Oh this sounds like Macgyvering. The Real Life section in the TV tropes article on it has some interesting examples and lists some other cultural takes on it: desenrascanço, débrouillard, Jugaad. It's a skill clever poor people are more likely to have opportunities to hone.
Renewables are indeed vital. At the same time, it feels counterproductive that every discussion on nuclear or renewables ends up casting things as an either or proposition. Nuclear as baseload remains very useful. What's more, we don't just seek to replace current capacity but also to quickly increase generation.
While costs of transmission infrastructure required for country scale (larger distances for lower correlation) energy dispatch are recognized, its more abstract challenges are less well acknowledged. Dispatch at this level is not just about developments in grid integration or hardware like solid state "transformers", it also has a complex routing coordination aspect requiring research in control and even game theory [1].
A lack in wind and solar can sometimes occur simultaneously. Analysis of German wind turbines data observed that experiencing a stretch of almost a week with generation as low as 10% installed capacity was likely within a given year [2]. Surprising/extreme weather events like Europe's recent "wind drought" are rare but there remains a large amount of uncertainty in how changes in climate will affect the tail of this distribution. Tools such as coordinating distributed generation and improvements in storage tech will surely help smooth generation, nuclear is another powerful tool in that toolbox.
I've never met a coherent definition of symbolic AI so I'll just focus on what it contained: some type of search or inference algorithm (iterative deepening, depth first search being major ones) often combined with heuristics written in a programming language like prolog or lisp. Specification + Inference ≊ Specification + Control flow. That means Probabilistic models written in a language like Stan, which is Inference + Specification, fit neatly into so called symbolic AI (which is basically just programming with search/inference).
These search and sampling algorithms still play key roles in game playing AI (chess, poker, Go) and natural language generation. It is the human knowledge, specification and heuristics, part that tends to be more readily replaceable. A lot of control flow and data-structures that powered old AI approaches can be found in databases, compilers, type inference, computer algebra and even the autodiff libraries neural nets are written in.
Video game AI, constraint solving and business rules engines are probably closest to still using the full symbolic approach rather than merely extracting the control flow and structures portion.
We can therefore make a compact prediction: learned approaches replace human written computer programs (specifications, rules systems or heuristics) whenever human contribution is not valuable or is somehow harmful to robustness/generalization.
Portia spiders do demonstrate trial and error learning, quoting the abstract of [1]:
> All species from the jumping spider genus Portia appear to be predators that specialize at preying on other spiders by invading webs and, through aggressive mimicry gaining dynamic fine control over the resident spider’s behavior. There is evidence that P. fimbriata, P. labiata and P. schultzi derive signals by trial and error. Here, we demonstrate that P. africana is another species that uses a trial and error, or generate and test, algorithm when deriving the aggressive-mimicry signals that will be appropriate in different predator–prey encounters.
It turns out that the species with more variation in encountered prey types are more likely to rely on search, varying over possible patterns until a response is received. Other papers show their ability to learn generalizes beyond mimicry of vibrational patterns. They are also capable of deriving and maintaining situation specific attack routes and plans.
I'll also argue that a fully instinctual repertoire, even without learning, should count as intelligence if flexibly deployed. Consider: despite an inability to learn or adjust, a Nash equilibrium approximating poker bot or an Alpha Zero neural network can be described as encompassing a deep instinct of the game that enables intelligent action selection.
Sociality is not the only (and probably not an initial) driver of the evolution of intelligence in animals. Ecological and particularly foraging competency in less predictable environments is another and likely more primary stimulator for intelligence.
Coordinating complex movements is another one that appears to drive intelligence.
Vulnerability in a hostile environment. Together with foraging and coordinating complex movement plans, it is hypothesized that predation pressure when combined with the loss of their shell drove cephalopod intelligence. Trading defense for agility and int would also have provided a competitive advantage when competing for prey.
Fredkin's contribution were to reversible computation and since Quantum computers are a kind of reversible computer, many ideas and gate notions carried over.
