Showing posts with label Anatoly Karlin. Show all posts
Showing posts with label Anatoly Karlin. Show all posts

Thursday, April 06, 2017

Neural lace

Elon Musk tweaks the technosphere again (from The Economist):
"Ever since ENIAC, the first computer that could be operated by a single person, began flashing its ring counters in 1946, human beings and calculating machines have been on a steady march towards tighter integration. Computers entered homes in the 1980s, then migrated onto laps, into pockets and around wrists. In the laboratory, computation has found its way onto molars and into eyeballs. The logical conclusion of all this is that computers will, one day, enter the brain.

"This, at least, is the bet behind a company called Neuralink, just started by Elon Musk, a serial technological entrepreneur. Information about Neuralink is sparse, but trademark filings state that it will make invasive devices for treating or diagnosing neurological ailments. Mr Musk clearly has bigger plans, though. He has often tweeted cryptic messages referring to “neural lace”, a science-fictional concept invented by Iain M. Banks, a novelist, that is, in essence, a machine interface woven into the brain. ..."
Neural lace was Iain M. Banks's technology as used by citizens in "the Culture" to 'telepathically' converse with Culture Minds, the AIs which actually ran their civilisation. In one novel, a Mind mentions matter-of-factly that no more exquisite torture device has ever been conceived of.

Anatoly Karlin, quoting Nick Bostrom, is profoundly skeptical:
"We do not need to plug a fiber optic cable into our brains in order to access the Internet. Not only can the human retina transmit data at an impressive rate of nearly 10 million bits per second, but it comes pre-packaged with a massive amount of dedicated wetware, the visual cortex, that is highly adapted to extracting meaning from this information torrent and to interfacing with other brain areas for further processing.

"Even if there were an easy way of pumping more information into our brains, the extra data inflow would do little to increase the rate at which we think and learn unless all the neural machinery necessary for making sense of the data were similarly upgraded. Since this includes almost all of the brain, what would really be needed is a “whole brain prosthesis–—which is just another way of saying artificial general intelligence.

"Yet if one had a human-level AI, one could dispense with neurosurgery: a computer might as well have a metal casing as one of bone."
This reminds me of all the ways experts tell you that something can't be done.

My vote goes with James P. Hogan, who in his excellent and seminal SF novel, "The Genesis Machine", described a brain-computer symbiosis delivering enhanced imagination/visualisation.

Amazon link

A system of equations would be projected onto an interior whiteboard in your mind. The computer would generate a VR-type solution space which you could navigate freely while discussing subtleties and points of interest with the accompanying AI system. It was, if you like, virtual reality in the head - the abstract rendered concrete.

And that is definitely going to happen.

Friday, March 03, 2017

Diary: programming is hard!

I remember a post by James Thompson (can't find it - sorry) describing the joy of giving IQ tests to bright students. Their smug condescension as they breeze through the early items; their horror as they hit their cognitive limits and start to sweat: "Wait a mo, .., just can't seem to see it ... ."

In artificial intelligence textbooks it's standard to cover the coding of theorem provers for propositional-calculus. Hidden away in the advanced exercises is the suggestion to extend the program to cover the full predicate-calculus. Inevitably, this is flagged as hard.

Tell me about it. Today I got basic binary resolution to work. Great! I can do inference. And yet this is not the central mystery of a predicate-calculus resolution theorem prover (RTP).

The conceptual problems arise from the layers of abstraction around the proof procedure. At the risk of boring you even further, binary resolution is driven by the unification of complementary literals (basic formulae). These two literals occur in the two clauses offered to the resolution function (hence binary resolution).

A  problem addressed by an RTP comprises: assumptions  = the axioms, a set of clauses, plus the problem itself, a goal clause.

