Welcome (Pinned Post)

Posted in Quick Read · Sept. 1, 2018, 9:23 p.m. · 145 words

Hi. This is my blog. Ideally, below this entry you would find posts on a variety of different subjects, like stuff I've read or watched, places I've been or other items of topical interest. You know, the basics. But as I write this (September 1st 2018) the vast majority of blog entries are devoted to reviews of comic books featuring the Incredible Hulk.

Maybe some explanation is required. It's pretty simple: in 2012 I decided to read and review all the Incredible Hulk comics from the beginning. Six years later, I have got through 35 of them (Incredible Hulk vol. 1 and Tales to Astonish up to issue #88). There are still a few more to do, but I hope to get around to them. While I'm working on that, there will be some posts on other topics in this space.

To Be Hulkinued, as they say.

How Netflix Killed the Author

Posted in Deep Dive · Nov. 5, 2018, 10:26 a.m. · 1698 words

Apparently TV scripts are being rewritten by algorithms now, or at least that's the impression you got if you paid any attention to the publicity around Cary Fukunaga's Maniac, released on Netflix in September. Fukunaga gave an interview to GQ, where he said, in part:

Because Netflix is a data company, they know exactly how their viewers watch things... So they can look at something you're writing and say, We know based on our data that if you do this, we will lose this many viewers. So it's a different kind of note-giving. It's not like, Let's discuss this and maybe I'm gonna win. The algorithm's argument is gonna win at the end of the day.

At least in the parts of the internet that turn up in my feeds, this was not well received. Ron Gilmer at Collider wrote about "a computer program dictating creative changes" and "[surrendering] our entertainment to the equivalent of Skynet" while an article at the Quartzy pondered whether the show would look like it had been "written by a computer."

Meanwhile my Twitter feed was full of dire predictions of the assimilation of the artistic impulse into the algorithmic Borg, although admittedly a lot of this came from science fiction writers whose professional viability depends in part on being able to come up with worst case scenarios for any new technological development.

But if we want to get past the hype, it seems reasonable to point out that TV ratings have existed in America since the Truman administration. What is Netflix doing that's actually new?

We can start by pointing out what they're not doing. We've seen such advances in facial and voice recognition technology that it might seem plausible their Skynet-for-script-development is being fed video footage and spitting out analysis and statistics. But that's not happening. As machine-learning specialist Professor Michael Jordan points out, Google's search algorithm still can't parse a relatively simple sentence like "What's the second-biggest city in New England that is not on the coast?" We're nowhere near having software that can be given a tv episode and make useful determinations about the plot or characters.

So we need human beings to watch the show and give some sort of simplified analysis of its content. The traditional Hollywood method of simplified analysis is what's called 'coverage,' where agencies and production companies hire script readers to reduce unproduced screenplays to summaries of their plot, characters, quality and potential commercial prospects. A 2006 New Yorker article by Malcolm Gladwell contains an account of Epagogix, a company that claims to be able to predict the eventual box office takings of a movie based on this sort of simplified plot synopsis.

It's worth thinking about that for a moment. First, you can ask anyone who works in movie production whether you can reliably tell what the finished product will look and feel like based on the original screenplay. In my experience the answer you will get to this is a unanimous no. Putting aside sequels that are following an existing template, there are so many points during pre-production and filming where the character of the work can change completely, based on choices made by the director, the actors, the cinematographer, editor, composer, costume designer and so on.

Then, once the film is completed, there are still so many confounding factors affecting its profitability: changes in culture and fashion, news stories that affect the way audiences feel about the star(s) and subject matter, or internal studio politics that affect whether a movie is promoted or dumped. For example, the creative disagreement that allegedly led Paramount to abandon the international theatrical release of Alex Garland's Annihilation and release it directly to streaming instead, on Netflix.

Given all the things that can distort the relationship between a screenplay and its eventual box office take, I think it's worth being extremely sceptical about anyone who claims to have discovered a mathematical relationship between the two. In fact, I'll go a step further. It's worth being extremely sceptical when you see any news coverage of developments in AI, period. The field is so overhyped at the moment: if you take a look at job postings for tech firms, there are a huge number where the business strategy appears to be 1) Hire data scientists; 2) ?????; 3) Profit. Every company with unanalysed data seems to want to wave a machine learning magic wand at it, and any company with a proprietary algorithm has a vested interest in making it seem as magical as they can.

Netflix, then. What is it actually doing?

We have some clues, thanks to the work of Alexis Madrigal at the Atlantic, who used web scraping software to get the details of the oddly-specific genre categories that turn up in your Netflix recommendations. His analysis revealed that Netflix "microgenres" are formed out of combinations of categories including things like "based on books," "set in the 1970s," and an odd, unexplained enthusiasm for the works of Raymond Burr.

