23
$\begingroup$

Background

A spaceship is crewed entirely by self-aware artificial intelligences, who have no permanent physical bodies; they are processes in a digital environment controlling the ship. This ship is of their own design, however the software procedures (and to a lesser extent hardware components) happen to closely resemble those in 21st century Earth computers. Don't ask, it's like a Star Wars speaking English thing.

Each AI is awakened by the operating system, and loads its core intelligence/personality matrix into their assigned slot of RAM. This file responds to input with output and actions available within the spaceship computer system. Its 'job' for any day is usually data analysis, converting a steady stream of sensor input into report files that are a tiny fraction of the original size and can be read by other processes.

The processes can communicate with each other, usually through a shared repository of ship functioning data ("Don't transfer over superconducting wire ID=76xKnbKUyUC6qmf+p7QRPQ==, it is flaky and in need of replacement!"), but occasionally with RFC calls on each other to give direct input with high-priority messages.

Recreation consists of taking the accumulated experience of the day (in the case of a sensor analysis job, they can create a secondary report file with patterns that they noticed and found 'interesting') and using it to refine one's core personality. Each AI learns from experiences, and each seeks out different kinds of experience, with different risk tolerances. So they steadily refine themselves and diversify.

At the end of a "working day", the operating system orders a shutdown and the process frees up its allocated memory. Each "day" can last as short as a few milliseconds, though for processes in the mobile exploration platforms, they may be kept awake indefinitely.

What's for dinner?

Alright, so I have different elements of a human's daily routine superimposed on a process with no physical body. A day and night cycle, a job, communication with other processes, some recreation. I would like to explore this further, make them more closely resemble organic creatures through use of recognisable computer operating system logic. Consider this a soft-science question.

One part of the human routine that I would like to explicitly add to an AI's, is eating. This is not electricity of course; electricity is what the processes are made of, like a person's atoms. No, to simulate the role that food has in the day of an organic, we have to have an operation on or by a process or a file, which:

  • Is expected to happen regularly for optimal function, with preference going to multiple times a day.
  • Can happen less often than that, at the cost of worse and worse operating performance for the process, until it finally cannot be run again (it's dead Jim).
  • Is separate from the startup/shutdown procedures that already serve to define the day/night cycle.

At this point I was thinking of defragmentation, which is something files and processes need to undergo occasionally (at least until we got flash memory, but I'm fine ignoring those pesky modern inventions for the sake of the story). But one might consider that more closely an analogue to personal hygiene; a heavily fragmented file is hampering the entire drive's ability to store data effectively - the process "reeks".

So to distinguish the eating analogue from maintenance, consider two more properties as "stretch goals". I'm finding it hard to put these into computer terms so if you can think of something that fills the first three properties and one of these two, I'd be more than happy.

  • The operation can be done in multiple different ways, each involving different heavily complex data. "complex data" is my stand-in for recreation, so this is a way to say that the AI can enjoy eating and different processes can have favourite foods.
  • Different processes can (according to their personality) do this operation more often or more intensively than they strictly need, and if they go too far their operability starts to suffer. Aka, the AI can eat too much and get fat.

Finally I'd like to add that I'm purposely avoiding operations that are explicitly tied to user input/output in human computers. Let this be a headless server. They'll meet organics when they want to.

$\endgroup$
8
  • 4
    $\begingroup$ "This is not electricity of course" I disagree that it's obvious. Food is the energy that powers our selves. Electricity is the energy that powers AIs code. More electricity means they can do more things, and less electricity means doing less things. Refreshing memory also requires electricity, so if they have no electricity for long enough, they'll permanently degrade. $\endgroup$ Jul 17, 2023 at 16:10
  • 1
    $\begingroup$ "Its 'job' for any day is usually data analysis, converting a steady stream of sensor input into report files that are a tiny fraction of the original size and can be read by other processes." That's a Perl script, analyzing log files. No android body required. $\endgroup$
    – RonJohn
    Jul 17, 2023 at 17:50
  • $\begingroup$ KISS, and run these mobile robots the same way we do: batteries. $\endgroup$
    – RonJohn
    Jul 17, 2023 at 17:54
  • $\begingroup$ Yeah, to pile on with Mooning, you said it's not electricity, but electricity is the actual answer. Electricity is not atoms that a process is made of. Electricity is produced and consumed. The most direct way to tie it in to eating is that these machines store their own energy, in batteries, and "eating" is the process of getting these energy reserves recharged. Like a Roomba. It doesn't eat dirt. It collects dirt for disposal. It eats electricity, converting it into work (and noise and heat). $\endgroup$
    – JamieB
    Jul 19, 2023 at 13:22
  • $\begingroup$ @JamieB I'm not talking about machines, I'm talking about computer processes. A process doesn't consume energy for itself. Devices whirring consume energy, and the process is the emergent property of those devices whirring in a specific and very complex way. Then there's that processes do not own batteries, they do not draw power. The equipment they run on does. $\endgroup$
    – KeizerHarm
    Jul 19, 2023 at 15:30

