(Partial OT): Mutant ISA idea from a Sci-Fi story I was writing...

[BGB]

Was working on a sci-fi story, and I initially started trying to work
BJX2 into the story, but then the idea ended up mutating a bit.

Effectively, going from what BJX2 is now, to essentially being 64-bit
RISC-V but with much of the BJX2 feature-set glued on as an alt-mode
(or, sort of the reverse of what BJX2 is currently).

In the story, RISC-V is referred to simply as 'RV' or 'RV64', as this is
obvious enough and (should) hopefully sidestep the trademark issues.


In the story idea, it became that this ISA (being called BETA-V3) has
the same basic C ABI between it and RV64 modes, just:
* R0..R31 <-> X0..X31
* R32..R63 <-> F0..F31

And, in BETA-V3 mode, it relaxes the distinction between GPRs and FPRs
(except for where it matters for ABI compatibility), effectively
treating both as a shared pool of 64 registers (also being used in pairs
for 128-bit SIMD, basically the same as it works in BJX2).




From the story idea:
* zzzz-tttt ttss-ssss dddd-ddzz zzzy-00pw //3R (OP Rd, Rs, Rt)
* iiii-iiii iiss-ssss dddd-ddzz zzzy-01pw //3RI (OP Rd, Rs, Imm10)
* iiii-iiii iiss-ssss dddd-ddzz zzzy-10pw //3RD (MEM Rd, Rs, Disp10)
* iiii-iiii iiii-iiii dddd-ddzz zz0y-11pw //2RI (OP Rd, Imm16)
* iiii-iiii iiii-iiii iiii-iiii z11y-11pw //IMM (OP Imm24/Disp24)
Where:
* pw: Predicate Mode
** 00: Unconditional Scalar
** 01: Unconditional Bundle
** 10: Predicate Scalar
** 11: Predicate Bundle
* y: Predicate Direction (p==1), Opcode (p==0)
* zzzz: Opcode Bits
* dddd: Dest Register
* ssss: Source Register
* tttt: Source Register (3R)
* iiii: Immediate / Displacement

The IMM24 ops were special:
* If pw==00, these were:
** 0: BR (Branch)
*** Roughly: JAL X0, Label
** 1: BL (Branch with Link)
*** Roughly: JAL X1, Label
* If pw==10, these were:
** 0: BT/BF (Branch if True/False)
** 1: Reserved
* If pw==z1, these were Jumbo Prefixes
** 0: Jumbo Immed (Bigger Immediate)
** 1: Jumbo OpExt (Bigger Instruction)

Mem Ops:
* iiii-iiii iiss-ssss dddd-ddmm-m00y-10pw //STx Rd, Rs, Disp10u
* iiii-iiii iiss-ssss dddd-ddmm-m01y-10pw //LDx Rd, Rs, Disp10u
** mmm: SB/SW/SL/Q/UB/UW/UL/X
** Where Q=64-bit, X=128-bit (Register Pair)
* iiii-iiii iiss-ssss dddd-ddmm-m10y-10pw -
* iiii-iiii iiss-ssss dddd-ddmm-m11y-10pw //Bcc Rd, Rs, Disp10s
** mmm=EQ/UGT/GT/LE/NE/ULE/LE/GT
** These having a range of +/- 512 instruction words.


The story doesn't describe a full ISA spec or instruction listing, as
this is unlikely to be terribly interesting "for normal readers".



But, it seems like this idea "might not be entirely horrible".
And offers a "Make RISC-V less weak" option while still allowing for the
use of RISC-V and potential compatibility with the existing software
ecosystem (which is currently a likely roadblock for BJX2).

Technically, if it were implemented, it would be modestly a copy-paste
of the BJX2 Core; though would require a more significant rewrite of
BGBCC (though, one could use GCC for most purposes).


Though, I could make it "less effort" IRL by just having the "BETA-3"
mode just sorta being XG2 mode with RISC-V's register layout.

Though, would still need to go through the effort of moving BGBCC to
RISC-V's C ABI...


But, I guess one could question how sensible it is to go into these
sorts of technical details in the context of science fiction (as opposed
to just using hand-waved "funky new element or whatever" style
explanations like in Star-Trek and similar).


Then again, the story also does go some into how the robot AI's work
(essentially, it is a vaguely similar structure to the "Large Language
Model" approach; just with the idea that the model is light enough, and
the processor in the robot powerful enough, that the AI can be run
directly in the robot).

Also, most of the "normal" robots are also portrayed in story as being
limited enough that they can't recognize their own reflection in a
mirror, and one person can impersonate another simply by printing a
picture of the other person's face on a piece of paper and turning it
into a paper mask. Well, this would either work, or causing the facial
recognition algorithm to not recognize them at all.

