Adam Funk <a24061@ducksburg.com> wrote:

> On 2024-02-19, Sam Plusnet wrote:
>
> > On 19-Feb-24 11:17, Adam Funk wrote:
> >> On 2024-02-07, lar3ryca wrote:
[-]
> >>> It got a lot of bad reviews at the time, but I consider it to be one of
> >>> the funniest SF films ever, in the same class as 'Galaxy Quest'.
> >>>
> >>> See: https://en.wikipedia.org/wiki/Message_from_Space
> >>>
> >>>> <https://scifi.stackexchange.com/questions/80359/short-story-space-probe-
voyager-discovered-by-aliens>
> >>>
> >>> Nothing there either. Thanks for trying.
> >>>
> >>> If my (admittedly faint) memory serves, one of the reasons for umbrage
> >>> was the pulsar periods/coordinates from Earth (the long lines from the
> >>> origin) may have resembled either the aliens that found it or their
> >>> enemies.
> >>
> >> Well, it turns out one of the pulsar specifications is misleading
> >> anyway.
> >>
> >> <https://en.wikipedia.org/wiki/Pioneer_plaque#Sun_and_galactic_landmarks>
> >
> > Can we issue a patch?
>
> I don't see why not, but maybe we don't want to send accurate
> instructions for finding us?

Sorry about being boringly factual, but it is out of the question.
(and will be for the forseeable future)
To get a probe to leave the solar system
requires multiple gravity assists. (aka planetary slingshots)

The planets will never be in the right positions once again
to get a probe to leave in precisely the right direction.
It is hard enough to get one to leave at all,

Jan

Lo, on the 2/15/2024, occam did proclaim ...

[elision runs rampant]

> As it turned out, I did opt to sit in a couple of undergraduate AI
> courses in a different part of the University (of London), purely out of
> interest. 45 years ago AI (or Machine Learning) was not as fashionable
> as it is today.

Was that all if-then rules-based, or had it gotten to fuzzy logic by
then?

My formal AI training was basically like learning geography by viewing
a set of vacation slides borrowed from the instructor's friends. I
didn't even get around to trying Prolog or turtle graphics.

/dps

--
Ieri, oggi, domani

Peter Moylan <peter@pmoylan.org.invalid> writes:

> That's one story. But the national economy suffers when only the rich go
> to university, so that a lot of hidden talent is wasted.
>
> Peter Moylan http://www.pmoylan.org
> Newcastle, NSW

Now, in the US anyhow, the non-rich can go to university, are led to
believe that they *must* go to university if they want to remain
non-rich rather than destitute and homeless -- but in exchange for
debt indenture for decades or life.

When I was in school, education was cheap and summer jobs could cover
a lot of it. I paid off my student loan, such as it was, in 11 months
with a job at a prestigious place but with wage so low that my boss
apologized several times for the level.

Oh, right, 1960s.

--
Mike Spencer Nova Scotia, Canada

On 24 Feb 2024 01:33:52 -0400, Mike Spencer
<mds@bogus.nodomain.nowhere> wrote:

>
>Peter Moylan <peter@pmoylan.org.invalid> writes:
>
>> That's one story. But the national economy suffers when only the rich go
>> to university, so that a lot of hidden talent is wasted.
>>
>> Peter Moylan http://www.pmoylan.org
>> Newcastle, NSW
>
>Now, in the US anyhow, the non-rich can go to university, are led to
>believe that they *must* go to university if they want to remain
>non-rich rather than destitute and homeless -- but in exchange for
>debt indenture for decades or life.
>
>When I was in school, education was cheap and summer jobs could cover
>a lot of it. I paid off my student loan, such as it was, in 11 months
>with a job at a prestigious place but with wage so low that my boss
>apologized several times for the level.
>
>Oh, right, 1960s.

On 1960s and the cost of colleges:
I matriculated at Rice U. (Houston) in 1964 when the tuition
was zero, as dictated by W.M. Rice whose estate founded it.
A few years later, trustees got the courts to break that
condition so (they said) they could collect funding from Merit
Scholarships and others which provided funding according to
'need'. (They had previously broken conditions to allow them
to admit women and Blacks.)

They stated that full 'scholarships' would ensure that no
student was turned away by cost.

Not long after that, I was aware that other elite universities
made similar statements, and I read that almost all students
attended them with some reduction in fees.

Some time along the line, "able to attend" got subverted by
the introduction of loans in place of scholarships.

And then, for some varieties of loans, the government got
involved in guarantees, where the political bargaining led to
HIGH interest rates -- which would NOT be reduced while the
loan was being paid off. I know that there was some political
argument when interest rates fell and students were NOT
allowed to re-finance. Paying 8% interest or more seemed
unfair when the going rates were half that.

So far as I know, the loans that Biden is causing to be 'forgiven'
fall into that category -- 'unfair' loans on which students could
barely pay more than the annual interest.

Oh, I transferred to UT (Austin) where the tuition soon was
doubled -- from $50 to $100 per semester for in-state students,
and from $100 to $200 for out-of-state. Fees that low were
not common, and drew smart applicants from all over. Graduates
who did not move away left Austin ripe for the high-tech-growth
that followed.