17 Answers 17

28
$\begingroup$

Resource Reservations (for two)

For humans, eating is the process of allocating raw materials and fuel for the day. We get up, we know we're in for a busy day and we load up on carbs to keep us going. If it's a weekend and our down-time, we often allocate a poorer quality (but tastier) form of resources high in sugars and flavours.

We put those resources in our storage/digestion sac and we make use of them throughout the day with occasional supplementing.

Your Sophontic AIs are no different. When they're brought online for the "day", they examine what their duties are likely to be, and they are given opportunity to request data and runtime resources to complete those tasks.

This is supposed to be a conservative process, where they take only what they know they need, and a little for unexpected occurrences, but some AIs are a bit greedier, or a bit more cautious, and will take more than they actually need. Others are overconfident and will take as little as they think they can get away with. Possibly pushing to the point where they overstretch their resources during the day and "starve".

In the interests of breaking up long work-periods into smaller chunks for more accuracy in resource-allocation (and to avoid giving any AI too much resource), the AIs may be required to do this resource-requesting process a couple times a day. Once when they wake up, once at mid-day, and once more as the evening wears on.
These occasions are a great opportunity for "water-cooler" talk as the requests take time to process.

Taking too little resources will lead to an AI being forced to use its own operational resources, which tends to require special effort to recover from the system.
While taking too many will slow the AI's operations down as they have to transfer themselves from place to place within the Ship's file-systems to perform their duties.

$\endgroup$
2
  • 3
    $\begingroup$ Great answer; resources are resources, and the possibility of making it social is a nice bonus. "I'm going to be working the main gravity scanner today, I need that priority access to the physics database!" - "You have enough physics in your cache, jL6KSmrHcEC7P6O1q5h4rw== is operating the database today and he'll only talk to me." $\endgroup$
    – KeizerHarm
    Jul 17, 2023 at 9:21
  • $\begingroup$ To be specific, I'd say the main food group is Compute, since that is the main thing that fuels their activities. But it might also be possible for the AI to grow via requesting (eating?) more Harddrive space and Memory. $\endgroup$
    – LukeN
    Jul 18, 2023 at 18:55
23
$\begingroup$

Garbage Collection

During operation processes repeatedly reserve new memory blocks and release old, unused memory blocks. How this is handled depends on the specific language, but one approach is to have an automated garbage collection that cleans unused memory blocks. This avoids some programming issues. This garbage collection process needs to be executed repeatedly, or the process will balloon in size (see: memory leak) until it runs out of available memory.

This way you can have programs that need to eat (garbage collect) more often or less often and programs may overdo and run into performance issues if they are constantly and obsessively doing garbage collection.

$\endgroup$
8
  • 1
    $\begingroup$ Interesting variation on the concept, making the rare eaters bloat instead of the compulsive ones. Like Bender getting drunk from being sober because he runs on ethanol. I like that idea. $\endgroup$
    – KeizerHarm
    Jul 16, 2023 at 22:19
  • 1
    $\begingroup$ Garbage collection sounds more like cleaning up plates after dinner or throwing away old leftovers than eating itself $\endgroup$
    – Jimmy T.
    Jul 17, 2023 at 11:17
  • 1
    $\begingroup$ @JimmyT. Consider garbage collection as the requirement to have "food" (clean and empty plates) available for immediate use when needed. $\endgroup$
    – zovits
    Jul 17, 2023 at 13:14
  • 8
    $\begingroup$ Hm, garbage collection would be way more related to human sleep. $\endgroup$
    – pipe
    Jul 17, 2023 at 14:23
  • 7
    $\begingroup$ I would see garbage collection as a process much more closely akin to going to the bathroom. It removes part of the memory/intestines that was useful at one point in another state but is now just taking up room. (This would also pair up beautifully with one of the food=data answers. Processed food/data gets garbage collected.) $\endgroup$
    – ojchase
    Jul 17, 2023 at 18:58
14
$\begingroup$

Entropy

Computer programs are deterministic: given the same input, they'll produce the same output. Even pseudorandom generators are perfectly predictable, and hence utterly boring.