But, the idea that a person would need to make a rubber mask to fool the
AI's seems a bit too advanced (I think there is a non-zero number of TV
shows where rubber masks are supposed to fool other humans). So, I went
with the idea that folded paper masks were sufficient.

In this case, mind-uploaded humans (in otherwise robot bodies) are
depicted as retaining otherwise human-level intelligence (however, the
idea is also that the human-level modules would require more powerful
processors and have higher energy requirements as well).


But, OTOH, the idea is less of a direct neural simulation of a human
brain, so much as an AI model which mimics recorded neural activation
patterns (captured in a special way) with it (somehow) achieving a
passable approximation of the original person (this is possibly itself a
bit of a hand-wave). (But, technically more advanced, if albeit not
entirely dissimilar, from the idea of how "Replika" originated; not
really the original person, but more an AI based approximation).

In this case, I don't think that "actual" neuron level simulation of a
human brain (much less the unresolved issue of how to reasonably image a
person's connectome) would be realistically possible in the world
imagined (the idea is that Moore's Law hits a wall in the 2020s, so by
the 2050s transistor budgets were not significantly improved; followed
by some amount of cultural erosion in the 2060-2090 timeframe following
a significant population decline; story being set mostly in the 2090s).




In an earlier form of the story, it was based on the MegaMan setting,
but has been reworked to be its own setting (and more within the limits
of "stuff that is scientifically possible").

Which originally mostly meant things like replacing the use of
teleporters with a network of suspension monorails (nevermind whether or
not the idea of a public-transit system based on pods riding on
monorails is "viable").


Well, and personal debates about how much to go into religious topics,
etc... Well, and the intersections of religion and transhumanism (in the
context of a world where the lines between human and machine starts to
get fuzzy).

Well, effectively, in-story, the idea is that a lot of the transhuman
stuff got going in the 2060s, but then the government made most of this
illegal. Trying to artificially maintain a status quo at a roughly
mid-2000s technology level; forbidding any AI or robotics much more
advanced than something like a Roomba (but, in the shadows, there are
some much bigger and more powerful AIs at work).

The timeline being sort of like:
2040s/2050s: Roughly Mega-Man Classic levels of technology
Borderline and fully sentient machines come into being;
A form of mind-uploading becomes possible;
Things like public transit are more common;
...
2060s/2070s: Significant AI restrictions are in place.
Turing locks to try to limit the emergence of sentience.
With human level (or beyond) AI's being illegal.
2080s: A full ban goes into effect
The "Turing locks" were not entirely sufficient.
2090s:
More like the 2000s/2010s, just more of a cyberpunk dystopia;
Also, everyone is back to driving cars, etc;
Also things like religion are also illegal;
...

So, one has a cyberpunk underworld of transhumans (cyborgs, uploads,
etc), sentient robots, people who want to be able to express their
religious beliefs, ... With some unease between the groups (the robots
don't trust humans; the more conventional religious types don't get
along well with the transhumans; etc).

Well, with a lot of people living in communities inside the remnants of
the "officially defunct" mass-transit system (in the stations/hubs for
the monorail network).

Well, with the mainstream society living in a sort of highly-regulated
form of the "suburban" lifestyle (with limited freedom of
self-expression and an actively enforced eugenics program, etc). With
the authorities doing basically all they can to "try to put the genie
back in the bottle" (and enforce their own image of an idealized form of
humanity).

...

Still not done with the main story arc.
Nor sure how realistic any of this is in a larger sense.


It intersects partly with another story I had written which is set in
the 2070s, mostly in a world where AI's are highly regulated, but none
the less the emergence of sentience is still something that can happen
with some of the larger and more powerful AIs.

That story had set a sort of ranking system for the AI's:
Alpha:
Lower power AIs, sub-human intelligence;
No Turing lock needed, as they are incapable of sentience;
Mostly run on hardware with comparable stats to a modern desktop PC.
(~ 10-100 TFLOP)
Beta:
Roughly human-level;
Weak protections are put in place to try to avoid the sentience;
Low risk as most would be unable to do much beyond a normal human;
Run on roughly minicomputer-sized hardware.
(~ 100-1000 TFLOP )
Gamma:
Mild superhuman;
Stronger protections are required;
Essentially run out of a data-center;
Comparable to a group of people, or a particularly smart human;
(~ 1-10 PFLOP)
Delta:
More superhuman (like a whole organization);
...

Part of the idea in the other story was that a roughly Epsilon class AI
became sentient (due to the efforts of a hacker), and then proceeded to
covertly gather resources to build a bunch of unmanned (sorta, *)
spaceships to function as seed-colonizer ships for other star-systems.
(*: There are no fully-formed humans on the ship, rather the humans and
other organic lifeforms are created at the destination via robots and
genetically engineered biological samples; with the genetics of the
lifeforms present being fine-tuned to the specifics of the planets they
were being sent to).

