What Would the Father of Cybernetics Think About A.I. Today?
slate.com
slate.com
I think this misunderstands the meaning of "control" in a control system / cybernetic sense, mistaking it for a more popular sense of the term. It's the difference between control-as-behaviour and control-as-power. The latter is strictly about humans, the former can include power structures but extends to any circular causality with at least one balancing loop.
Cybernetics thinkers were well aware of emergence and multi-layered autonomy. Some of them made it a central concept (eg Stafford Beer).
His book Cybernetics even has a whole chapter on random circuits being able to self-organize with some kind of learning algorithm for adjusting weights.
The cybernetic definition of control which covers self-organization in technology is the same one which includes self-organization in cells, brain structures, and even social groups.
They were still working on a planned economy, but their goal was to create a system where economic components could send signals to dictate their requirements upwards, essentially self-organizing necessary inputs. (The differences between that and capitalism being that a central administrator could set priorities, and the value of final goods wasn't to be determined by "what people are willing to pay" but "what's been allotted".)
Any system that can adjust its behavior to circumstances involves 'control'. Really, anything that responds to stimuli at all has some control component. Confusing that with centralization or a lack of autonomy suggests that Lloyd is missing something pretty fundamental about the history of cybernetics.
https://en.wikipedia.org/wiki/Project_Cybersyn
> Project Cybersyn was a Chilean project from 1971–1973 during the presidency of Salvador Allende aimed at constructing a distributed decision support system to aid in the management of the national economy.
> After the CIA-backed military coup on September 11, 1973, Cybersyn was abandoned and the operations room was destroyed.
Today, I think that the opposite may be true. That worries me.
Apparently the main use of computers there was more or less workers collectives using telex to "email" numbers to central.
"Cybernetics" reminds me a bit of "artificial intelligence", in this sense that it's "whatever doesn't work yet".
When Kantorovich brought linear programming to industrial production, it was clearly just a mathematical optimization. When Glushkov proposed running whole planned economies on it, that was cybernetics because the required computations were unachievable. But when businesses and militaries use it for design now, it's just operations theory.
When Jay Wright Forrester applied models of feedback and imperfect information to supply chains, that was simple economics. When he brought them to urban design and the environment, and Beer brought them to business and government management, the appropriate computing power and datasets were unavailable - so that was cybernetics. Now that environmental models are meaningfully grounded in reality, they're back to being mere models.
Of course, I have a bit less sympathy for the cyberneticists than for AI researchers. ML/AI work frequently follows a pattern in which research is explicitly speculative, then realized, then dismissed from the outside as "just algorithms". Its worst excesses are predicting major advances like NLP on very short timelines, not misrepresenting the present. Cybernetics seems more prone to over-promising what it can do now. And rather than researchers making advances which are then dismissed, the common pattern seems to be outsiders implementing techniques, at which point cyberneticists retreat to some new speculative claim where there's no competition.
FWIW, when I talk about Cybernetics I pretty much mean what's in Ashby's "Introduction to Cybernetics". I don't find much value personally in the "second-order" cybernetics and all that.
I figure that Beer believed his own shtick, but, man, you make it sound bad.
One of the chief critiques of central planning has been its inability to adapt to constantly changing conditions and its lack of information about individual preferences.
Amazon is already a centrally planned algorithmic system for delivering goods to people. Data collection is used to personalize someone’s experience in a way that wasn’t possible before. The feedback loops are extremely quick in terms of the ability to do constant A/B tests.
Perhaps USSR 2.0 might have its own amazon app for citizens that would collect preferences for what they want in real time and automatically feed it all through the economy. While also allowing to correct for certain things which the market might not be able to so easily as for eg medicine for poor people is more important than fidget spinners.
You might also be able to experiment with different algorithms to see which of them lead to more growth and productivity.
After a few years you might have enough data to do predictive analytics to improve things ahead of time.
It would never take off because of its totalitarian implications but we might get to witness a real world test of this idea in China in the coming years. Life in China is already quite convenient in ways not possible in the west.
Summarising a complex history, the USSR did have folks who argued for a cybernetic approach to production planning, but it was never tried. An entertaining semi-fictionalised account is Red Plenty. A drastically less entertaining but more scholarly read is Newspeak to Cyberspeak.
Not in the USSR - but there was an attempt elsewhere:
Hope China or anyone else doesn't seriously try running this experiment. Hope someone explains to power hungry rulers that they won't end up ruling or having any power over anything if they succeed in building such monstrous systems...