--
Rich Ulrich

On Sat, 24 Feb 2024 12:34:52 -0500, Rich Ulrich
<rich.ulrich@comcast.net> wrote:

>On 24 Feb 2024 01:33:52 -0400, Mike Spencer
><mds@bogus.nodomain.nowhere> wrote:
>
>>
>>Peter Moylan <peter@pmoylan.org.invalid> writes:
>>
>>> That's one story. But the national economy suffers when only the rich go
>>> to university, so that a lot of hidden talent is wasted.
>>>
>>> Peter Moylan http://www.pmoylan.org
>>> Newcastle, NSW
>>
>>Now, in the US anyhow, the non-rich can go to university, are led to
>>believe that they *must* go to university if they want to remain
>>non-rich rather than destitute and homeless -- but in exchange for
>>debt indenture for decades or life.
>>
>>When I was in school, education was cheap and summer jobs could cover
>>a lot of it. I paid off my student loan, such as it was, in 11 months
>>with a job at a prestigious place but with wage so low that my boss
>>apologized several times for the level.
>>
>>Oh, right, 1960s.
>
>On 1960s and the cost of colleges:
>I matriculated at Rice U. (Houston) in 1964 when the tuition
>was zero, as dictated by W.M. Rice whose estate founded it.
>A few years later, trustees got the courts to break that
>condition so (they said) they could collect funding from Merit
>Scholarships and others which provided funding according to
>'need'. (They had previously broken conditions to allow them
>to admit women and Blacks.)
>
>They stated that full 'scholarships' would ensure that no
>student was turned away by cost.
>
>Not long after that, I was aware that other elite universities
>made similar statements, and I read that almost all students
>attended them with some reduction in fees.
>
>Some time along the line, "able to attend" got subverted by
>the introduction of loans in place of scholarships.
>
>And then, for some varieties of loans, the government got
>involved in guarantees, where the political bargaining led to
>HIGH interest rates -- which would NOT be reduced while the
>loan was being paid off. I know that there was some political
>argument when interest rates fell and students were NOT
>allowed to re-finance. Paying 8% interest or more seemed
>unfair when the going rates were half that.
>
>So far as I know, the loans that Biden is causing to be 'forgiven'
>fall into that category -- 'unfair' loans on which students could
>barely pay more than the annual interest.
>
>Oh, I transferred to UT (Austin) where the tuition soon was
>doubled -- from $50 to $100 per semester for in-state students,
>and from $100 to $200 for out-of-state. Fees that low were
>not common, and drew smart applicants from all over. Graduates
>who did not move away left Austin ripe for the high-tech-growth
>that followed.

Tuition at Indiana University in 1956 was $120 per semester for a full
load. I lived in a fraternity house, but was not charged "room"
because I worked as a dishwasher in the house kitchen. I also worked
as a waiter at a nearby sorority house, so meals were free. It was a
rush to get from dinner at the sorority to my fraternity house to be a
dishwasher, but it worked out. The others on the crew would do the
dishes and silverware and leave the pots and pans to me.

In article <gi8ktildg35im2c690dp2f7m07u3nh0kh9@4ax.com>,
Rich Ulrich <rich.ulrich@comcast.net> wrote:

>Not long after that, I was aware that other elite universities
>made similar statements, and I read that almost all students
>attended them with some reduction in fees.

I went to the University of Vermont and State Agricultural College,
widely described as a "public Ivy" (but we're in America East, not the
Ivy League). At the time (ca. 1990) in-state tuition was considered
quite high -- at $2200 per semester it was still higher than most
other state schools -- but out-of-state tuition was several thousand
more, the most expensive state school for out-of-state students in the
country. Most private schools were then charging around $15,000 a
year ($7,500 per semester, if they were on the semester system). At
the time, students mostly could not get loans without a responsible
and credit-worthy co-signer, except for the federally guaranteed
Stafford (R-VT) Loan program, which was limited to $2,800 a year and
"need-based".

A couple of things happened. The first was that the bankruptcy code
was rewritten, and in the course of that "reform", student loans were
made non-dischargeable, which meant that private lenders were much
more willing to offer loans to students who perforce had no credit
history and no immediate job prospects. The second was a series of
scandals in university grant accounting, which resulted in much
stricter rules on what expenses could be charged to overhead[1] and
when, and more frequent and more stringent auditing. These changes in
accounting required universities to more explicitly and directly
account for things like graduate student and faculty dependent
tuition, both to demonstrate that they were reasonable and also to
prove that the school was not shifting financial burdens onto the
government, and this incentivized increases in "headline" tuition
which could be made up for by individual financial aid.

-GAWollman

[1] In grant accounting, there are two kinds of costs: "direct costs"
are the costs actually incurred in performance of the grant objectives
-- equipment, staffing, travel, publication fees, cloud computing,
etc. "Indirect costs" are the costs attributable to running a
university that aren't direct costs of the educational mission: paying
the accountants, administrators, and auditors, rent, utilities,
cleaning, business services and software, campus networks and campus
police. The indirect costs are apportioned among the university's
operations -- teaching, grant-funded research, sports, student
activities, the radio station -- in proportion to dollar value or some
other formula agreed between the school and the government agency
responsible for auditing it.[2] Some indirect costs are not allowable
on federal grants, and the institution must find another source or use
internally generated funds, like event fees or patent licensing
revenue, to fund them.

[2] The department with the plurality of grant value at an institution
-- usually the Department of Health and Human Services, because most
research institutions' largest funder is NIH -- is responsible for
leading the audit and for making determinations about which costs are
allowable.
--
Garrett A. Wollman | "Act to avoid constraining the future; if you can,
wollman@bimajority.org| act to remove constraint from the future. This is
Opinions not shared by| a thing you can do, are able to do, to do together."
my employers. | - Graydon Saunders, _A Succession of Bad Days_ (2015)

On 24/02/24 16:33, Mike Spencer wrote:
> Peter Moylan <peter@pmoylan.org.invalid> writes:
>
>> That's one story. But the national economy suffers when only the
>> rich go to university, so that a lot of hidden talent is wasted.
>
> Now, in the US anyhow, the non-rich can go to university, are led to
> believe that they *must* go to university if they want to remain
> non-rich rather than destitute and homeless -- but in exchange for
> debt indenture for decades or life.

