AI that learns from human conversation such that every interaction makes it a bit better at surviving the Turing Test, then logically it should eventually be able to pass at least the finite restrictions set forward in the original specification (No more than 70% chance of being identified after a 5 minute test). (Note: there is definitely a counter-point to be made here about exponential blowup in a conversation model, but more on that later)

To do this, I must show a base case and an inductive step. The base case is simply showing that my machine can learn to respond to a question. The inductive step is showing that if it can learn to respond to some number of questions, it can learn to respond to one more question, and thereby can respond to any question. A key assumption here is that there is a countable number of questions - as I mentioned earlier this is a core belief of mine: that given enough training time it is highly unlikely that most questions wouldn't be repeated (especially with a vectorizer that maps similar questions together - though perhaps that breaks the Blockhead property).

The below gives an overview of the system and a description of its interfaces:

def best_response(dictionary):
    if settings.STOCHASTICITY:
        responses = Counter()
        for response, data in dictionary.iteritems():
            p_fail = data["failures"] / data["frequency"]
            responses[response] = 1 - p_fail
        responses.normalize()
        return sample(responses)
    else:
        best, best_num = None, float('inf')
        for response, data in dictionary.iteritems():
            p_fail = data["failures"] / data["frequency"]
            if p_fail < best_num:
                best_num, best = p_fail, response
        return best

best_convo = random.choice(conversations)
for conversation in conversations:
    sim = cosine_sim(conversation.question, question)
    if best_similarity < sim:
        best_similarity, best_convo = sim, conversation

Conversation.objects.create(question=question,
                            responses=best_convo.responses)

While I knew exactly what was going on, it was still a bit of a shock when the system I'd built started talking to me. The prime experiment I ran was devised to prove it could learn. The test would go as follows:

1. Ask a question. Get a response.
2. If I thought it was something a human would say, I continued.
3. If I did not, I indicated that I thought I was talking to a machine.
4. Then, as I continued using the site, I would see the AI asking me questions.
5. I would answer them naturally.
6. Then I would ask my questions again, and I would see a good human response.

Of course, I knew that those were just my own responses being parroted back to me, but once I got some friends using it I was truly surprised sometimes when it felt like I was talking to an (occasionally obnoxious) human. When I implemented stochasticity and cross-pollination, this process sped up dramatically. Soon, we couldn't help but refer to the machine as "Alan."

On a more technical level, I created a view into the backend at turingtest.io/api/model, where I could verify that specific beliefs were being updated. So for instance, I could set up a human-human conversation and watch as each question-response was added to the model. I could also watch the model's perception of p(success) for each response that was rewarded or punished. I'm a big believer in backend visualization as a development tool (and just being neat in general), and it was a big help for this project.

It would take a long time and plenty of human interaction for Alan to have a chance at passing the Turing Test, and even longer for the Blockhead he's generating to do so. However, I believe the system I've created shows how it could be done in finite space. The above experiments are the core of the induction proof - it learns. Furthermore, because there are a finite number of questions that humans will ask, the database can exist. And if the system is always getting better at human conversation, then it will eventually surpass the finite bounds of the Turing Test.

Something I find interesting about Block's story about Jones the Dumb Grandmaster is that he doesn't consider the intelligence of Jones' two opponents. While Jones most certainly isn't intelligent, the chess players he is getting his moves from certainly are. Similarly, when a ton of human thought has been condensed into this database, the program that does the dictionary lookup is Jones, and the database is the grandmaster. This feels somewhat counterintuitive - a database surely can't be sentient, can it? But perhaps it possesses a sort of passive consciousness that can only be reinvigorated by an acting program - much like a legally dead person can occasionally be brought back to life, the program is that "spark" that revives the dead brain and produces something with consciousness and thought.

I don't know how long it would take for this system to get close to passing the Turing Test, or how much human interaction it needs. I do believe, however, that this project presents a good counter to both Block and Shieber's arguments. I still do not know whether the Turing Test is sufficient for intelligence, but when the time comes to spin Alan down I think the real measure of sentience will become clear - will I be able to pull the plug?

• [1] Ned Block. Psychologism and behaviorism. Philosophical Review, 90(1):5-43, 1981.
• [2] John R. Searle. Minds, brains, and programs. Behavioral and Brain Sciences, 3:417-424, 1980.
• [3] Stuart M Shieber. There can be no turing-test-passing memorizing machines. Philosophers' Imprint, 14(16):1-13, 2014.
• [4] Alan M Turing. Computing machinery and intelligence. Mind, 59(236):433-460, 1950.

Code: github.com/condaatje/turingtest.io

1. install the correct linux headers (if not already present)

   # select the correct version for your kernel,
   # e.g. linux66-headers-?? for 6.6.32-1-MANJARO
   yay linux-headers

   note: if there are multiple options, you can find the right one like so:

   # get your kernel version and match it to the correct package name
   uname -r

   6.14.4-arch1-2
   

2. install custom driver

   git clone https://github.com/lawstorant/hid-logitech-hidpp.git
   cd hid-logitech-hidpp
   sudo make install

1. install Berarma's Oversteer GUI tool from the AUR

   yay oversteer

   • https://github.com/berarma/oversteer

1. test the wheel for proper force feedback (FFB)

   • run oversteer and select the 'Tests' tab, then 'Start New' and follow the instructions
   • your results should look somewhat like the following:

• Valve Index
• AMD GPU

• Wayland (but not Gnome)

  • I recommend starting with default Endeavoros (KDE Plasma): https://endeavouros.com/

Don't use flatpak or snap, they aren't ready yet.

yay steam

At this point, Steam should pick up your headset and lighthouse and the rest is guided. Small deviations from this setup may produce complex issues - fair warning.

gitlab.com/christian-ondaatje/public/site/video/stack-demo.mp4

gitlab thread

gitlab.com/christian-ondaatje/public/site/video/k8s-controller-demo.mp4

gitlab thread