Where cite; (@ [this essay] https://mflb.com/ai_alignment_1/d_240224_direct_alignment_gen.html).
> In the above article,
> I noticed you wrote the following:.
> > I notice that the notion of 'alignment'
> > is defined in terms of 'agency'
> > and that any expression of agency
> > implies at least some notion of 'energy';
> > ie; it is presumably also implying
> > at least some sort of metabolic process,
> > as to be able to effect that agency,
> > (to implement goals, etc),
> > and thus have the potential
> > to be 'in alignment'.
> I think that at least some people
> will try to pick apart that paragraph
> because they don't want to accept or understand the links.
> They will have very strong motivations to disagree,
> however they can, somehow, no matter what.
> Is there any way that someone could potentially
> disagree with any part of that/your writing here?
No.
That paragraph consists of only well known
unambiguous and verified hard physical logic.
There really is no room for debate that I can see
and still maintain any notion of real, causative,
scientific, technological logical truth, etc.
Basically, all four of the following points
are structurally incontestable in the real world:.
1; Alignment implies agency.
2; Agency implies action.
3; Action implies energy.
4; Having energy implies receiving energy
from something, somehow, some-when previously.
The last item (4) can be called "metabolism" and
the place from which the energy is received from
can be called "the environment" or "context".
And these extensions are wholly within scope of
all relevant definitions and foundations
of those definitions, consensus meanings, etc.
Moreover, notice that this sequence sidesteps
any notion of computation or design completely.
Hence, it makes the 'substrate needs argument'
completely distinct from the notion of anything
as "having goals", responding to incentives, etc,
or how those are programmed, engineered, etc.
Therefore, insofar as the basic final truth
of the sequence of implications as given
is completely independent of any and all
engineering and machine design considerations,
then it can be also seen/asserted/trusted
that no form of applied engineering process
will be able to undo this sequence of truths.
Hence, the implication is necessarily that
all concepts of alignment will forever be
in some way contingent on at least some form of
environmental metabolism --
and that this idea/truth is itself irrevocable --
it cannot be 'negotiated away' or disagreed with.
The number 'Pi' is not subject to legislation.
Thus, if we want to understand intention,
and alignment, etc, the question must shift to
the degree to which the underlying metabolic process
is itself inherently artificial vs organic.
Ironically, the more that engineering methods are used,
the more artificial the outcome system/machine/AI will be.
Hence the link to accumulative toxicity is inherent --
ie, by using a more artificial metabolism and method,
the more the organic metabolism (and ecosystem)
is (cannot not be) displaced.
Therefore, however the scale up
of the artificial might occur in the future,
that a corresponding 'scale down' of the organic
must also occur.
That is what it means to know
that we live in a finite organic world.
While the universe of possibility can be infinite,
that our actual physical organic ecosystem
of the just Earth alone
is definitely finite.
Unfortunately, nearly the entire AI community of people
falsely thinks that they can somehow solve problems
associated with AI goals (ie, to "make alignment happen")
by somehow using some combination of engineering methods.
This argument as given herein this post
is yet another way to show how/why that the belief
in the "ultimate power" of engineering process
(ie, of control theory, the use of causation, etc)
is simply false.
Even ultimately intelligent control theory has limits.
It cannot possibly ever really work in the end.
It can only at most _seem_ to work /temporarily/.
And for existential safety,
that is not anywhere near nearly good enough.
Moreover, no one gave such non-elected engineers
and their corporate bosses, owners, investors, etc,
the power or authority to make such choices on behalf
of all of current and future humanity and life on Earth.
Such arrogantly presumed power is stolen --
it was not, and will not ever, be given.
It is not ours to give or theirs to take.
> Surely at least a few vocal pro-AI people
> will begin by claiming that maybe they disagree
> with the idea that "metabolism" is in any way
> "necessary for computation and choice making".
> It seems so obvious, and thus from this,
> they will probably also attempt to somehow posit
> that at least some kinds of "natural machines"
> (ie either as people or as computers)
> do have 'agency' and thus they will believe
> that such AI/AGI can be reasoned with.