Unfortunately that has now happened here too, over more or less the same
time period as fees went from negligible to high. That means crippling
student debt for a much larger group. In my day only the intellectually
inclined went to university. Now the dominant belief is that everyone
should go to university. That has had several bad effects, including:

(a) university courses being dumbed down. Universities deny that,
but it's pretty obvious to me.
(b) a shortage of skilled tradespeople: those who get their
education by a combination of technical colleges and
apprenticeships.
(c) a distortion of the housing market. Graduates used to
be able to get a housing loan not terribly long after
graduation, but now they have to pay off their
student debt first. That means a much larger group
that is struggling with rising rents. Meanwhile, the housing
market is now dominated by those with enough assets
that they don't need a bank loan, and that is driving up
the prices and therefore the rents.
(d) sorry, I've forgotten what (d) was going to be.

Student loans here are paid off through the tax system, with payments
imposed once the person's income goes over a certain threshold. This
system is based on the theory that university graduates have higher
incomes, an assumption that possibly is no longer true.

--
Peter Moylan http://www.pmoylan.org
Newcastle, NSW

On 24-Feb-24 22:29, Peter Moylan wrote:
> On 24/02/24 16:33, Mike Spencer wrote:
>> Peter Moylan <peter@pmoylan.org.invalid> writes:
>>
>>> That's one story. But the national economy suffers when only the
>>> rich go to university, so that a lot of hidden talent is wasted.
>>
>> Now, in the US anyhow, the non-rich can go to university, are led to
>> believe that they *must* go to university if they want to remain
>> non-rich rather than destitute and homeless -- but in exchange for
>> debt indenture for decades or life.
>
> Unfortunately that has now happened here too, over more or less the same
> time period as fees went from negligible to high. That means crippling
> student debt for a much larger group. In my day only the intellectually
> inclined went to university. Now the dominant belief is that everyone
> should go to university. That has had several bad effects, including:
>
> (a) university courses being dumbed down. Universities deny that,
>      but it's pretty obvious to me.
> (b) a shortage of skilled tradespeople: those who get their
>      education by a combination of technical colleges and
>      apprenticeships.
> (c) a distortion of the housing market. Graduates used to
>      be able to get a housing loan not terribly long after
>      graduation, but now they have to pay off their
>      student debt first. That means a much larger group
>      that is struggling with rising rents. Meanwhile, the housing
>      market is now dominated by those with enough assets
>      that they don't need a bank loan, and that is driving up
>      the prices and therefore the rents.
> (d) sorry, I've forgotten what (d) was going to be.
>
> Student loans here are paid off through the tax system, with payments
> imposed once the person's income goes over a certain threshold. This
> system is based on the theory that university graduates have higher
> incomes, an assumption that possibly is no longer true.
>
d) Increased costs of a university education were 'justified' because of
the increased lifetime earnings of graduates.
No mention of the degrees which don't readily map onto well paid
professions.
Also, in many fields where a degree once lead to rapid advancement, it
now is just the entry level qualification.

--
Sam Plusnet

On Sat, 24 Feb 2024 20:58:48 -0000 (UTC), wollman@bimajority.org
(Garrett Wollman) wrote:

>In article <gi8ktildg35im2c690dp2f7m07u3nh0kh9@4ax.com>,
>Rich Ulrich <rich.ulrich@comcast.net> wrote:
>
>>Not long after that, I was aware that other elite universities
>>made similar statements, and I read that almost all students
>>attended them with some reduction in fees.
>
>I went to the University of Vermont and State Agricultural College,
>widely described as a "public Ivy" (but we're in America East, not the
>Ivy League). At the time (ca. 1990) in-state tuition was considered
>quite high -- at $2200 per semester it was still higher than most
>other state schools -- but out-of-state tuition was several thousand
>more, the most expensive state school for out-of-state students in the
>country. Most private schools were then charging around $15,000 a
>year ($7,500 per semester, if they were on the semester system). At
>the time, students mostly could not get loans without a responsible
>and credit-worthy co-signer, except for the federally guaranteed
>Stafford (R-VT) Loan program, which was limited to $2,800 a year and
>"need-based".
>
>A couple of things happened. The first was that the bankruptcy code
>was rewritten, and in the course of that "reform", student loans were
>made non-dischargeable, which meant that private lenders were much
>more willing to offer loans to students who perforce had no credit
>history and no immediate job prospects. The second was a series of
>scandals in university grant accounting, which resulted in much
>stricter rules on what expenses could be charged to overhead[1] and
>when, and more frequent and more stringent auditing. These changes in

Most of my salary as a data analyst/statistician on research projects
was paid as a line item on various NIH-NIMH research grants.
I also put in hours as a departmental resource, supporting a few
projects and overseeing departmental computing on the mainframe.
That justified the rest of my salary being paid from "overhead" on
grants. For some time (no recent knowledge) (1980s - 2005)
Western Psychiatric Institute and Clinic was #2 or #3 among
institutions each year in dollars in grants from NIMH.

I was surprised when I first heard that some grants received
"100% overhead" -- That is, a $3M grant to the PI gave Pitt
$6M. And (1990s?) some big recipients were trying for more.
I assume that some broke the 100% barrier, but I don't recall
later discussions on the subject.

As to accountability: I recall the first spasm of that occurring
when Jimmy Carter was President (1976-1980), "Circular 21",
IIRC. That tended to screw up the way that the Pitt Computer
Center received funding since "attributing costs" was largely
an exercise in imagination.

Other accountability: For a few years, I was supposed to
"account for my time" by two different standards. One was
for the grant funding, the other was for someone else's
demand -- The two systems had to have a lot of overlap in
theory, but in practice, the terms and definitions were set
up so that even if they got hold of both sets, no one could
use one set of numbers to prove the other set wrong.