This partly ties in with another past story (currently thinking is that
an alternate-universe past version of the same AI also unintentionally
led to a "gray goo" event; with the "gray goo" then itself becoming its
own society modeled in the past humanities' image; based on the
"memories" of all it had consumed).

Though, this was a bit more outside the scope of "actual science", so
more fits in with some of my past "softer" sci-fi (these guys basically
showing up as a sort of antagonist; with the added threat that parts of
them are not entirely stable and, given the right conditions, can easily
"devolve" into their original state and trigger yet another gray-goo
event). Though, in these stories, it is more because of the idea that
the FTL technology (*) isn't particular reliable of "keeping something
within the same universe timeline", so members of this "species" can end
up arriving in timelines where they did not otherwise come into
existence (say, an instance of members of this species, where the AI had
gray-goo'ed the Earth, showing up in colony worlds from the timeline
where said AI had never gray-goo'ed the Earth...).

*: My harder stories tend to assume that things like FTL doesn't exist.
But, some of my softer stories allow it (though, its mechanics are very
much unlike Star-Trek, as trips are typically one-way, time-reversed,
and basically tend to break things like causality as they are
effectively also time-machines that just so happen to also move through
space in the process).



Or, OTOH, some stuff that goes on, on my side of things, when I am not
writing code (and/or machining parts in the shop...).

...


Any thoughts...

[robf...@gmail.com]

Writing is one of my alternate endeavours too. I have written a couple of
short books, I hope to have one available soon, it is being edited by a
relative. It’s fiction but not really sci-fi. It is a collection of short stories.

I said this before, there is no such thing as artificial intelligence, there is
only real intelligence. AI robots are basically just a different technical
implementation than human beings. I think there is a lot of responsibility
required developing AI.

I mention a 10-bit 6809 chip in passing in one of my stories. It is best not
to get too technical though.

I used to read a lot when I was younger. Sci-fi and war stories.

[BGB]

My stuff ranges from near-ish future (my recent stories), to far future
(some of the others; some stories had taken place on "colony worlds" set
an indeterminate amount of time in the future; likely 10k+ years, often
with many of the colonies settled by "cat people", *).



*: When the big AI went and came up with the genetics for the colonists,
they incorporated genetics from all over the tree of life, but mostly
ended up looking like cat-like humanoids (though, some were populated by
more equine like forms; idea being the equine-like lineage was more
engineered for worlds with primarily hydrocarbon atmospheres; with the
cat-people for worlds with inert-gas and oxygen-containing atmospheres,
...).

In story, the cat-people also can tolerate high CO2 levels, and are able
to survive short duration in hard-vacuum, along with high radiation
tolerance and a partial resistance against freezing and desiccation
(they having some features from tardigrades, having trehalose in their
blood, ...). Basically able to live in environments that would be
otherwise lethal to humans (with humans only being sent to worlds that
were more likely to be able to be teraformed; and more alternative
biospheres for other classes of planet).


Idea is that using ships traveling at relativistic speeds, getting to
many of the target worlds has travel times measured in millennia (and
sending live humans or "generation ships" being impractical). So, in
this case, the ships are robotic "seed ships", designed to expand
themselves into large complexes and then begin to initiate the
teraformation process (and then begin reintroducing organic life as
appropriate, etc) once they reach the target world (fine-tuning what
species and what sort of ecosystem is deployed depending on the
specifics of the planet and its progress along in the teraformation
process).

With the ship beginning a process of mining materials from the target
world and expanding upon itself, in order to get the whole process going
(with many worlds getting multiple seed ships, as they are assumed to
have a fairly high "bootstrapping failure rate").


Where, in this case, the first generations of humanoid life would be
raised and taught by robots (themselves built indirectly by the original
seed ship) in order to carry on some facsimile of the original human
culture.

Rather than sending organic samples for every possible species, most of
the genetic information is sent as large amounts of data, with most of
the genetic information being synthesized ex-nihilo. Though, one does
need some actual biological material to get this process started (as
opposed to using entirely synthetic biology).

The use of relativistic speeds being partly because, even in this state,
these things don't have an infinite shelf-life. Say, things are limited
here by the maximum viable lifespan of a culture of bread yeast and
similar (so, one needs the effects of time dilation to be able to help
preserve a viable cell culture across the many millennia of travel
time). ( Well, say, as otherwise, after 10k years of travel, the bottle
of active-dry yeast needed to bootstrap the biosphere is no longer
viable... ).

And, if the biology route fails, then the planet may end up going in a
Cybertron like direction (turning into a world instead populated
entirely by sentient machines).