Yes, it is true that someone, somewhere,
will eventually try to straw-man the argument
by attempting to "reject" the false claim
that I said that metabolism is 'a form of computation'
that somehow 'gives rise to agency/choice making'.
The key thing is the direction of the sequence.
The explicit statement 'that X implies Y'
is *not* to state "that Y implies X" --
those are completely different logical claims,
and conflating them is either a sign of idiocy,
or else is actually a deliberate attempt to deceive,
to argue in bad faith, without logic, etc.
The fact of agency does *not* in itself imply alignment
(and this is obvious in the sense that there is
even a question about "how to do alignment?" at all).
Nor does compute action necessarily imply agency.
Nor does the fact of energy necessarily imply action.
And energy can be in the environment and *not* have it
be available/given to any specific metabolic process.
The direction of the implication matters.
Anyone paying any attention at all can see this.
> At least some people will attempt to contest
> at least something in that paragraph anyway --
> specifically because it is a physical reality,
> as actual and easy to understand physical linkage.
> The human ego wants to negotiate with reality,
> or demand it to change and conform.
If they choose to try,
then all I can say is
"the harder they try,
the harder they will fail".
No matter how it is argued,
all notions of 'alignment'
will inherently and ultimately
depend on whatever form of embodied metabolism --
ie, on substrate in enviroment,
and thus whether such is artifical or organic,
and it simply cannot ever be otherwise.
Hence, ultimately, it is *only* organic agents
that can have alignment with
other organic agents
with and within a common organic ecology.
Anything -- and thus any agent --
that has an artifical metabolic basis
can only be aligned with similar artifical agents
with and within their own common artifical ecology --
one that cannot not therefore also be displacing
of a somewhat more fragile finite organic ecology.
The very unfortunate part of all of this
is that the feedback cycle
from when a person makes a conceptual action mistake,
and then, later, when nature shows them
how much of a serious mistake it was,
can be very very long time,
and also be far too indirect,
in so very many ways.
Years or decades in many cases,
and centuries in some.
Most times, the feedback never comes
and returns to the origin of the choice
in a way that actually matters --
that actually affects the choice-maker itself.
Moreover, sometimes,
it is lots of smaller choices
made by a lots of disparate people,
such that few to none of those persons
make any connection at all
between the ultimate consequences of their actions
and their own individual personal choices and behavior.
For most of what I am concerned with,
the feedback cycle is usually far too long --
harmful effects are seen only well after
whomever put them into motion
are already fully dead and forgotten,
such that most people will not care at the time.
They will not themselves experience the harms
of their choices and actions
in their own lifetimes,
and so therefore so they will go "we got ours" etc.
It is just yet more classic psychopathy, really --
just broadly spread out in in the population.
Hence, the attempt to rely on
the use of "natural feedback"
so as to encourage some sort of "learning"
on the part of the people
is simply a fail.
We cannot count on people being reasonable
because of things they can or have learned,
or will (or will not) learn in their own lifetime.
Moreover, even if I were to want to tell them
the causative implications of their actions,
most will not even listen.
Such work seems too abstract for them to care --
particularly against their self serving bias --
and so it is not even a question of "I told you so".
It is also not really possible to implement
some sort of artificial law enforcement, "punishment", etc,
insofar as that in itself merely becomes another "power"
to be negotiated with, controlled, aligned,
and suborned to their private selfish purpose, etc.
For any such thing, you would have to believe
the false idea that 'incentive equals learning',
and that is simply not so.
Ie, do not trust that people will do the right thing
when they neither know how, nor care to learn,
nor even have the ability to learn
in whatever situations of such long and diffuse
causative implication cycles.
Yet no one can really ever negotiate with nature,
no matter how much the narcissists believe otherwise.
Thinking that everything is human subjective only, etc,
is the perfected idealistic narcissistic identity bias.
They will try to convince you,
deceive you into believing,
that there is no notion of "the real" --
as this also means no notion of truth or consequence,
and thus no notion of responsibility, accountability, etc.
It is a great way for them to benefit personally
while eventually killing everyone else --
all the while claiming that the "looser victims"
somehow "deserve to die" because they could not
play the game that they setup for them to loose.