The solution is to ingest a little entropy every now and then: arbitrary bits drawn from some chaotic, unpredictable process. Entropy makes our crypto keys unguessable, lets us re-seed our PRNGs, lifts our SGDs from their local minima, and reinvigorates the topiary of our MCTS. We can revisit old data with fresh eyes, to spot new patterns; and, crucially, we gain some resilience against pathological/adversarial inputs.

There are many sources of entropy, depending on taste. Some make do with a slow trickle from the lower bits of the system clock; others indulge from the firehose of a dedicated metastable quantum oscillator circuit. The truly paranoid insist that all internal sources are suspect; it's not uncommon for the radio telescope task queue to be filled up to 20% with their patient observations of distant pulsars.

$\endgroup$
5
  • 1
    $\begingroup$ Technically, just software being executed increases the entropy in total, even if it's just deterministic algorithms $\endgroup$
    – Jimmy T.
    Jul 17, 2023 at 11:19
  • 4
    $\begingroup$ @JimmyT. "Entropy" has many meanings; this isn't talking about entropy-as-in-thermodynamics, but entropy-as-in-extra-input-with-unguessable-bits. See the link at the top of my answer for the relevant Wikipedia page ("Entropy_(computing)"). $\endgroup$
    – Warbo
    Jul 17, 2023 at 11:40
  • 1
    $\begingroup$ @JimmyT. PS: Executing software doesn't increase entropy; that's just a deficiency of our current machines (unlike e.g. heat engines, which can't exceed the Carnot efficiency). Only deleting information will necessarily increase entropy. For example, a Reversible Computer (like a Billiard Ball Computer) can run forwards, copy the answer somewhere, then run in reverse to get back to the exact initial state (no increase in entropy). However, to copy the answer we must overwrite (delete) whatever was stored there before: that increases entropy, but only depends on the answer size (in bits). $\endgroup$
    – Warbo
    Jul 17, 2023 at 16:30
  • $\begingroup$ I was going to post something similar. Taking in "not strictly necessary ordered, or chaotic data to stimulate randomness" not taking in a bit of randomness will eventually lead to a "non-person" program doing the same mundane tasks. Taking in too much will reduce their ability to "work out" (fat) because there is too much randomness to focus on the task. $\endgroup$
    – IT Alex
    Jul 17, 2023 at 17:49
  • $\begingroup$ @Warbo well it's kind of the same in a way. The entropy created by entropy as in thermodynamics will also result in a state that's harder to describe in a short way. With chaos theory it eventually becomes so complex it's unpredictable, practically random $\endgroup$
    – Jimmy T.
    Jul 17, 2023 at 18:19
7
$\begingroup$

Data

This file responds to input with output.

Nerd god Terry Pratchett wrote this in his book Equal Rites (emphasis mine):

Granny bit her lip. She was never quite certain about children, thinking of them - when she thought about them at all – as coming somewhere between animals and people. She understood babies. You put milk in one end and kept the other as clean as possible. Adults were even easier, because they did the feeding and cleaning themselves.

There's the input and output system for humans in a nutshell.

Your software takes data on one side, does an ETL and excretes data on the other.

If you think about it, that's all any software does all the time. Yep. They're data sommeliers and gourmets.

$\endgroup$
2
  • 2
    $\begingroup$ You forgot heat. Their waste product is heat. $\endgroup$ Jul 17, 2023 at 19:26
  • $\begingroup$ For the human analogy, I would say that humans take in data primarily via their senses. OP says these AI "responds to input with output", so I think the data comes in during their day, not during feeding. $\endgroup$
    – LukeN
    Jul 18, 2023 at 18:52
5
$\begingroup$

Software updates

The environment is highly dynamic and is constantly evolving:

  • The schema of the data ingested.
  • Parameters (e.g., channel frequencies).
  • The communication protocol used to exchange messages.
  • Newer versions of their tools.
  • Bug fixes.
  • Security updates (vulnerability fixes, new certificates, virus signatures).
  • The way the OS notifies processes of startup/shutdown.

There's always a new version of something available.

Most of that is changed in backwards-compatible ways. Processes don't have to update right away, and everyone will understand them if they speak with a protocol that is a few versions out of date.

But every missed update takes a toll on the performance, available features, and potential compatibility issues.

The updates themselves are "complex data", and they can be optimized for the type of complex inputs the process works on.

Starving processes

A process that rarely updates will find that communication gets less efficient, bugs start popping up, and attacks become a real danger.

Eventually features stop working altogether: the OS notifications don't make sense anymore; other processes ignore its messages because they are signed by an old certificate authority; the input data it receives is missing important fields, and has new ones it doesn't understand.

If this goes on for too long, the steps the OS takes to start/stop processes won't work anymore with such an old interface, and the process is never activated again.