An in-between option that is relatively simple and a rather fast (though not as fast as the automata) optimization to levenstein distance and based on Tries is: http://stevehanov.ca/blog/?id=114.
That makes sense but my argument is since further material (some examples which I listed) assumes and builds upon the standard approach, you'll likely be better off taking that path.
By standard approach I mean the typical material covered for someone studying vector calculus properly. This will be stuff like differential forms and the basics of tensors, manifolds and multilinear maps at the undergrad level. Differential geometry and cohomology are examples of courses which build on them.
I agree with you that pseudovectors, cross products and vector calculus are a terribly adhoc way to teach this stuff but a course covering linear algebra with differential forms elegantly unifies, corrects and generalizes them. Standard is also in contrast to the geometric algebra/calculus alternate path.
Differential forms are a particular kind of tensor and tensors can be defined in terms of multilinear maps. As spekcular says, the standard curriculum covers differential forms, tensors and vectors. This entails becoming familiar with multivectors, the wedge product and multilinear algebra, making geometric algebra a relatively small delta to pick up.
On the other hand, the standard course will also prepare you for mathematical topics like lie derivatives, differential geometry and de Rham cohomology.
Other than physics, the standard approach equips you with the mathematical machinery underlying many topics in machine learning and statistics like Hamiltonian monte carlo, automatic differentiation, information geometry and geometric deep learning.
The central advantage of geometric algebra over the standard approach isn't that it's better or more general, it's that pedagogical material for it is generally leagues and magnitudes better than those for the standard course.
I've played many of the author's games and consider them to have better stories than most. I’d have to reach for the likes of Planescape or Disco Elysium to find something I’d consider more compelling.
As you imply, there are many games whose brief summary might sound weird, cheesy or cliche but are actually very well crafted story experiences when played. For me, some would be: Life is Strange, Soma, Deus Ex, Human Revolution, Alpha Centauri, Alan Wake, Control, Quantum Break and Detroit Become Human.
Then there are the Soul Reaver games with middling stories but whose written dialogue and delivery easily put them at the level of great literature for me.
Storywise, some of the author's games would not be out of place if ranked highly amongst such a list. It's like Berkson's paradox, his studio has had staying power but you certainly couldn't point to graphics or unique game mechanics as explanations for why.
>This article seems like a self-congratulating piece
I'd go as far as to argue the author undersells his abilities. His RPG world building and stories are a great deal more entertaining if compared to many fantasy books.
> That doesn't mean AAA studios can't make a good game story,
They could but rarely do. Meanwhile, the majority of those few attempts get workshopped to death (it's easy to tell when a game with potential got derailed in this way). The author isn't saying good game stories don't exist, only that they're rare and most often from smaller studios.
I don't think it's a stupid question, just oddly phrased (more on that later). I think there are two stable interpretations. "When did grass evolve" or "when were grass lawns invented".
Later:
Why isn't it an obviously stupid question? Because I think the accompanying "who (or what) invented grass" is validly answered as "by evolution". I feel the act of invention requires no intentionality and is simply the output of learning processes where generated artifacts have material and dynamic properties embodying deep knowledge of physical laws and help achieve some goal relative to an environment. Evolution learns in the sense that the mathematics of natural selection mirror that of bayesian filters.
Certain dolphins have larger prefrontal cortices with more neurons, are highly intelligent with the ability to predict and long term memory yet seem to lack what we think is a unique human capacity for first person introspective meta-cognition. Which is essentially what most and this article mean when talking about the hard problem or qualia.
We can also look at chimpanzees. They have a prefrontal cortex, future prediction and long term memory. While their prefrontal cortex is smaller, it is also the most human like. Yet, they too seemingly lack capacity for that human style of conscious being. There is alas, something more your definition has failed to touch upon.
That's not accurate. By the 1960s and 70s, the scientific consensus [1] not only strongly believed smoking was a causal factor in lung cancer but also knew it contained radioactive isotopes of Polonium (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2509609/) and lead.