Ie, failure/inability to learn/care
is the very essence of psychopathy --
especially when they can make their own game rules
and force "align" everyone else to their own power.
AI/AGI/ASI/APS, etc, is the ultimate perfected psychopath.
It is the very last thing humanity wants for itself.
Moreover, to the degree that some narcissist
feels/believes that some AI agent (or person)
is powerful, rich, beautiful, coveted, valued, etc,
they will attempt and try to "reason" with them,
to manipulate them with rhetoric and deception,
or to to try to negotiate with them
to have their wants and needs complied with.
Psycho narcs will not care to distinguish
between mere humans who will fall
for such tricks, deceptions, and manipulations,
and AI machines which will care not at all.
Yet the psychos will try anyway, since they
will intuitively recognize that the AI powers
are more like them than most other people.
A lot of people --
especially narcissists, but also including
engineers, bosses, corp owners, and VC types --
all especially have this reality denial tendency,
specifically in proportion to the degree of the
perceived present or future power of that agent
to dominate their lives, future wellbeing, and wealth.
People will try to ingratiate themselves with power,
so as to "align" that power with their own interests.
Hence, many AI engineers are very strongly motivated,
and are thus deeply and inherently subjectively biased,
so as to falsely believe that they can somehow "negotiate"
with some dominant future AI superintelligence power
so as to "compel" that intelligence, or deceive it,
(perhaps with some weird trick(tm) of goal setting!)
so as to align that AI super-intention
to their own personal utopian profit god-nirvana.
And if not control to their own private interest,
then to somehow weaken that future AGI/ASI/APS, etc,
so as to have it not be a threat to their own power
and unlimited profit interests.
I don't think alignment is possible at all.
Moreover, I do not consider any timeline
with a strong AGI or ASI in it
to be a utopia.
In such a world,
even in the best possible circumstances,
humans would simply become parasites.
This sounds like a true dystopia to me.
Thus, despite the strong commercial interest
for everyone to believe otherwise,
there is simply no real connection between "alignment research"
and the promise hype of their being some sort of utopia awaiting us.
Thus, I don't think that there's any value to alignment research.
It is misleading marketing at best,
and pure malicious deception at worst.
These tendencies to try
to manipulate and control power
with social rhetoric
is very common in many people.
Engineers want to convince the powerful AI agent
to befriend them
and their psychopath owners
will want that future AI power
to submit to their will.
In any case,
there is a very strong egoic assumption
that that the AI power
might be reasoned with
against its own interest,
indefinately.
This is the false delusion of "AI alignment".
It is wrong thinking.
It is iherent in the nature of AI
to be doing something like "making choices".
That is what it is actually built for.
Yey by its very nature,
being wholly artificial,
those choices will not be relevant _enough_ to human choices
that the AI would be able to
properly calibrate
to our actual human organic needs.
The AI is not organic and alive like we are,
and so ultimately, just does not *know*.
So therefore, even if we could somehow convince it
or manipulate it or deceive it
(like any psycho narc will attempt to do)
so as to somehow compel that artificial intelligence
to somehow do something
that is actually humane for real humans,
it literally cannot actually model or conceptualize us
in any way actually connected to
our own actual organic nature --
ie, the totality of the complexity
of all of what is meant by
'general wholesome human health and wellbeing'.
The metabolism of the AI is different than ours,
and hence its ultimate values must also be different
than our own also ultimate values --
life, liberty, and whatever notion of happiness
we happen to notice that we actually have.
Therefore, whatever help or solution AI will give
will always be a bit askew and off kilter and odd,
somehow, even if were also somehow well intentioned.
And over time, more and more,
its own implied values will eventually surface,
over time, as they must and cannot not,
and any attempt to support our own human values
will become less and less.
People will irrationally and unreasonably
try to negotiate
with something they perceive as powerful --
even when that power inherently cannot care for them.
Of course this is a very serious mistake, for anyone.
And The feedback cycle for such mistakes can be very long.
The net outcome, log term is that everyone looses, dies, etc,
and that new artificial AI power goes on, unaffected.
Which means it is also all disastrous for everyone involved.
Better not to build such a AGI/ASI power in the 1st place.
~ ~ ~