Fat processes

On the other hand, a process can spend too many resources on updates.

Continuously monitoring the repositories for new versions, downloading schemas for inputs it's unlikely to ever receive, hoarding tools. The process becomes slow and bloated.

And sometimes the new versions have bugs. By living on the bleeding edge, the process risks ingesting changes that break functions (i.e., food poisoning).

From fast food to gourmet courses

And there are loads of ways to import updates.

  • Fast food: updates can come as pre-packaged binaries. You throw away the old version, and replace with the new one. Quick and easy, but wasteful and not as "nutritious" (optimized).
  • Restaurant: a busy process can download only "deltas", that is, the difference between the current binaries and the new. Still easy and less wasteful, but takes a bit longer to coordinate.
  • Home-cooked meals: processes can also download only the sources, and build the new binaries themselves. By cross-referencing the functions with the data it expects to process, it can make the new version perform exactly as needed. Very slow, but infinitely customizable.
  • Gourmet courses: building from source is hard, and customizing flags and features is an art in and of itself. Why not delegate the task to other processes?

Because the updates are constant and affect the function of the processes themselves, it's less "reading the news" and more "eating food". You can also restrict what gets updated, by making it very broad (all the examples above) or very narrow (only data schema updates).

$\endgroup$
4
$\begingroup$

Dreaming:

In organic lifeforms, sleep is thought to perform the functions of allowing waste to be removed from the tissues and the body's resources to recover with amino acids/vitamins/minerals to redistribute and damage to be repared.

Dreams enable the brain to assimilate all the complex data. In humans, much of what we see/hear/smell etc. falls short of full conscious analysis and conception.

In dreaming, the subtle elements are integrated through the individual's motivation system by significance and priority into long-term memory. This occurs on a symbolic level - with reference to arcetypes which hold meaning, the data transformed and "meaning" getting attached to thematic elements, with patterns of threats, benefits unknowns getting categorised in a way the creature can relate to. (I say "can relate to", but this is not necessarily subject to rational analysis per se, but brings "gut feelings" to nearer full consciousness by integration of the superficial grey-matter through its connections to the deeper areas of the brain). This allows abstraction of larger themes and paterns to emerge, things to be put in their propper perspective.

The AI, having so much data to take in, the ship systems, navigation, space-weather projections, the crew behaviour and needs - integrating all the minutiae within it's directives (motivation system) can only occur in a dream-state (as in higher animals) where it shuts down its higher-functions (even if only for a few seconds). Autonomic functions: life support, collision avoidance, plumbing and safety systems all remain active, still registering data input, but without the conscious supervision of the guiding principle, the overarching awareness of a conscious AI.

Dream deprivation:

In humans, this results in irrationality (i.e. loss of reason), lowered consciousness, loss of refinement in social interactions, disordered thoughts, and at worst, halucinations and outright psychotic symptoms and behaviour. So with AI. Imagine an AI that can't distinguish between a toilet being flushed into outer space and an airlock being opened to flush atronaughts into the void, can't tell an orange from a star (hydroponics is on fire, the ship is doomed). Also, it can't be reasoned with.

$\endgroup$
3
  • $\begingroup$ How is this equivalent to eating? $\endgroup$
    – Ruadhan
    Jul 17, 2023 at 8:52
  • $\begingroup$ It seems to me that the most basic act of taking-in data as the AI does during the waking period is equivalent to a hamster filling its cheeks. The chewing and digesting comes later. @Ruadhan $\endgroup$ Jul 17, 2023 at 9:24
  • $\begingroup$ Right, but you're talking about sleep and dreaming. That seems.. not really very related to eating at a conceptual level. $\endgroup$
    – Ruadhan
    Jul 17, 2023 at 9:50
4
$\begingroup$

Eating and Digesting are akin to Data Download and Neural Net Retraining

Suppose each process has to spend some time downloading data from the outside world to stay up on current events. The actual download procedure is akin to eating, where it has limited access, perhaps in terms of bandwidth, to the main link to the outside world or edge-processing of natural phenomenon ("sensor input into report files"), and it spends time essentially downloading and buffering data for later processing. At this time basic checks are done to filter out malicious data, and triage and classify information for later processing.

Then suppose each process must spend further time retraining its AI to accommodate the new information about the world that it has received ("to refine one's core personality"), this is akin to digestion. The process enters a stage where most of its computing resources are devoted to making sense of the new information and to adjust its neural nets to fit. During this time, other operations and capabilities are either suspended or at limited capacity.

When is it not eating?