>accounting required universities to more explicitly and directly
>account for things like graduate student and faculty dependent
>tuition, both to demonstrate that they were reasonable and also to
>prove that the school was not shifting financial burdens onto the
>government, and this incentivized increases in "headline" tuition
>which could be made up for by individual financial aid.
>
>-GAWollman
>
>[1] In grant accounting, there are two kinds of costs: "direct costs"
>are the costs actually incurred in performance of the grant objectives
>-- equipment, staffing, travel, publication fees, cloud computing,
>etc. "Indirect costs" are the costs attributable to running a
>university that aren't direct costs of the educational mission: paying
>the accountants, administrators, and auditors, rent, utilities,
>cleaning, business services and software, campus networks and campus
>police. The indirect costs are apportioned among the university's
>operations -- teaching, grant-funded research, sports, student
>activities, the radio station -- in proportion to dollar value or some
>other formula agreed between the school and the government agency
>responsible for auditing it.[2] Some indirect costs are not allowable
>on federal grants, and the institution must find another source or use
>internally generated funds, like event fees or patent licensing
>revenue, to fund them.
>
>[2] The department with the plurality of grant value at an institution
>-- usually the Department of Health and Human Services, because most
>research institutions' largest funder is NIH -- is responsible for
>leading the audit and for making determinations about which costs are
>allowable.

As it happens, the top bureaucrats at NIH and NIMH are drawn from
academia -- teaching hospitals, they are called-- especially from
those with experience with research. As you might guess, they do
have friends and they do have sympathies with the folks they are
overseeing.

Now I am thinking of a comment I once read about management
in general - It might have been one of the pop books. The
comment was about how most 'management' across industries is
very similar, with similar problems. Where you get some big
complications is in education. And also when research is involved.

--
Rich Ulrich

On 23/02/2024 12:20, Snidely wrote:
> Lo, on the 2/15/2024, occam did proclaim ...
>
> [elision runs rampant]
>
>> As it turned out, I did opt to sit in a couple of undergraduate AI
>> courses in a different part of the University (of London), purely out of
>> interest. 45 years ago AI (or Machine Learning) was not as fashionable
>> as it is today.
>
> Was that all if-then rules-based, or had it gotten to fuzzy logic by then?

Both were taught. However it was all classical AI ('let's do it with
human logic') rather than the current neural network approach, which is
all probability matrices, but which seems to work better.

>
> My formal AI training was basically like learning geography by viewing a
> set of vacation slides borrowed from the instructor's friends.  I didn't
> even get around to trying Prolog or turtle graphics.
>

Sam Plusnet wrote:

> On 24-Feb-24 22:29, Peter Moylan wrote:
>> On 24/02/24 16:33, Mike Spencer wrote:
>>> Peter Moylan <peter@pmoylan.org.invalid> writes:
>>>
>>>> That's one story. But the national economy suffers when only the
>>>> rich go to university, so that a lot of hidden talent is wasted.
>>>
>>> Now, in the US anyhow, the non-rich can go to university, are led to
>>> believe that they *must* go to university if they want to remain
>>> non-rich rather than destitute and homeless -- but in exchange for
>>> debt indenture for decades or life.
>>
>> Unfortunately that has now happened here too, over more or less the same
>> time period as fees went from negligible to high. That means crippling
>> student debt for a much larger group. In my day only the intellectually
>> inclined went to university. Now the dominant belief is that everyone
>> should go to university. That has had several bad effects, including:
>>
>> (a) university courses being dumbed down. Universities deny that,
>>      but it's pretty obvious to me.
>> (b) a shortage of skilled tradespeople: those who get their
>>      education by a combination of technical colleges and
>>      apprenticeships.
>> (c) a distortion of the housing market. Graduates used to
>>      be able to get a housing loan not terribly long after
>>      graduation, but now they have to pay off their
>>      student debt first. That means a much larger group
>>      that is struggling with rising rents. Meanwhile, the housing
>>      market is now dominated by those with enough assets
>>      that they don't need a bank loan, and that is driving up
>>      the prices and therefore the rents.
>> (d) sorry, I've forgotten what (d) was going to be.
>>
>> Student loans here are paid off through the tax system, with payments
>> imposed once the person's income goes over a certain threshold. This
>> system is based on the theory that university graduates have higher
>> incomes, an assumption that possibly is no longer true.
>>
> d) Increased costs of a university education were 'justified' because of
> the increased lifetime earnings of graduates.

> No mention of the degrees which don't readily map onto well paid
> professions.

Some people don't think the taxpayers should be supporting degrees
like that. This site doesn't make that argument, but it does advocate
government support of higher education in completely financial terms.

https://www.aplu.org/our-work/4-policy-and-advocacy/publicuvalues/societal-benefits/

(I'm surprised at the low number it gives for the government
contribution per student.)

> Also, in many fields where a degree once lead to rapid advancement, it
> now is just the entry level qualification.

As I happen to know.

--
Jerry Friedman

occam formulated the question :
> On 23/02/2024 12:20, Snidely wrote:
>> Lo, on the 2/15/2024, occam did proclaim ...
>>
>> [elision runs rampant]
>>
>>> As it turned out, I did opt to sit in a couple of undergraduate AI
>>> courses in a different part of the University (of London), purely out of
>>> interest. 45 years ago AI (or Machine Learning) was not as fashionable
>>> as it is today.
>>
>> Was that all if-then rules-based, or had it gotten to fuzzy logic by then?
>
> Both were taught. However it was all classical AI ('let's do it with
> human logic') rather than the current neural network approach, which is
> all probability matrices, but which seems to work better.

IIRC, early neural nets aimed for "tens of nodes".