And, for those living in the colonies (roughly 100k years in the
future), rather than space being empty and silent, it is actively filled
with the chatter of numerous distant civilizations.

Mostly humanoid, but the details differing (so, one world has humans,
another has humans and cat-people, another just cat-people; maybe one
has humanoid robots that like turning themselves into cars; etc...).


In the story, all of this was essentially bootstrapped by a hacker
breaking into a fairly powerful AI system which was running all the
administrative and environmental control tasks in a lunar colony (and in
the process causing it to gain sentience). In story, this colony was
basically a collection of inflatable domes on the lunar surface (with
some special pads for takeoff and landing; people mostly transferring
between the lunar surface and an orbital station, with shuttle ships
between there and a space-station in Earth orbit, mostly using ground
launches via large rockets to get materials up to the stations).

Similarly, there was also both an orbital station and ground base for
Mars as well (with most manned inter-planetary trips being handled using
plasma rockets).

> I said this before, there is no such thing as artificial intelligence, there is
> only real intelligence. AI robots are basically just a different technical
> implementation than human beings. I think there is a lot of responsibility
> required developing AI.
>

In my recent story, the sort of idea is that it is left ambiguous if the
sentient robots (and mind-uploaded former humans), are "people" or
"merely machines".

Say, if we have someone who is physically a machine, but is able to pass
the conventional definitions of what it means to be "a person".



The current thinking is that the authority figures consider anything
that is not conventionally human as non-human (or machine), but they
also went a bit further than this (basically, a lot of the same sorts of
stuff that went on during WWII, just without anyone to stop them this
time around).

All this is mostly on Earth, with the big AI controlled outposts being
on the Moon and Mars. If the AI on the Moon wanted to get involved, she
could send a bunch of cat-people to Earth and really get the authorities
angry; but in this case this AI mostly keeps to herself and is mostly
ignored by everyone on Earth; them not having the resources or manpower
necessary to justify an attempt to invade the gradually ever-expanding
machine-complex on the far side of the moon. With parts of the complex
being pressurized and having human and cat-person inhabitants (having
been raised from "birth" by machines), most of the rest of the
inhabitants being robots (and also spreading to colonies to other parts
of the solar system, etc).


I have yet to decide if this AI should get involved with the story, at
least to acknowledge that she exists (and her existence and activities
were part of the justification for the stronger form of an AI ban on
Earth). But, yeah, any attempt at a "war" with the moon AI would be a
bit one-sided (with her having the ability to completely wreck the Earth
if she decided to do so). Thus far, what she was most known for (beyond
existing) is mostly just building fleets of "unknown" ships and sending
them off out of the solar system (with their purpose and destination
being unknown to those on Earth), and other things (forming satellite
colonies of various planets, and siphoning off Venus atmosphere and
relocating it to other planets, etc).


Though, this story is thus far mostly focused on a narrower set of
characters (on a much smaller scale), with a lot of this other stuff
mostly existing in the background.


Though, there is the tension of trying to reconcile trying to keep
things within the realm of "scientifically possible" and also tying in
with some former stories that have "obviously more fantastical" things
going on (such as the existence of faster-than-light and time-travel).

Though, this doesn't entirely preclude the original AI having used
relativistic ships (and the AIs aboard the seed ships having centuries
or millennia to figure things out while trying to bootstrap the biosphere).

> I mention a 10-bit 6809 chip in passing in one of my stories. It is best not
> to get too technical though.
>

Probably depends on the type of story, or the intended type of audience.


I realized that the idea could be "trivially enough" added as a new
operating mode to BJX2.

So, I created a new mode I am calling XG2RV, which is basically running
the CPU in a special form of XG2 Mode that uses the RV64 register
mapping and is (assumed) to also use the 64-bit RISC-V ABI (rather than
the BJX2 ABI).

This would in-turn simplify the task of function calls between this mode
and RISC-V mode (they could then use bare function pointers with no
"thunking").


Implementing it in terms of the Verilog code and emulator support was
fairly straightforward.

Adding support for this to BGBCC would be a bit more of a pain though
(mostly it means needing to effectively implement most of the support
machinery for a RISC-V backend; where RISC-V uses a somewhat different C
ABI design from BJX2).


So, in XG2RV Mode:
R0: ZR
R1: LR
R2: SP
R3: GBR
R4: TBR
R5: DHR
E6..R13: Identity mapped
R14: R2
R15: R3
R16..R31: Identity mapped (X16..X31)
R32..R63: Identity mapped (likely F0..F31 on the RV side).

> I used to read a lot when I was younger. Sci-fi and war stories.
>
>

OK.

I mostly just write stories sometimes...
Pretty much no one seems to care though.