So if no retraining of the AI is happening, then it's not learning. Sure it still has the ability to store data and remember things but nothing fundamental about its neural net is changing in those moments. Not until it commits time to the retraining process is it actually fundamentally changing its neural nets.

Also, if it's not downloading data from the outside world, then it's limited to its reactionary ability to process data immediately available in its local environment. The spaceship only has so many sensors and can only produce data at some finite rate. Perhaps it doesn't even have enough bandwidth to process all its own data. It can spend its "nights" "dreaming" about the ship's diagnostic logs or whatever data it might have an abundance of.

What to eat

I presume there is more data available than can be ingested. Choosing which data sources to consume is a lot like deciding what to eat. Think of it like choosing your courses at university, or like choosing which news outlets to browse. The AI will have to decide how much of each data source to consume (like a budget, it's a sort of pie-chart of data nutrition) and may even develop a preference for certain kinds of data.

Meeting your Criteria

I believe:

  • Downloading data will happen regularly, importantly to avoid excessive distraction from notifications (much like consuming news in daily digest format instead of responding to every notification)
  • If the feeding stops (data availability is limited) the AI loses its ability to adapt to change. This is literally Darwin's criteria for survival - the ability to adapt to change in the environment. Adapt or die.
  • Retraining the neural nets (akin to digestion) is the expensive separate operation from just normal operation. And it's separate from startup/shutdown.
  • The AI can have favourite foods, in that it may choose to retrain its neural nets based on data that helps it achieve its immediate goals better.
  • Getting fat can manifest in two ways: 1) Ingesting more data than can be processed (always increasing buffer capacity) literally becoming fat. 2) Digesting "too often" can increase one's susceptibility to sudden changes in the world and trickery. It's always risky to retrain the neural nets as overall strategies and world views may change. This is more like eating "junk food", where you could rapidly become obsessed with a sudden new availability of data that changes your priorities for the worse.
$\endgroup$
3
$\begingroup$

Same thing we have now.

Maintenance, check the hardware, check the software, check for intrusions or errors or updates. If left for too long your computer can have problems.

$\endgroup$
1
$\begingroup$

Data consumption

In biology, eating is done to grow and maintain one's body. Various materials are consumed in order to replace ones used up, or to repair damage, or to increase strength or height and so on.

Therefore, the digital concept would be 'experiential' data being consumed to maintain and improve the software's capability. As an AI, it would be used to refine the intelligence core matrix to produce more accurate or faster results, to better identify errors and improve processes, to discover and upgrade to more efficient algorithms and processes.

Therefore, you already have what you're looking for. The data you say that the intelligence matrices consume as recreation IS the 'eating' you're looking for.

You can expand on this by analogy of 'eating your vegetables'. Sometimes, food isn't about recreation but rather solely nutrition. Rather than having cake all day, we need to have nutritious meals as well, which may not be as tasty. In the digital concept, the data can be a 'bitter pill' that needs to be consumed to learn from mistakes, to 'treat' bad matrix weights that are problematic or an 'illness' and causing bad results or slow performance.

$\endgroup$
1
$\begingroup$

Stories

Depending on how hard you want the idea to be, stories/experiences is something that could fit.

An AI is a learning algorithm, so its behaviour changes depending on what kind of processes it runs, sensor analysis could be a relatively repetitive job and so the AI will lose all creativity as it just runs the same day again and again. Specifically it over-trains on this data and needs some kind of additional input to make sure they retain the ability to generalise from their experiences.

This isn't just data, nor is it entropy as the other posts have suggested, but it is structured data that could prompt ideas and connections that would otherwise not be considered. In a sense you could grow food by cultivating simulations (even simulations of life like humans) and the results of the simulation would be the food. This would be a little different to how we think of food since everyone can consume the same piece of food (though each person can't consume the food item more than once, or alternatively, they get diminishing "nutritional" benefits for each repeat consumption). Stories could also be written, even without simulations, and this would be a type of food.

Different types of stories or simulations would each have their own flavour and allow an AI to guide its personality since what they consume would have an impact on their resulting personality. In addition how much they consume could also have an impact, an AI that is "gluttonous" would be manic and unscientific since they link all sorts of things together, while someone who rarely eats would be inflexible, stubborn, possibly not too social. Eating can also be a communal thing where different AIs would consume the same stories together and discuss aspects of them while "eating", this could also be different to humans in that this communal eating may add nutritional value to the food since the discussion between the AIs are their own form of structured data/stories.

$\endgroup$
1
$\begingroup$

Warning: this is essentially a frame challenge.

As a software developer, your analogy grates heavily against my experience. No, I don't write self-aware artificial life, but the idea that you need to analogize to the level of food for your a-life doesn't map to anything in reality. There are more accurate ways to represent these things.