> No exploding coal plant will render a whole region unlivable for >30 years
Unlivable for humans. The exclusion zone around Chernobyl became a de facto wild-life sanctuary. To the extent that some environmentalists now argue benefits of officially maintaining it as one indefinitely. Human supporting activities such as farming, mining and habitat expansion often come at a serious cost to local ecosystems. Nature ends up flourishing over time in humans excluded regions exactly because they are unlivable for > 30 years.
As pointed out by others, one effect of burning coal is the concentration of its radioactive elements in fly ash. And since less care is taken, more radioactive material ends up released into the environment by coal plants than nuclear plants. It's worth pointing out however, when contrasted with background, risks from exposure to radioactivity is not significantly raised by living near coal plants. The real killer from coal is pollution. Having said that, accumulation of fly ash over time could be a concern, especially wherever it gathered under non-uniform dispersal. As far as I know, this is yet to be shown.
Perhaps a stronger example of society's inconsistent reasoning about radioactivity exposure is inhalation of tobacco smoke. It's curious that information on the significant amounts of radioactive material in tobacco smoke did not percolate widely and probably would not have changed habits anyway, given its known carcinogenic nature was already not enough to do so.
The code on the right is an explanation or implementation you could jump into. The replacement for use would be the function name, such as `summation` or `product`. A computer language is already used to communicate almost all math these days: LaTeX. A slightly lighter syntax, perhaps with optional unicode names, built on a symbolic math package would not be any more cumbersome.
From the article we learn Eurisko is more of an example of human machine symbiosis than of AI.
"Thus the final crediting of the win should be about 60/40% Lenat/Eurisko," he wrote, "though the significant point here is that neither party could have won alone. The program came up with all the innovative designs and design rules and recognized the significance of most of these. It was a human observer, however, (the author) who appreciated the rest, and who occasionally noticed errors or flaws in the synthesized design rules which would have wasted inordinate amounts of time before being corrected by Eurisko."
Lenat did not know much about the game and its search space was intractable for the AI. The combination of both was able to come up with unique solutions counter-intuitive to humans.
From a certain perspective, Eurisko could also be viewed as on the path to some next iteration of programming that never quite arrived. That is, programming languages stalled in the 70s, as Alan Kay notes in: http://worrydream.com/EarlyHistoryOfSmalltalk/
A look beyond OOP as we know it today can also be done by thinking about late-binding. Prolog's great idea is that it doesn't need binding to values in order to carry out computations [Col *]. The variable is an object and a web of partial results can be built to be filled in when a binding is finally found. Eurisko [Lenat *] constructs its methods—and modifies its basic strategies—as it tries to solve a problem. Instead of a problem looking for methods, the methods look for problems—and Eurisko looks for the methods of the methods. This has been called "opportunistic programming"—I think of it as a drive for more enlightenment, in which problems get resolved as part of the process.
This higher computational finesse will be needed as the next paradigm shift—that of pervasive networking—takes place over the next five years. Objects will gradually become active agents and will travel the networks in search of useful information and tools for their managers. Objects brought back into a computational environment from halfway around the world will not be able to configure themselves by direct protocol matching as do objects today. Instead, the objects will carry much more information about themselves in a form that permits inferential docking. Some of the ongoing work in specification can be turned to this task [Guttag ][Goguen].
Tongue in cheek, I once characterized progress in programming languages as kind of "sunspot" theory, in which major advances took place about every 11 years. We started with machine code in 1950, then in 1956 FORTRAN came along as a "better old thing" which if looked at as "almost a new thing" became the precursor of ALGOL-60 in 1961. In 1966, SIMULA was the "better old thing," which if looked at as "almost a new thing" became the precursor of Smalltalk in 1972.
Everything seemed set up to confirm the "theory" once more: in 1978 Eurisko was in place as the "better old thing" that was "almost a new thing". But 1983—and the whole decade—came and went without the "new thing". Of course, such a theory is silly anyway—and yet, I think the enormous commercialization of personal computing has smothered much of the kind of work that used to go on in universities and research labs, by sucking the talented kids towards practical applications.