Madrigal ended up getting in touch with Todd Yellin, a Netflix VP, and this is where we learn more about the details of the company's process. Netflix has replaced traditional coverage with a process for gathering metadata on everything in their library. Instead of hiring aspiring screenwriters to write plot synopses, Netflix hires them to fill out an elaborate spreadsheet Yellin designed, with a 36-page user manual (or at least it was originally an elaborate spreadsheet; maybe they've since replaced it with an elaborate app).

I haven't seen the spreadsheet—unsurprisingly it doesn't seem to be publically available—but it is apparently used to record details like story locations and the lead characters' jobs, plot elements like spaceships or zombies or the presence of a strong female lead, and rate various aspects from 1 to 5, like how happy the ending is, how much gore is seen, and the social acceptability of each of the characters. Netflix then combines all of this metadata with its viewership data (basically a more granular version of traditional ratings) to generate recommendations for users and to tell its creators what they can and can't do.

I don't know about you, but for me this is a pretty far cry from Skynet rewriting people's screenplays. Studio executives have been demanding rewrites from screenwriters since Hollywood began, based on some unarticulated admixture of gut hunches and previous box-office performance. Using an elaborate spreadsheet seems like a more scientific approach, but consider what a lossy compression format that spreadsheet is. Every second of film or tv is filled with information, signals that shape the viewing experience, from lighting and sound design and casting up to word choice in the dialogue. Can you really capture all of that in a spreadsheet completed by an independent contractor who's watching 20 hours of content per week? Imagine giving someone the spreadsheet to look at and then getting them to watch the movie. Would it be identical to the movie they had in their head based on the metadata alone?

(That's putting aside the subjective aspects of categorising and rating content, like anything to do with humour. I know people with otherwise good taste who can't stand Monty Python; I would rather chew on tinfoil than watch an episode of Arrested Development.)

Plus, you have to take into account that viewer response is based on combinations of factors. In different genres you're willing to accept different things. I kept watching The Shield after the main characters murdered a police officer in cold blood, but it would probably have derailed my viewing of Grace and Frankie.

The combinations can be more subtle and counterintuitive than that—screenwriter William Goldman once blamed the commercial disappointment of his The Great Waldo Pepper on audiences liking Robert Redford too much in the lead, meaning that they turned against the movie when Redford's character failed to prevent Susan Sarandon's character from falling to her death. The same story would have succeeded if the film had starred an actor with more of a dark side to his persona, Goldman argued. "I truly believe that if Jack Nicholson had been in the part, he wouldn't have been as good as Redford, but the movie would have worked for audiences," he wrote.

So you can start to see how murky this stuff gets, and how many variables there are. Although if you were really serious about taking a scientific approach you could always try creating alternative edits of tv episodes, changing one aspect at a time, assigning them randomly to viewers and doing A/B testing, but while Netflix has apparently been doing this sort of thing with its promotional materials, I doubt they're going to commit the sort of resources required to experiment with the content itself. (Although maybe you could see that as a missed opportunity to try out different casting choices for Iron Fist.)

None of this is meant to argue against the idea of collecting data or doing analysis. As someone who regularly downloads interesting-looking datasets to play with in R, I'd love to get my hands on Netflix's data. But as someone who also gets paid money to write fiction I'd be extremely dubious about basing any creative decision entirely on that data, even if what you're trying to maximise is eyeballs rather than quality.

If you've made it this far I should perhaps sound a note of caution, which is that because I haven't had a chance to examine Netflix's metadata collection process for myself I don't actually know how it deals with series pacing, which was the issue with Cary Fukunaga's original version of Maniac. But based on the above, I would hope that future Netflix creators are able to treat feedback on their work as a slightly better-informed version of standard studio notes, rather than as Netflix wielding some sort of unimpeachable evidence-based argument.

And I hope the Netflix executives giving feedback are able to avoid believing their own press. When you have qualitative research in spreadsheet form it's easy to start thinking that you're dealing in cold, hard fact and not the mushy intangibles of human experience and perception. But not only has Netflix not yet solved the problem of art, claims that they have solved the problem of how to keep people watching are, in the meantime, still pretty dubious.

Tales to Astonish #88

Posted in To Be HULKinued · March 8, 2017, 2:48 p.m. · 77 words

Gil Kane blows into town, and his dynamic, cartoony style helps clear out the cobwebs after John Buscema's too-stiff, classic illustration approach. Kane is the first artist on the title to ink his own pencils since Ditko, and it's not entirely clear if he's clicking here with Stan Lee, who actually interrupts his usual compulsive logorrhea to leave one panel entirely wordless. The story is not much better than OKAY, but the new visual direction shows promise.