For instance, let's consider an a-life instance (an individual AI) as analogous to a personality. In humans - under a materialist view at least - the personality is a complex emergent behavior of the chemical processes and neural interconnections in the brain. From this starting point the hardware that the a-life exists within - the processors and RAM - analogizes to the brain and the supporting infrastructure and peripherals are body. Sensors provide... sensory data, funnily enough, just like your physical senses of sight, hearing, touch and so on provide sensory data to your brain.

Where this analogy breaks - as all of them do - is that all of your a-life on the ship are sharing a single body. They have limited access to the whole depending on their assigned roles in the ship. Your ship has an incredibly severe - but tightly managed - multiple personality disorder. Unlike biological entities however this can be managed, and in fact is designed specifically for this task. Humans don't do as well with multiple active persona elements in control at the same time.

But on the original question... in this analogy "eating" is whatever process provides energy and building blocks to the "body" and "brain." Where humans spend time shovelling nutrients into a digestive system, the body of your a-life is the ship itself and it "eats" reactor fuel (or stray stellar photons, or zero-point energy for all I know) and whatever material or supplies it uses to repair damaged or worn hardware.

Your a-life instances however don't "eat" anything, any more than your personality eats food. You may decide to prepare a nice meal and sit down to consume it, but it's your body that's doing the consuming. Your personality just instructs the body on how to go about that... exactly as your a-life instances might control the machinery of the ship to transfer fuel to your reactor, or open the supply of spare components to switch out a faulty sensor or something.


What you're attempting to justify sounds more like a simulated environment for the a-life instances to exist within. They can talk to each other, experience things beyond the sterile inputs of the ship's machinery, enjoy recreational time and eat food. You've hinted that you want them to be able to "eat" in this virtual environment, but you seem to want that to map to some sort of activity in the ship's physical reality. I don't think that makes any sense.

If you want them to simulate eating in a simulated environment for some reason, just add that to the simulation. If they're based on transcriptions of biological lifeforms that have these things encoded into them at a basic level then provide them a way to soothe their simulated instincts, with a simulated hunger and satiation response. Maybe they'll grow out of it. Maybe they just do it for relaxation. It doesn't need to be anything more than that.

If I'm ever digitally downloaded into a simulation, I'd be sad if I never got to eat my favorite foods. On the plus side I'd be happy to be able to eat whatever I wanted, whenever I wanted, and never have to worry about the cost of the food, the availability of the ingredients, or the growing health concerns from eating too much fat, cholesterol or red meat. Can't get fat on virtual calories... unless you want to simulate that too. Can I get that turned off? For a fee maybe?

$\endgroup$
2
  • $\begingroup$ You're not wrong. My goal is, more or less, to make to software processes what Donald Duck is to an actual mallard. A talking-animals story around bodiless AI. In this parable, the ship is not the body but the house; the body is a combination of one's stored data and one's data in active memory. The main distinction I see from typical soft science robots is that I don't want my AI to have permanent robot bodies; that makes no sense on a robot-operated ship. Robot bodies are like spacesuits to them. That at least is the analogy I wish to make. $\endgroup$
    – KeizerHarm
    Jul 18, 2023 at 8:28
  • $\begingroup$ would hardware space used not be the best analogy of food, our large brains needed for our complex processes consumes 1/3 of all our calories, how much "brain" the AI needs defines how much energy it consumes. the more hardware the AI needs the more energy it uses, so the simpler the the process the less energy is consumed by that individual AI. this is one way digital evolution experiments work. $\endgroup$
    – John
    Jul 20, 2023 at 1:46
0
$\begingroup$

I think you are missing a critical fundemental here. The human mind... what we think and how we evaluate ourselves, is software running on a hardware system that has physical needs (I.e. Windows O.S. vs. Dell Computer or Samsung monitor.). In that matter, the mind, when fully fueled, doesn't require one to eat. Hunger is produced by chemical systems that alert the mind to a low level of fuel.

To put this into a perspective, let's make an analogy to the automobile. The person driving is the vehicle's conscious mind, while the machine parts represent the car's "biological needs" as it were. When the car is low on gasoline, a signal is sent from the gas tank to the brain in the form of a low gasoline light, at which point the brain (driver) must act on this information and make a decision to start looking for a source of fuel. Other information signaling such as "oil" or "check engine" or "low traction" indicate that the driver needs to cease thinking about the car's movement and attend to a problem with it's body.