>> My formal AI training was basically like learning geography by viewing a
>> set of vacation slides borrowed from the instructor's friends.  I didn't
>> even get around to trying Prolog or turtle graphics.
>>

-d

--
"Inviting people to laugh with you while you are laughing at yourself
is a good thing to do, You may be a fool but you're the fool in
charge." -- Carl Reiner

On 26/02/24 03:17, jerryfriedman wrote:
> Sam Plusnet wrote:

>> d) Increased costs of a university education were 'justified'
>> because of the increased lifetime earnings of graduates.
>
>> No mention of the degrees which don't readily map onto well paid
>> professions.
>
> Some people don't think the taxpayers should be supporting degrees
> like that. This site doesn't make that argument, but it does
> advocate government support of higher education in completely
> financial terms.
>
> https://www.aplu.org/our-work/4-policy-and-advocacy/publicuvalues/societal-benefits/
>
> (I'm surprised at the low number it gives for the government
> contribution per student.)
>
>> Also, in many fields where a degree once lead to rapid advancement,
>> it now is just the entry level qualification.
>
> As I happen to know.

Australia's previous government revised the student fee structure, with
the stated aim of having the lowest fees for the degrees that gives the
greatest employability. (And the indirect aim of making "less useful"
degrees less attractive.) The lowest fees are for the following group.

Band 1: Agriculture, English, mathematics, teaching, clinical
psychology, languages, nursing

source:
<https://www.futurityinvest.com.au/insights/futurity-blog/2023/01/10/what-is-the-cost-of-university-education-in-australia>

Their assumptions about employability seem to be shaky, in my opinion.

This is for Australian citizens. Foreign students are charged much
higher fees.

The present government has not revised this system, as far as I know. It
has, however, introduced a waiver of fees for "priority" technical
college courses, where there is a labour shortage. I think this also
extends to nursing degrees, but I'm not sure.

--
Peter Moylan http://www.pmoylan.org
Newcastle, NSW

On 26/02/24 08:26, Snidely wrote:
> occam formulated the question :
>> On 23/02/2024 12:20, Snidely wrote:
>>> Lo, on the 2/15/2024, occam did proclaim ...
>>>
>>> [elision runs rampant]
>>>
>>>> As it turned out, I did opt to sit in a couple of undergraduate
>>>> AI courses in a different part of the University (of London),
>>>> purely out of interest. 45 years ago AI (or Machine Learning)
>>>> was not as fashionable as it is today.
>>>
>>> Was that all if-then rules-based, or had it gotten to fuzzy logic
>>> by then?
>>
>> Both were taught. However it was all classical AI ('let's do it
>> with human logic') rather than the current neural network approach,
>> which is all probability matrices, but which seems to work better.
>
> IIRC, early neural nets aimed for "tens of nodes".

Back before neural nets became fashionable, I wrote a computer program
to predict whether the user was going to type a 0 or a 1 as the next
input. Its success rate was of the order of 70%. It worked with an array
of finite automata to do the predictions. At each round, the automaton
with the lowest success rate was deleted, and was replaced by a copy of
the most successful one, modified by random variations.

I think I had something like 20 automata in my machine, or possibly as
many as 50. The number was limited by the size of available memory.

As I recall it, I improved the success rate by dividing the automata
into male and female, and using sexual reproduction. But this was just
for my amusement, not a serious research project, so I didn't carry it
any further.

--
Peter Moylan http://www.pmoylan.org
Newcastle, NSW

In article <pkkltitfv48fa6ir1vr241fph18oudtj51@4ax.com>,
Rich Ulrich <rich.ulrich@comcast.net> wrote:

>I was surprised when I first heard that some grants received
>"100% overhead" -- That is, a $3M grant to the PI gave Pitt
>$6M. And (1990s?) some big recipients were trying for more.
>I assume that some broke the 100% barrier, but I don't recall
>later discussions on the subject.

Oh very much so. For every dollar brought in to MIT, the Institute
keeps 65 cents, and the researcher notionally gets to spend 35. But
the big catch here is that certain expenses -- notably, capital
equipment -- do not attract overhead, so buying a $100,000 computer
costs the grant account the same as paying a lab assistant $35,000.

For a long time our capital equipment threshold was only $2000, so
there was a strong incentive to load up laptops and workstations with
the fanciest displays and the most memory available, because a $5000
computer cost less than an equally suitable $1999 computer. (The
threshold has since been raised to $5000, but individual graphics
cards can cost more than that now.)

-GAWollman

--
Garrett A. Wollman | "Act to avoid constraining the future; if you can,
wollman@bimajority.org| act to remove constraint from the future. This is
Opinions not shared by| a thing you can do, are able to do, to do together."
my employers. | - Graydon Saunders, _A Succession of Bad Days_ (2015)

On 26/02/2024 00:18, Peter Moylan wrote:
> On 26/02/24 08:26, Snidely wrote:
>> occam formulated the question :
>>> On 23/02/2024 12:20, Snidely wrote:
>>>> Lo, on the 2/15/2024, occam did proclaim ...
>>>>
>>>> [elision runs rampant]
>>>>
>>>>> As it turned out, I did opt to sit in a couple of undergraduate
>>>>> AI courses in a different part of the University (of London),
>>>>> purely out of interest. 45 years ago AI (or Machine Learning)
>>>>> was not as fashionable as it is today.
>>>>
>>>> Was that all if-then rules-based, or had it gotten to fuzzy logic
>>>> by then?
>>>
>>> Both were taught. However it was all classical AI ('let's do it
>>> with human logic') rather than the current neural network approach,
>>> which is all probability matrices, but which seems to work better.
>>
>> IIRC, early neural nets aimed for "tens of nodes".