This is incorrect. There are no “cybernetics” department at universities, nor people with this title or role. However the concepts and ideas from cybernetics have infiltrated and influenced most areas of science. If you trace the ideas you will find them, you will recognize the DNA. The ideas from cybernetics + information theory was a paradigm shift in science at the time. I would say the cybernetic ideas were a victim of its own success. It became subsumed into many other disciplines by other names. It is like a snake eating its own tail. See the Ouroboros symbol [0] .
“Finally, because of his emphasis on control”
Here he is assuming control means too-down with a naive view of control. Control here could be replaced with regulation, direction, steering, facilitation. I would define control as the mechanism that allows a system to achieve it's purpose in an efficient and effective way. This definition says nothing about top-down. In actual fact there is a concept within cybernetics of "self-regulation" or "internal control". This means that control must be exhibited from within the system and not from an external party.
"Wiener could not foresee a technological world in which innovation and self-organization bubble up from the bottom rather than being imposed from the top."
Within the field of there is a ton of material on self-organization. This was one of the topics that was specifically investigated and written about at length. In particular how do you design for "self-organization"? Staford Beer and the organizational cybernetician were specifically focused on maximizing freedom, by balancing of both bottom up, and top-down feedback loops.
While Weiner did give the field the name i would hesitate in calling him the "father" of the field. Alot of the ideas were developed in collaboration with others, particularly at the Macy conferences [1]. The ideas have a long history and cybernetics has been heavily influenced by Ashby, Beer, Shannon, Maturana, Van Forester etc. Cybernetics itself born out of the tradition of operational research and systems engineering.
[0] https://commons.m.wikimedia.org/wiki/File:Serpiente_alquimic...
“The world’s best chess players are neither computers nor humans but humans working together with computers.”
https://deepmind.com/blog/article/alphazero-shedding-new-lig...
I've been developing a startup project building a cybernetic control system in JavaScript that makes extensive use of feedback loops that get dynamically spawned between devices and services and it is truly marvelous to see the full system in action. I'm getting capabilities out of the system that I didn't even foresee being able to do, like being able to write fully voice-controlled chess game:
https://www.youtube.com/watch?v=faOfKd1eAQA
That was just a fun side-project too. With a true cybernetic control layer running between the speech recognition system and the things you want to control building apps like this becomes relatively straight-forward, perhaps not easy but a lot easier than it would be without one.
I am quite busy in a rewrite of the whole system, not a whole lot of documentation yet:
https://github.com/jaxcore/jaxcore
I'll be rewriting the speech recognition system as well and integrating it better into the new version, I wrote a bit of a writeup here about how Mozilla's DeepSpeech can be hooked into the web:
https://github.com/jaxcore/jaxcore-listen
And there's hardware aspect of the project as well, I created a new cybernetic control dial that gets hooked into the Jaxcore to control all sorts of things like stereo equipment, wireless speakers, and can control pretty much anything in a super elegant way.
[1] https://en.wikipedia.org/wiki/Anthony_Stafford_Beer
and his
[2] https://en.wikipedia.org/wiki/Viable_System_Model
Not sure why (not)?
It's an odd line to take, though, since artificial intelligence has never been the sole or even central focus of cybernetics. There was a relatively explicit split in 1956 at the Dartmouth workshop, but the piece follows the AI thread from there on with no real acknowledgement that there was any other path in effect. Beer and Forrester began their cybernetics-related work around 1960, and so the article seems to drop them entirely as no longer interesting. This is convenient for the overall thesis: Lloyd works on the limits of computation and Moore's Law, so he structured this article as "Wiener underestimated advances in raw computing power, but now that's dying down and so he's especially relevant".
But the offhand references to Wiener's relevance to unemployment, human-computer interfacing, environmental disasters, and warfare are cheapened by the speculative "what would Wiener think?" tone. His work has actively shaped the current state of economic planning, environmental modelling, and military strategy, so it doesn't make much sense to read him like an outside commentator on those topics.
Holy shit this author is unutterably clueless. Wiener in fact wrote a whole book more or less on how innovation and self organization bubble up from the bottom. It was called "Invention: The Care and Feeding of Ideas."
> Wiener saw the world in a deeply pessimistic light.
Gee, a Jewish guy in 1950 was pessimistic and worried about totalitarianism. I wonder where he might have acquired this attitude? Maybe, you know, all them people recently killed in WW-2 might have spooked the man?
I used to respect Seth Lloyd ... until he published the ridiculous "on the computational capacity of the universe." Google informs me he also takes money from child molesting gangsters, which sounds about right. Dude should stop grubbing money from pedophiles and read a book.
I'm really interested to know more about the critic of this paper. Please detail or share some links.
(Note: not part of of the academic field, and I really don't care at all about the politics and characters involved, but I am interested a lot in the concepts in the paper)