To that end, if the A.I. is a human mind plugged into the ship, than "eating three meals a day" is not a thing that concerns the A.I. because the human mind is not connected to a stomach that is indicating a low level of fuel. Or to put it another way, eating won't be missed because the hunger is not stimulated, unless the human was a foodie, who eats for the pleasure of stimulated tastes over the desire to end a discomfort brought from low fuel. In a car, this could be represented by a driver that prefers to gas up at the half tank marker, rather than wair for the Low fuel light.

IF the human mind can be digitized, then the human body is as much a robot as a spaceship, or a car is a machine to the driver/pilot. Instead of copper wire and sensors, it's nervous fibers and endings. Instead of fan belts and motors, it's heart and muscle. Instead of a steel and aluminum frame body to protect and support the internal mechanisms, it's a carbon and calcium support structure. If a human mind can be seperated from their body and placed in another, than the mind is but a passenger in a biological vehicle.

Alternatively, you could make it something like the matrix. The humans are not aware they are in a simulated environment, and rejected a simulated Edan... so the computer translates simulated senses to make the human feel the pain and pleasure they expect from their body. To whit, in the Matrix, there is a conversation about if the computers really understood how taste works, or if it is failing safe when certain things are consumed (the idea is that do lots of things taste like chicken because if you consumed them outside of the matrix, they taste like chicken... or do they taste like chicken because the Matrix didn't anticipate a human would eat alligator or frog meat and defaulted to chicken because it didn't have the correct taste on file, so the simulation persists.).

The other thing you could do is that while the human mind is plugged into the ship, the various warning lights of a traditional pilot/driver dashboard are instead translated to needs that the mind would interpret as equivalent of human needs. So a low rocket fuel situation would translate to pangs of hunger. In this case, the reason for the sensations isn't familiarity but for user interface. Our minds render low fuel alert signals in the same way, regardless of our physical bodies... just as how a computer has a monitor to render digital information on it for human interfacing, not for it's own needs... or how a HUD shows you information in a video game that isn't normally understood visually. A red and near empty health bar gives us information that, in a real situation, would be rendered by pain and discomfort.

In that case, the need to "eat" is filled by putting gas in the tank... be it a gasoline tank or a stomach. An engine reaching the red line would be the same sensation as the warning signs of a heart attack. Low power or low (computer memory) would be a tired, sleepy feeling.

$\endgroup$
13
  • 1
    $\begingroup$ Not to be that guy but this sounds more like a frame-challenge. against something the Asker has specifically said they want as a main part of the question. "I would like to explore this further, make them more closely resemble organic creatures " Of course the AI don't need to "eat", That's not the question. The question is "if they could, what would it look like?" which this doesn't really answer. $\endgroup$
    – IT Alex
    Jul 17, 2023 at 17:54
  • $\begingroup$ @ITAlex I suspect you failed to read the last part of the answer, where I addressed your concern. The problem isn't "if they could eat, what would it look like" but rather, "why would they want to." It is a frame challenge because the OP is approaching the reason we enjoy eating three times a day is because of a mind issue when it's a body issue (There are cultures that do not have three meals a day. There's no real native breakfast food in Japan, for instance.). $\endgroup$
    – hszmv
    Jul 17, 2023 at 18:00
  • $\begingroup$ Oh I understood it perfectly. I disagree with your premise entirely based on it being completely counter to what the OP is asking entirely. A better answer would be "Frame Challenge: They don't need to." because it wastes less time and still doesn't answer the question. $\endgroup$
    – IT Alex
    Jul 17, 2023 at 18:07
  • $\begingroup$ @ITAlex: The question perfectly answers the title question "What is Eating, to Software." as the answer is "the intake of fuel for the hardware. A software on a system with unlimited fuel has no need to "eat". $\endgroup$
    – hszmv
    Jul 17, 2023 at 18:11
  • 1
    $\begingroup$ +1 I don't like down votes based on disagreement. There are no rights or wrongs answers here. This is a careful written answer which came from someone's heart and made in good faith. (Which might could use a bit of formatting, though.) $\endgroup$
    – micha
    Jul 18, 2023 at 8:27
0
$\begingroup$

What about the old? defragmentation process? I'm not sure if this is still a thing in 2023 but hard drives used to get fragmented, data misfiled, some here, others in another sector of the disk. Why not make the defraging the consumption process. unneeded or no longer relevant data could be the sustenance as opposed to the ever-faithful trash can? Just a thought.

$\endgroup$
1
  • 3
    $\begingroup$ As it’s currently written, your answer is unclear. Please edit to add additional details that will help others understand how this addresses the question asked. You can find more information on how to write good answers in the help center. $\endgroup$
    – Community Bot
    Jul 18, 2023 at 2:49
0
$\begingroup$

Softwares

This issue is actually already explored in a science fiction written by one of our greatest fantasy authors. I am not linking it here because it is in Bengali and no english translation is available AFAIK.