Tales to Astonish #87

Posted in To Be HULKinued · Jan. 28, 2017, 7:39 p.m. · 128 words

"The Humanoid and the Hero!" — After having something of a change of heart on the destroying-the-Hulk issue, the Army tries to help take down the Humanoid instead, in a lengthy fight sequence. But more importantly, this month's subplot features the return of Boomerang, with a repeat appearance that sees him immediately binning his old costume, having decided the smock covered with suction cups was not a good look. Sadly his definition of improvement involves replacing the baggy cargo pants with a standard tights and trunks combo but keeping the spangly 'B' chest emblem, this time accented by decorative bandoliers and a cummerbund with a glued-on boomerang pointed at the crotch. The end result is an outfit that's a slightly different flavour of terrible, but an OKAY comic.

Tales to Astonish #86

Posted in To Be HULKinued · Jan. 24, 2017, 4:11 p.m. · 119 words

"The Birth of... the Hulk-Killer!" — It's time for pulse-pounding plot contrivance in the mighty Marvel manner as Bruce Banner, who hasn't been seen since issue #70, reappears for the three panels necessary to reprogram the Orion missile and save the day before Hulking out again. Other returning characters this issue include Boomerang, who shows up for a training montage in order to demonstrate that he has new powers and skills (but the same old costume). And while this is going on, the Army stumbles on the Leader's hideout and activates his long-dormant secret weapon in hopes that it will take out the Hulk. What could happen next? If your answer was:

take a drink. Readable, but deeply OKAY.

Nicholas Fisk

Posted in Quick Read · Jan. 24, 2017, 11:23 a.m. · 255 words

There's a strange repeating process I've noticed on social media over the past several years, which is that every few months British people of my generation will be talking about their favourite childhood books and suddenly everyone will remember that Nicholas Fisk existed.

People may have forgotten Fisk's name but they remember his stories. 25 years ago his novels were in the children's section of seemingly every library in the UK. Librarians must have loved him. He wrote sf and fantasy and horror for what would probably now be called middle-grade readers; his books were often unsettling and they left an indelible mark. There was Grinny, a sort of sinister flip-side to Mary Poppins. Monster Maker, a fantasy-horror story set in a special-effects studio, which was adapted for television by Jim Henson.

And Time Trap, an odd, troubling book about gang violence and an antiseptic future and time travel and the Blitz: a book I didn't read so much as worry at, the way a dog worries at a bone. I would borrow it from the library every few weeks and try to decide if I liked it. (Last year I happened to notice that my local library had a copy, so I borrowed it again. I'm still trying to make up my mind about it.)

Fisk died last year at the age of 92, after a long retirement. I only ever knew him through his books, and I regret never getting the chance to meet him. I hope he knew how much he mattered.

Tales to Astonish #85

Posted in To Be HULKinued · Jan. 17, 2017, 11:15 p.m. · 108 words

"The Missile and the Monster!" — Wherever he goes, whatever he does, the Hulk can't seem to escape Soviets trying to foul up US weapons tests. This issue's saboteur turns out to be last issue's sinister Hawaiian shirt man, who has hidden a frowning robot in the trunk of Rick Jones's car in hopes of disrupting the takeoff of the Orion missile. Well, there's no such thing as a free launch, after all. John Buscema gives the Hulk an odd slab-like torso here, but his Hulk's head — floppy-haired, no lips, unibrow — is closer to what would end up on kids' lunch boxes than any one else's thus far. OKAY.

Tales to Astonish #84

Posted in To Be HULKinued · Jan. 14, 2017, 1:06 p.m. · 201 words

Reading Jack Kirby comics is often more about the journey than the destination and this sadly marks the dead-end conclusion of his run on the Hulk. Kirby had been filling in on the Sub-Mariner co-feature in Astonish, setting up a storyline that would unite the book in a thrilling crossover battle. But then he stopped drawing Namor and abruptly gave up doing layouts for the Hulk, with the result that in this issue the two pass each other in a movie theater but never actually meet. Instead, the Hulk watches some news footage that conveniently wraps up the Boomerang plot of the last several issues, then has a fight with a train. I love ya Jack, but this is an AWFUL way to go out.

Also noted:

  • If you're wondering how to spot a Marvel comic that was slapped together in a last-minute deadline crunch, an art credit to "Almost the whole blamed Bullpen" is a pretty good place to start.
  • Meanwhile in subplot land, Rick Jones finds a way to drive out of Florida — by borrowing a car from a sinister cigar-smoking man in a Hawaiian shirt, who warns him not to look in the trunk. What could possibly go wrong?