Here's my working example. First come the axiom clauses:
(defvar *c3*   (mk-fact-clause '(likes rod horvath)))
(defvar *c4*   (mk-fact-clause '(likes sally renner)))
(defvar *c5*   (mk-fact-clause '(likes sally rod)))
(defvar *c6*   (mk-fact-clause '(likes hardy renner)))
(defvar *c7*   (mk-fact-clause '(likes hardy rod)))
(defvar *c8*   (mk-fact-clause '(amusing hardy)))
(defvar *c9*   (mk-fact-clause '(likes horvath moties)))
(defvar *c10* (mk-rule-clause '(likes sally ?x) '((likes ?x moties)) ) )
(defvar *c11* (mk-rule-clause '(likes rod ?x) '((likes ?x renner) (likes ?x rod)) ) )
(defvar *c12* (mk-rule-clause '(likes ?x ?y) '((amusing ?y) (likes rod ?y)) ) )
(defvar *c13* (mk-fact-clause '(likes ?x ?x)))

(defvar *g1* (mk-goal-clause '((likes sally ?who)) ) )  ; this is the problem clause
You may notice a "Mote in God's Eye" theme here.

Despite the simplicity of these facts and rules, you'll struggle a bit to figure out the complete list of who, exactly, Sally Fowler likes.

I expect better from my program - once completed.

---

So a clause might look like this (which you will recognise as a Prolog-style Horn-clause):
'( (likes sally ?x) ← ((likes ?x ?y) (likes ?y moties)) )   ; a rule clause
which is already a bit complicated.

Then we have to set up structures for the knowledge-base (facts and rules), the goals which emerge from the resolution process, and the pointers and bindings which allow a reconstruction of proofs.

---

In telling you this, my intention is not to document the program, my point is more psychological.

You need to keep a very great deal of abstract structure in your head simultaneously if you hope to put this not-entirely-trivial program together. Plus lots of functions.

It's very g-loaded, much like doing mathematics.
"It's quite hard getting figures but ... it appears that the average IQ of students applying for PhD programmes at good universities in computer science is around 128.

"That sounds right to me. Most software development companies don't take people without good first degrees or a higher degree in a STEM subject - and that's just the pool we're talking about here."
---

Parenthetically, thank God for Lisp, a gift from the deity for clear, conceptual, exploratory thinking.

---

I read endless texts in the media from innumerate journalists about how unemployed or newly-redundant workers are going to retrain as software developers. Those writers couldn't hack it themselves and sadly, neither will most of those no-longer-required workers.
"Wait a mo, ..., just can't seem to see it ... ."
At an IQ threshold of 128, we're talking about 2.5% of a population normed at 100.

A caveat. It's intellectually hard writing machine learning programs or configuring recalcitrant artificial neural nets on novel problem-domains .. or programming RTPs. Yet for everyone doing those things, there are ten people paid to use graphical web design tools or load databases, or do a bit of scripting. Things which are not that hard.

Like many occupations, there has been mission-creep in the definition of software developer.

---

And why am I making such heavy weather of this wretched theorem-prover? This.

Thursday, June 02, 2016

The aircraft carrier in total war


Interesting discussion about that exciting new weapon, the US navy railgun. A lot of feedback in the comments, mostly skeptical, about new 'wunderwaffe'.

I think it boils down to this: against a third-world opponent, nothing works like a carrier group for overwhelming global force projection; conversely, if you're fighting a first-world state with up-to-date weaponry, your carrier group is toast.

First-world state-on-state warfighting doctrine doesn't seem to have kept up with drone & missile technologies. Drones and missiles are orders of magnitude cheaper than the platforms (aircraft, tanks, ships) they target.

Offensive modes involving pre-positioning and saturation-attacks seem impossible to reliably defend against. And if all else fails, comment 13 observes:
"when articles talk about a naval war between China and USA in 2050 using some sexy new weapon, they never seem to cover that old weapon - nuclear ICBMs."
Five missiles in an area-spread with megaton warheads - no carrier group is going to look too clever after that.

---

My favourite fictional account: "Firelance" by David Mace (although the Vindicator is a cruise missile platform, not a carrier).



Thursday, January 15, 2015

Predicting IQ across the world from genotypes

Early days for this - I wrote about it last November, where I tried to 'predict' my own IQ. Now a much better article has appeared, written by Anatoly Karlin. Interesting stuff, highlighting the ground-breaking research of Davide Piffer.