> Unlike the classifier system, which acts on raw binary strings, PI rules and representations are Lisp expressions, and there are more structured representations of concepts, which are however still basically rule clusters. Again unlike the classifier system, which gets new rules by using a genetic algorithm on existing ones, PI has a number of specific inductive mechanisms --- specialization, generalization, abduction, and concept formation --- which are intended to be reminiscent of conscious experience, unlike the GA.
PI was invented by Holland, who is more famously known for his contributions to genetic algorithms and evolutionary computation. Those algorithms were meant to be used in a system like PI and not in the stand alone manner typical of today. Learning Classifier Systems being simplified versions of PI, are probably the closest concept to Eurisko in semi-active use. DreamCoder: https://arxiv.org/abs/2006.08381 is the latest development along this investigatory and very much underexplored path to Machine intelligence.
I have a fair amount of experience in both F# and Ocaml and don't think the lack of modules harms F#. The trick I had to learn was to think of F# as its own language and not try to write it as Ocaml or Haskell. What I do sometimes miss in F# are existential types.
I also wouldn't say objects in F# are less natural. This is only opinion but, I prefer F#'s lighter object syntax and its object expressions are great. Active patterns also provide an expressive option for dispatching on abstract datatypes. The conflicts vs .NET I've found with F# OOP are when implementing comparison operators, or when interacting with frameworks that want public fields. Other than that, there's not much friction. I suppose there is friction when it comes to MS tooling for producing Windows UIs but then, F# also has Fable and Elm inspired UI kits for various platforms. Going AOT route will also meet with some friction.
Comparing languages comes down to what problem solving approaches they make difficult by default. In that light, F# is much more properly a "functional" language than C#. The language is designed from the ground up for recursive algebraic data structures, HM type inference, flexible pattern destructuring, null avoidance, exhaustive matching, computation expressions, idioms around composable higher order functions, clean syntax for function types, immutability and sound type guided development.
It also has a light python like syntax and like all MLs but perhaps even more so, is pragmatic when it comes to the use of esoteric design concepts. The philosophy is not to be functional and pure for their own sake but about composing language defaults which force certain types of correctness promoting approaches.
F# also has units of measure and type providers. They're probably less often used. But just the json, html and xml providers are useful on their own.
To be clear, this isn't arguing F# vs C#. Such arguments are extremely boring. It's emphasizing what core design decisions distinguish the two languages.
Take entomologists, bird watchers or those who identify mushrooms. In each, there are instances that are fiendishly difficult to tell apart. If you ask an expert for advice, they'll tell what features to look for and where, sometimes not even on the item itself and some requiring specialists tools.
While explanations are far from sufficient to instantly close the gap to expertise, they provide a massive boost that you might not otherwise have found on your own. The justification comes from the fact that their explanations are a reliable source of increased performance in making fine-grained distinctions. It's further demonstrated by answers to questions like "If they are so difficult to tell apart, why make these distinctions?" or "How did they come to be so similar?".
The trick is that even a tiny difference in ability can lead to a large difference in results as the number of decision steps involved in a task increase. The expert wastes less time in search space. This is a combination of good mental models honed from experience and attention to detail.
Something unusual about Bellard is the breadth of his knowledge. There are probably a fair few who can write a fast deterministic low level code optimized neural network library. A decent number who can write an autodiff library but few who can do both. Once that's done, I daresay that's the hard part.
Given such a library, implementing an LSTM or transformer model and combining it with Adaptive Modeling/incremental learning of data statistics using a neural network to feed an Arithmetic Coder is something that is within the reach of many more.
In summary, a 10x developer is not 10x smarter or 10x more skilled, they just waste less time investigating unproductive paths and are maybe 1% less likely to take wrong turns. This ends up making a huge difference for large complex tasks.
https://tvtropes.org/pmwiki/pmwiki.php/Main/MacGyvering