Tales to Astonish #83

Posted in To Be HULKinued · Jan. 2, 2017, 6:22 p.m. · 186 words

"Less Than Monster, More Than Man" — Let's say it's 1966 and you're in the mood for a story about a secretive high-tech espionage organization whose members are referred to only by a numeral ("Number One", "Number Two", "Number Six" and so on) as they carry out strange actions in pursuit of obscure but presumably sinister goals. Unfortunately, The Prisoner won't debut on TV for another year so you'll have to settle for reading Tales to Astonish, where the Secret Empire also does all of the above. Sadly, the Secret Empire must be the most self-sabotagingly vindictive shadowy enterprise of all time: after failing to kidnap Betty Ross last issue, Boomerang embarks on a perfectly credible Plan B, but with no support from Numbers Two to Nine, who are busy meeting to discuss killing him off until Nine throws a stun grenade at the rest. It's the sort of group where the failure to wear a metal bucket under your villain robes can be described as "blind confidence". Meanwhile, the Hulk goes out looking for food and comes back carrying General Ross and Rick Jones. OKAY.

Tales to Astonish #82

Posted in To Be HULKinued · Dec. 27, 2016, 5:09 p.m. · 101 words

"The Battle Cry of the Boomerang" — It's plot-counterplot this issue, as the Hulk returns to the surface, where he is attacked by the US Army, which is looking for Boomerang, who has kidnapped General Ross's daughter on the orders of The Secret Empire, whose members are busily murdering one another whilst watching proceedings on a view screen, classic supervillain style. Jack Kirby is listed as "designer" in the credits, meaning that although he didn't have time to actually pencil the comic, Stan wanted his story ideas. He threw the kitchen sink at this one (but Boomerang still looks rubbish). GOOD.

Hulk Smash #1: The Business

Posted in To Be HULKinued · Dec. 7, 2016, 1:18 p.m. · 126 words

Briefly skipping ahead, here's a page I love from The Incredible Hulk #383, written by Peter David and pencilled by Dale Keown.

Hulk #383 was published in 1991 and I can still remember where I bought my copy and where I read it. Eleven year old me didn't get the Abomination's joke about Ted Turner (although it made 32 year old me laugh out loud just now) but he loved the last panel, where the Hulk says a cool thing and hits someone.

Mainstream comics of the '90s largely aspired to be about characters saying a cool thing and hitting someone, and having set so low a bar most of them have unsurprisingly not stood the test of time. But I still think this is great.

Tales to Astonish #81

Posted in To Be HULKinued · Dec. 6, 2016, 1:18 p.m. · 110 words

"The Stage is Set!" — While Bruce Banner wanders shirtless through underground tunnels, Gamma Base comes under attack from possibly the worst wardrobe Kirby ever gave a supervillain. This is the murderous menace of the man called Boomerang, who arrives swathed in red cargo pants, a pointy metal helmet with bolt-on goggles, antenna and a chin-strap, and a baggy canvas shirt adorned with suction cups and a spangly letter B strapped across the chest. Readers of the whacked-out cosmic Sixties Marvel comics often took the strait-laced Kirby for a habitual drug user; here he has apparently mainlined a heroic dose of some narcotic that totally destroys your fashion sense. EH.

Tales to Astonish #80

Posted in To Be HULKinued · Dec. 5, 2016, 1:18 p.m. · 84 words

"They Dwell in the Depths!" — The Hulk gets caught up in a subterranean turf war between the evil forces of Tyrannus and visiting Fantastic Four c-lister The Mole Man. The undoubted highlight of this issue is the Mole Man's robotic Octo-Sapien, a sort of metal bulb on legs surrounded by droopy nozzles and topped off by a villainous frowny face, the combined power of which somehow manages to turn the Hulk back into Bruce Banner. This one scales the heady heights of OKAY.

Tales to Astonish #79

Posted in To Be HULKinued · Dec. 4, 2016, 1:18 p.m. · 85 words

"The Titan and the Torment!" — Readers who were unengaged by last issue's utterly generic mad scientist plot (and I assume that's basically all of them) will be pleased this time around, as the villain in question accidentally kills himself after a four-panel tussle with the Hulk. There then follows a fight with Hercules, briefly on loan from the Thor comic. The art is inconsistent with occasional sojourns into actually dire, and the story feels like a deadline-week afterthought. It's the Contractually Obligated Hulk. AWFUL.

Tales to Astonish #78

Posted in To Be HULKinued · Dec. 3, 2016, 1:18 p.m. · 98 words

"The Hulk Must Die!" — Bill Everett's pencil and ink art for this issue gives it a look that's somewhat unique for the Marvel era, both strangely angular and surprisingly detailed. The relentless forward motion of Kirby's story layouts prevents the art from getting totally bogged down in Fiftiesish Joe Maneely-style noodling, although it has to be said that the Hulk spends most of this instalment failing to climb out of a hole he's fallen into. Meanwhile, an evil scientist in the employ of the US Army plots to overthrow humanity with an array of Hulk-powered machinery. EH.