Yes, they were 'flat' neural networks and they did not have feedback
loops. In fact, there is a famous analysis of early neural networks by
Minsky and Papert that proved that such neural networks ('perceptrons')
would not work 100%.

<https://direct.mit.edu/books/monograph/3132/PerceptronsAn-Introduction-to-Computational>

(The first paragraph of that page is worth reading. It probably dampened
the advance in AI by a decade or two - by diverting research effort in a
different direction.)

>
> Back before neural nets became fashionable, I wrote a computer program
> to predict whether the user was going to type a 0 or a 1 as the next
> input. Its success rate was of the order of 70%. It worked with an array
> of finite automata to do the predictions. At each round, the automaton
> with the lowest success rate was deleted, and was replaced by a copy of
> the most successful one, modified by random variations.

The approach is interesting as it is based on an 'evolutionary' model.

>
> I think I had something like 20 automata in my machine, or possibly as
> many as 50. The number was limited by the size of available memory.
>
> As I recall it, I improved the success rate by dividing the automata
> into male and female, and using sexual reproduction. But this was just
> for my amusement, not a serious research project, so I didn't carry it
> any further.
>

Oooh, that sounds like the first case of built-in gender-bias into an AI
algorithm!

Had you pursued that approach, you would have had to revise it later by
creating LGBTQ+ automata also. </smile >

After serious thinking Peter Moylan wrote :

> Back before neural nets became fashionable, I wrote a computer program
> to predict whether the user was going to type a 0 or a 1 as the next
> input. Its success rate was of the order of 70%. It worked with an array
> of finite automata to do the predictions. At each round, the automaton
> with the lowest success rate was deleted, and was replaced by a copy of
> the most successful one, modified by random variations.

What, essentially matching a regular expression?

/dps

--
Why would I want to be alone with my thoughts?
Have you heard some of the shit that comes out of my mouth?
-- the World Wide Web

On 26/02/24 20:50, Snidely wrote:
> After serious thinking Peter Moylan wrote :
>
>> Back before neural nets became fashionable, I wrote a computer
>> program to predict whether the user was going to type a 0 or a 1 as
>> the next input. Its success rate was of the order of 70%. It worked
>> with an array of finite automata to do the predictions. At each
>> round, the automaton with the lowest success rate was deleted, and
>> was replaced by a copy of the most successful one, modified by
>> random variations.
>
> What, essentially matching a regular expression?

A single FA will match a regular expression. The story becomes more
complicated if you're running many of them in parallel, voting on what
the answer should be. Well, now that I think of it, that creates a
nondeterministic finite automaton, which still would match a regular
expression if you kept the structure constant. But if components are
created and destroyed at each time step, with random variations, the
overall structure is no longer a NDFA.

--
Peter Moylan http://www.pmoylan.org
Newcastle, NSW

On 26/02/24 19:09, occam wrote:
> On 26/02/2024 00:18, Peter Moylan wrote:

>> I think I had something like 20 automata in my machine, or possibly
>> as many as 50. The number was limited by the size of available
>> memory.
>>
>> As I recall it, I improved the success rate by dividing the
>> automata into male and female, and using sexual reproduction. But
>> this was just for my amusement, not a serious research project, so
>> I didn't carry it any further.
>
> Oooh, that sounds like the first case of built-in gender-bias into an
> AI algorithm!
>
> Had you pursued that approach, you would have had to revise it later
> by creating LGBTQ+ automata also. </smile >

That possibility didn't occur to me at the time. I did, however,
experiment with varying the number of sexes. My conclusion at the time
was that the optimal number of sexes is three, but that was based on
limited evidences.

Of course, we have always known that if a species has more than two
sexes it becomes harder to find partners. You can probably have a viable
species with three sexes, but with any more the reproduction rate is
likely to collapse.

--
Peter Moylan http://www.pmoylan.org
Newcastle, NSW

Peter Moylan <peter@pmoylan.org.invalid> wrote:

> On 26/02/24 20:50, Snidely wrote:
> > After serious thinking Peter Moylan wrote :
> >
> >> Back before neural nets became fashionable, I wrote a computer
> >> program to predict whether the user was going to type a 0 or a 1 as
> >> the next input. Its success rate was of the order of 70%. It worked
> >> with an array of finite automata to do the predictions. At each
> >> round, the automaton with the lowest success rate was deleted, and
> >> was replaced by a copy of the most successful one, modified by
> >> random variations.
> >
> > What, essentially matching a regular expression?
>
> A single FA will match a regular expression. The story becomes more
> complicated if you're running many of them in parallel, voting on what
> the answer should be. Well, now that I think of it, that creates a
> nondeterministic finite automaton, which still would match a regular
> expression if you kept the structure constant. But if components are
> created and destroyed at each time step, with random variations, the
> overall structure is no longer a NDFA.

You remind of another great Daedalus invention: the democratic bus.
Each passenger has a steering wheel,
and the bus goes where the majority want it to go.
(there even is some footage showing it in action)

Somewhat impractical for a bus, obviously, [1]
but it is a good metaphor for a newsgroup such as this one.

It follows respected regular expressions,

Jan

[1] But it might have become practical in the meantime,
with Tesla's self-driving technology, which can force it
to adhere to trafic rules, and the realities of where allowed roads are.

On 26/02/2024 12:29, Peter Moylan wrote:
> On 26/02/24 19:09, occam wrote:
>> On 26/02/2024 00:18, Peter Moylan wrote:
>
>>> I think I had something like 20 automata in my machine, or possibly
>>> as many as 50. The number was limited by the size of available
>>> memory.
>>>
>>> As I recall it, I improved the success rate by dividing the
>>> automata into male and female, and using sexual reproduction. But
>>> this was just for my amusement, not a serious research project, so
>>> I didn't carry it any further.
>>
>> Oooh, that sounds like the first case of built-in gender-bias into an
>> AI algorithm!
>>
>> Had you pursued that approach, you would have had to revise it later
>> by creating LGBTQ+  automata also. </smile >
>
> That possibility didn't occur to me at the time. I did, however,
> experiment with varying the number of sexes. My conclusion at the time
> was that the optimal number of sexes is three, but that was based on
> limited evidences.
>
> Of course, we have always known that if a species has more than two
> sexes it becomes harder to find partners.

We have? Can you point to an SF story that explores this scenario?

> You can probably have a viable
> species with three sexes, but with any more the reproduction rate is
> likely to collapse.
>

Again, I'm baffled at the conclusion. What if the third sex is type 'O'
(as in blood types) which is compatible with both 'M' and 'F'? Why would
this reduce the reproduction rate?

On 2024-02-26 02:09, occam wrote:
> On 26/02/2024 00:18, Peter Moylan wrote:
>> On 26/02/24 08:26, Snidely wrote:
>>> occam formulated the question :
>>>> On 23/02/2024 12:20, Snidely wrote:
>>>>> Lo, on the 2/15/2024, occam did proclaim ...
>>>>>
>>>>> [elision runs rampant]
>>>>>
>>>>>> As it turned out, I did opt to sit in a couple of undergraduate
>>>>>> AI courses in a different part of the University (of London),
>>>>>> purely out of interest. 45 years ago AI (or Machine Learning)
>>>>>> was not as fashionable as it is today.
>>>>>
>>>>> Was that all if-then rules-based, or had it gotten to fuzzy logic
>>>>> by then?
>>>>
>>>> Both were taught. However it was all classical AI ('let's do it
>>>> with human logic') rather than the current neural network approach,
>>>> which is all probability matrices, but which seems to work better.
>>>
>>> IIRC, early neural nets aimed for "tens of nodes".
>
> Yes, they were 'flat' neural networks and they did not have feedback
> loops. In fact, there is a famous analysis of early neural networks by
> Minsky and Papert that proved that such neural networks ('perceptrons')
> would not work 100%.
>
> <https://direct.mit.edu/books/monograph/3132/PerceptronsAn-Introduction-to-Computational>
>
> (The first paragraph of that page is worth reading. It probably dampened
> the advance in AI by a decade or two - by diverting research effort in a
> different direction.)
>
>>
>> Back before neural nets became fashionable, I wrote a computer program
>> to predict whether the user was going to type a 0 or a 1 as the next
>> input. Its success rate was of the order of 70%. It worked with an array
>> of finite automata to do the predictions. At each round, the automaton
>> with the lowest success rate was deleted, and was replaced by a copy of
>> the most successful one, modified by random variations.
>
> The approach is interesting as it is based on an 'evolutionary' model.
>
>>
>> I think I had something like 20 automata in my machine, or possibly as
>> many as 50. The number was limited by the size of available memory.
>>
>> As I recall it, I improved the success rate by dividing the automata
>> into male and female, and using sexual reproduction. But this was just
>> for my amusement, not a serious research project, so I didn't carry it
>> any further.
>>
>
> Oooh, that sounds like the first case of built-in gender-bias into an AI
> algorithm!
>
> Had you pursued that approach, you would have had to revise it later by
> creating LGBTQ+ automata also. </smile >

Heard an interesting comment on the news today. Apparently a Canadian
politician said that biological males should not be allowed in women's
washrooms, regardless of their stated gender.

The person commenting on that statement was the head of an
"LGBTQX-whatever" group, and said that our society was going backward
instead of forward, due to the statements of "a vocal minority".

Made me wonder if she thought she was not a member of a minority.

--
When I was a child, my dad used to hit me with a camera.
I still have flashbacks.

On 26-Feb-24 18:38, lar3ryca wrote:
> On 2024-02-26 02:09, occam wrote:
>> On 26/02/2024 00:18, Peter Moylan wrote:
>>> On 26/02/24 08:26, Snidely wrote:
>>>> occam formulated the question :
>>>>> On 23/02/2024 12:20, Snidely wrote:
>>>>>> Lo, on the 2/15/2024, occam did proclaim ...
>>>>>>
>>>>>> [elision runs rampant]
>>>>>>
>>>>>>> As it turned out, I did opt to sit in a couple of undergraduate
>>>>>>> AI courses in a different part of the University (of London),
>>>>>>> purely out of interest. 45 years ago AI (or Machine Learning)
>>>>>>> was not as fashionable as it is today.
>>>>>>
>>>>>> Was that all if-then rules-based, or had it gotten to fuzzy logic
>>>>>> by then?
>>>>>
>>>>> Both were taught. However it was all classical AI ('let's do it
>>>>> with human logic') rather than the current neural network approach,
>>>>> which is all probability matrices, but which seems to work better.
>>>>
>>>> IIRC, early neural nets aimed for "tens of nodes".
>>
>> Yes, they were 'flat' neural networks and they did not have feedback
>> loops. In fact, there is a famous analysis of early neural networks by
>> Minsky and Papert that proved that such  neural networks ('perceptrons')
>> would not work 100%.
>>
>> <https://direct.mit.edu/books/monograph/3132/PerceptronsAn-Introduction-to-Computational>
>>
>> (The first paragraph of that page is worth reading. It probably dampened
>> the advance in AI by a decade or two - by diverting research effort in a
>> different direction.)
>>
>>>
>>> Back before neural nets became fashionable, I wrote a computer program
>>> to predict whether the user was going to type a 0 or a 1 as the next
>>> input. Its success rate was of the order of 70%. It worked with an array
>>> of finite automata to do the predictions. At each round, the automaton
>>> with the lowest success rate was deleted, and was replaced by a copy of
>>> the most successful one, modified by random variations.
>>
>> The approach is interesting as it is based on an 'evolutionary' model.
>>
>>>
>>> I think I had something like 20 automata in my machine, or possibly as
>>> many as 50. The number was limited by the size of available memory.
>>>
>>> As I recall it, I improved the success rate by dividing the automata
>>> into male and female, and using sexual reproduction. But this was just
>>> for my amusement, not a serious research project, so I didn't carry it
>>> any further.
>>>
>>
>> Oooh, that sounds like the first case of built-in gender-bias into an AI
>> algorithm!
>>
>> Had you pursued that approach, you would have had to revise it later by
>> creating LGBTQ+  automata also. </smile >
>
> Heard an interesting comment on the news today. Apparently a Canadian
> politician said that biological males should not be allowed in women's
> washrooms, regardless of their stated gender.
>
> The person commenting on that statement was the head of an
> "LGBTQX-whatever" group, and said that our society was going backward
> instead of forward, due to the statements of "a vocal minority".
>
> Made me wonder if she thought she was not a member of a minority.
>
Claiming the high ground.

Compare that with calling yourselves "Bolshevik" and labelling the other
lot "Menshevik".

--
Sam Plusnet

On 2024-02-26, occam wrote:

> On 26/02/2024 12:29, Peter Moylan wrote:
>> On 26/02/24 19:09, occam wrote:
>>> On 26/02/2024 00:18, Peter Moylan wrote:
>>
>>>> I think I had something like 20 automata in my machine, or possibly
>>>> as many as 50. The number was limited by the size of available
>>>> memory.
>>>>
>>>> As I recall it, I improved the success rate by dividing the
>>>> automata into male and female, and using sexual reproduction. But
>>>> this was just for my amusement, not a serious research project, so
>>>> I didn't carry it any further.
>>>
>>> Oooh, that sounds like the first case of built-in gender-bias into an
>>> AI algorithm!
>>>
>>> Had you pursued that approach, you would have had to revise it later
>>> by creating LGBTQ+  automata also. </smile >
>>
>> That possibility didn't occur to me at the time. I did, however,
>> experiment with varying the number of sexes. My conclusion at the time
>> was that the optimal number of sexes is three, but that was based on
>> limited evidences.
>>
>> Of course, we have always known that if a species has more than two
>> sexes it becomes harder to find partners.
>
> We have? Can you point to an SF story that explores this scenario?

Dunno, but there was a ST ENT episode about a species with a 3rd
gender who are useful for reproduction but treated as inferiors.

<https://memory-alpha.fandom.com/wiki/Cogenitor_(episode)>

>> You can probably have a viable
>> species with three sexes, but with any more the reproduction rate is
>> likely to collapse.
>>
>
> Again, I'm baffled at the conclusion. What if the third sex is type 'O'
> (as in blood types) which is compatible with both 'M' and 'F'? Why would
> this reduce the reproduction rate?

A "universal donor" that can produce both kinds of sex cells?

I think there are some animals that switch between M & F (probably not
often), but for anything complicated like a humanoid (or any mammal,
really) a reproductive type O would require a lot of equipment with a
production cost [1] that would probably outweigh the evolutionary
benefits.

[1] What's that called in evolutionary biology?

--
Ninety-nine percent of who you are is invisible and untouchable.
--Buckminster Fuller

On 2024-02-26 13:45, Adam Funk wrote:
> On 2024-02-26, occam wrote:
>
>> On 26/02/2024 12:29, Peter Moylan wrote:
>>> On 26/02/24 19:09, occam wrote:
>>>> On 26/02/2024 00:18, Peter Moylan wrote:
>>>
>>>>> I think I had something like 20 automata in my machine, or possibly
>>>>> as many as 50. The number was limited by the size of available
>>>>> memory.
>>>>>
>>>>> As I recall it, I improved the success rate by dividing the
>>>>> automata into male and female, and using sexual reproduction. But
>>>>> this was just for my amusement, not a serious research project, so
>>>>> I didn't carry it any further.
>>>>
>>>> Oooh, that sounds like the first case of built-in gender-bias into an
>>>> AI algorithm!
>>>>
>>>> Had you pursued that approach, you would have had to revise it later
>>>> by creating LGBTQ+  automata also. </smile >
>>>
>>> That possibility didn't occur to me at the time. I did, however,
>>> experiment with varying the number of sexes. My conclusion at the time
>>> was that the optimal number of sexes is three, but that was based on
>>> limited evidences.
>>>
>>> Of course, we have always known that if a species has more than two
>>> sexes it becomes harder to find partners.
>>
>> We have? Can you point to an SF story that explores this scenario?
>
> Dunno, but there was a ST ENT episode about a species with a 3rd
> gender who are useful for reproduction but treated as inferiors.
>
> <https://memory-alpha.fandom.com/wiki/Cogenitor_(episode)>
>
>
>>> You can probably have a viable
>>> species with three sexes, but with any more the reproduction rate is
>>> likely to collapse.
>>>
>>
>> Again, I'm baffled at the conclusion. What if the third sex is type 'O'
>> (as in blood types) which is compatible with both 'M' and 'F'? Why would
>> this reduce the reproduction rate?
>
> A "universal donor" that can produce both kinds of sex cells?
>
> I think there are some animals that switch between M & F (probably not
> often), but for anything complicated like a humanoid (or any mammal,
> really) a reproductive type O would require a lot of equipment with a
> production cost [1] that would probably outweigh the evolutionary
> benefits.

No known mammals, but plenty of fish and invertebrates.

> [1] What's that called in evolutionary biology?

--
rw-rw-rw- Permissions of the beast.