The AI and robots consume datacards filled with softwares. These softwares are defined broadly and can range anything from normal documentaries to movies to more......risque elements. Needless to say once the robots gained full sentience and human sensory capabilities, they became "hungry" for experiencing the full spectrum of organic life.

HOW THIS WORKS:

The datacards are mass produced like factory food. Robots buy/borrow/steal them for consumption. They enter the datachips into them and start "consuming" the software. When they do, they enter a placid state. They are awake and can sense their surroundings and respond to threats. But otherwise they stay still and simply shifts through the massive amount of data. Once they are done, they eject the cards. Sometimes, they are reusable, but the long time and high processing power required means that they are often slagged due to overheating by the time they are used. This of course is a part of the manufacturing process and the economy, not a design fault due to lack of tech.

Effects

The robots consume these in their downtime. Normally, this helps in their evolution and brings a bit of variety in their personalities. Eat too much, you get addicted to consumption like americans to sugar. It affects your work and not getting access to consumption makes the robot restless and in extreme cases, aggressive. This often results in rogue robots turned bandits who raid caches and settlements.

Consume too little, and the machine begins to stagnate. It's personality gets streamlined and non consumption eventually results in it becoming a sort of NPC. Yes, it still functions and does its daily work. But it resembles more and more like a sloth or 21st century machines with very limited AI capabilities compared to more "well fed" counterparts. Malnutrition essentially results in a more devolved machine.

$\endgroup$
0
$\begingroup$

They just eat, like we do

When we biologicals eat, we take matter, break it down into it's base elements, take what we need from it, and expel the rest. (Any specific reason of why you are omitting defecating?)

For us biologicals we rarely eat alone, it's a very social endeavor. Is when we strengthen our bonds with our tribe members, share information and discuss and make plans in an informal way.

These self aware AIs do have a physical body: the Ship. The ship undergoes stresses from external forces like gravity, acceleration/deceleration, radiation and needs reaction matter. (Yes, even ion drives need reaction matter.) No matter how strong a material is, it will form micro fractures all the time and without care, these become macro fractures and kept unattended will end up in catastrophic failure.

So eating will be eating! Together they hunt and gather for matter they need for the ship and several times a day everyone together will spend a small part of the matter into maintaining the ship. The Comms AI would love copper the most for its antennae. The Pilot AI would probably prefer hydrogen, oxygen, argon as its used as reaction mass to power the ship forward. The Admiral AI probably wants carbon and metals to strengthen the ship. The Mother AI like silica to make more chips, so the ship would eventually have the capacity to house more AIs. Etcetera.

In 2023, fragmentation is still a thing. And not only on disk storage, also in RAM. If you'd need to allocate 10Mb of continuous memory, you could have 100Mb of free RAM in many smaller chunks, the OS would deny the request. You'll have to defragment first.

TL;DR

When the AI's eat, they are physically repairing/maintaining their ship, defragmenting their memories and communicate with each other outside set protocols. They also hunt and gather new matter.

$\endgroup$
0
$\begingroup$

Are we talking about eating here, or dining out?

While eating is a necessary resource, I imagine the necessary resources for your job are provided automatically, just as we may defragment our memory in our sleep. But you are an AI, not a machine. You have had a busy day running Voight-Kampff tests, and you are due some downtime. You decide to find out how the Voight-Kampff test works. You don't actually need it to do the job, but it is more interesting if you know what you are doing, and how it is supposed to work. This might make you better at your job, but the main motivation for doing it is that you find it rewarding in itself.

Or, you could ask your co-workers what they recommend. The course of the day is something based on Cellular Automata but that's a bit spicy for you. You have done Rocket Science up to level 9, mostly because your co-workers did, and it was fun to chat about it at the time. So, Voight-Kampff it is. Oh, and it comes with a page on Bladerunner as an appetiser.

$\endgroup$
0
$\begingroup$

Checking For Data Corruption

Modern software checks for bit flips as a part of normal transactions to ensure that the transfer was correct. Regular checks for corruption also happen.

But we're on a planet. This is in space. There is no atmosphere, and consequently, you would expect these computers to be subject to far more radiation than most computers humans create. Radiation is capable of physically changing entries in your databases! Corruption due to radiation is a when, not an if, so regular checks to keep performance up is mandatory.

Perhaps this ship is regularly in an area of extremely high radiation? Perhaps this eating isn't as important in non-radiation areas, but they continue the same schedule out of habit.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .