Super Artificial Intelligence, a naive approach

[fuelair]

And 13 or so others posting a lot in science!!!!! And all demonstrating either a heavy lack of knowledge of English (especially in terms of grammar but not limited to same!!!) usage, spelling and proper arrangement of words in sentences.

 

[The Don]

He moved the goalposts, hoping no-one would notice.:

Gosh, how unprecedented

Suddenly, for the purposes of this calculation, humans have the brain power of a chimp. He has altered the figure from (10^16 to 10^18) to 10^15, suddenly meaning we only have to make machines 10 times more powerful to get to our level.

Of course, as having a brain of power 10^18 was a requirement to qualify as a god, he has just completely undermined his humans-are-gods argument. Humans, chimps, dolphins, dogs, pigs and possibly parrots too* now qualify as gods under the new revised criteria.

* I can't be bothered checking.

Of course regardless of brainpower, cats have always qualified

 

[The Don]

.....and of course Moore's law is not guaranteed to keep on applying:

https://www.technologyreview.com/s/601441/moores-law-is-dead-now-what/

 

[fuelair]

PGJ, what are you hoping for with these threads? Fame? Recognition? Fortune? Friendship? Admiration? Respect?

I fear that whatever you're hoping for, it won't end well.

It already has functionally ended and as you note. And now it is at least 14.

It's like the people who deny Einstein's formula or Evolution because they just know it/they cannot be correct. But ghosts, psi-powers and Bigfeet have to be there because they believe!!!! And, there is among them a difficulty with English as it is properly written for good understandability!!!

My real question in those cases is do they use the words and write the way they do to obscure purposively or do they truly have a poor understanding of the written language?????

 

[abaddon]

... in any case it seems Moore's law no longer applies as we start to impinge on physical limits.

It's interesting that in the OP there was a factor of somewhere between 100 and 10000 between human brain and machines and yet in ProgrammingGodJordan's calculation that suddenly dropped to a factor of 3

What happened ?

He made up some convenient numbers. If we return to the op as he suggests, then he cites...

(iv)
Mankind has already created brain based models that achieve 10^14 of the above total in (iii).

His citation does not support that but let's run with it.

He also says...

(iii)
The creation of general artificial intelligence is so far, mankind's largely pertinent task, and this involves (i), i.e. optimization.

The human brain computes roughly 10^16 to 10^18 synaptic operations per second.

Since we have a range, let's go at the top end at 10¹⁸. Note that 10¹⁵ doesn't get a look-in. Why the top end of the range? Because PGJ wants an AI which at minimum equals human capability.

Inserting the values that PGJ claims in his OP we get...

10¹⁸/10¹⁴ = 2^(n/2)
10⁴ = 2^(n/2)
10000 = 2^(n/2)
Solve for n...

Log₂(10000) = n/2

13.3 = n/2 approx.

n = 26.6 years.

All of which ignores the limitations of Moore's Law.

 

[ProgrammingGodJordan]

... in any case it seems Moore's law no longer applies as we start to impinge on physical limits.

It's interesting that in the OP there was a factor of somewhere between 100 and 10000 between human brain and machines and yet in ProgrammingGodJordan's calculation that suddenly dropped to a factor of 3

What happened ?

The original post pointed to sources, containing the precise values used.
Sources will vary human brain speed from roughly 10^16 to 10^18 synaptic operations per second.

This is why I used the word roughly in the original post.

So, at minimum, we have roughly 10^15, which yields alt least 3 years, which outputd 2020 at least.

 

[ProgrammingGodJordan]

Since we have a range, let's go at the top end at 10¹⁸. Note that 10¹⁵ doesn't get a look-in. Why the top end of the range? Because PGJ wants an AI which at minimum equals human capability.

Inserting the values that PGJ claims in his OP we get...

10¹⁸/10¹⁴ = 2^(n/2)
10⁴ = 2^(n/2)
10000 = 2^(n/2)
Solve for n...

Log₂(10000) = n/2

13.3 = n/2 approx.

n = 26.6 years.

All of which ignores the limitations of Moore's Law.

(A)
The calculation with minimum brain speed (instead of maximum) yields 3 years roughly.

Keep in mind the original post expressed AT LEAST 2020.
HINT: Replace 10^18 with 10^15 roughly.


(B)
If one can read well, one will observe the values in the citation. (So I didn't need to conjure up numbers)

Moore's law runs out roughly a little while after 2020, the minimum calculations align with Moore's law.

 

[Porpoise of Life]

A factor 10 is not 'roughly'. I wouldn't call something operating at 10% human ability 'human level'.
You're makling up your own definitions again, because you've been caught lying.

 

[ProgrammingGodJordan]

A factor 10 is not 'roughly'. I wouldn't call something operating at 10% human ability 'human level'.
You're makling up your own definitions again, because you've been caught lying.

What are your estimations for the human brain speed?

Link your sources, then do the calculations on the minimum.

 

[Aridas]

I aim to minimize my ignorance...

Hmm? Truly? What if I pointed out that Moore's Law is actually about the number of transistors, not speed?

Or that -

HBS=human_brain_speed
CMS=current_machine_speed
n = YEARS_TILL_BRAIN_CHIP/rate
RATE = 2


So,
HBS = 2*10^15 CMS = 6.4*10^14

Disagrees significantly with the range that you cited in your OP?

(iii)
The creation of general artificial intelligence is so far, mankind's largely pertinent task, and this involves (i), i.e. optimization.

The human brain computes roughly 10^16 to 10^18 synaptic operations per second.

An order of magnitude off, at the minimum, notably changes things. Especially when transistors are likely to hit the atomic level in size and be able to shrink no further rather soon, by some estimates, ~2020... and assuming that economic forces don't prevent them from getting there because of prohibitive cost increases.

 

[abaddon]

(A)
The calculation with minimum brain speed (instead of maximum) yields 3 years roughly.

Keep in mind the original post expressed AT LEAST 2020.
HINT: Replace 10^18 with 10^15 roughly.

Nope. 10¹⁵ is well outside your claimed range of 10¹⁶ to 10¹⁸.

Furthermore, since you are suggesting an AI at least equal to human level, we must perforce use the maximum figure. Otherwise, your suggested 10¹⁵ figure is one thousandth what humans are capable of.

(B)
If one can read well, one will observe the values in the citation.

I intentionally quoted your numbers from the OP so that you could not dodge.

 

[ProgrammingGodJordan]

Nope. 10¹⁵ is well outside your claimed range of 10¹⁶ to 10¹⁸.

Furthermore, since you are suggesting an AI at least equal to human level, we must perforce use the maximum figure. Otherwise, your suggested 10¹⁵ figure is one thousandth what humans are capable of.

I intentionally quoted your numbers from the OP so that you could not dodge.

10^15 is listed in the source of the original post
Roughly 10^16 was mentioned, and precise values are in the original post Wikipedia links

 

[abaddon]

What are your estimations for the human brain speed?

Link your sources, then do the calculations on the minimum.

Sorry, no. We are working from YOUR claims.

 

[Aridas]

10^15 is listed in the source of the original post
Roughly 10^16 was mentioned, and precise values are in the original post Wikipedia links

Which link are you talking about? The one to http:// or the one to a wikipedia page that doesn't exist, apparently?

 

[ProgrammingGodJordan]

Hmm? Truly? What if I pointed out that Moore's Law is actually about the number of transistors, not speed?

Or that -



Disagrees significantly with the range that you cited in your OP?

(A)
The artificial synapses correspond with the speed.


(B)
The values mentioned in original post were rough.

Anyways, I was careful to point to original sources, with the precise values.

 

[ProgrammingGodJordan]

Sorry, no. We are working from YOUR claims.

See the Wikipedia links, for the precise values linked in original post.

 

[MikeG]

Sources will vary human brain speed from roughly 10^16 to 10^18 synaptic operations per second.


.........So, at minimum, we have roughly 10^15, which yields alt least 3 years, which outputd 2020 at least.[/U][/B]

Your own claim, which we were working on, was 10^16 to 10^18.

10^15 is not roughly 10^16, nor even faintly 10^18, being one tenth the former, and one thousandth of the latter.

A goldfish is roughly human by that yardstick.

 

[ProgrammingGodJordan]

Which link are you talking about? The one to http:// or the one to a wikipedia page that doesn't exist, apparently?

It appears that was a typo that put two links in one.

Right click and inspect the 10^18 link to see both links in one line.

 

[Porpoise of Life]

10^15 is listed in the source of the original post
Roughly 10^16 was mentioned, and precise values are in the original post Wikipedia links

Where? I can find 2 wiki links in your OP, one explaining orders of magnitude, the other exascale computing. Neither of them mentions synaptic operations per second.
The first one (you've botched your link by the way) does mention the number of synapses in the brain, but that is not the same.

 

[abaddon]

10^15 is listed in the source of the original post
Roughly 10^16 was mentioned, and precise values are in the original post Wikipedia links

Here is the OP. Please point out that figure.

(i)
Life's meaning probably occurs on the horizon of optimization:

(source: mit physicist, Jeremy England proposes new meaning of life)

(ii)
Today, artificial intelligence exceeds mankind in many human, cognitive tasks:

(source: can we build ai without losing control over it?)

(source: the wonderful and terrifying implications of computers that can learn)

(iii)
The creation of general artificial intelligence is so far, mankind's largely pertinent task, and this involves (i), i.e. optimization.

The human brain computes roughly 10^16 to 10^18 synaptic operations per second.

(iv)
Mankind has already created brain based models that achieve 10^14 of the above total in (iii).

If mankind isn't erased (via some catastrophe), on the horizon of Moore's Law, mankind will probably create machines, with human-level brain power (and relevantly, human-like efficiency), by at least 2020.

(v)
Using clues from from quantum mechanics, and modern machine learning, I have composed (am composing) a naive fabric in aims of absorbing some non-trivial intelligence's basis.

Paper + Starting Code (rudimentary): "thought curvature"

(vi)
Criticism is welcome/needed.

BTW, you totally borked your wiki links, but they don't mention such a figure either. Except for this one. Which simply doesn't have anything to do with synaptic anything. And this one, which cites 10¹⁸ as comparable to the human brain.