And I'm pretty sure I'm the only former SF editor who has been working in marketing for the past two decades and currently creates content about AI for a major tech company that is one of the leaders in its space. (OK, I work in legaltech rather than adtech, but I use adtech as a marketer every day, and I have been involved in multiple major product launches, including an upcoming one I can't talk about.) I will try not to nitpick too much, but I did have a lot of internal questions about this company's launch plan, among other things, while reading the book.
Actually, let me ground it all first: Monitored tries to do something impossible. Feinstein started writing it in late '22, and it's a near-future novel about a technology that is changing almost daily, and has been transforming entire industries over the past four years. I enjoyed it, and thought Feinstein did an awesome job, while still deeply believing he was writing about a 2022 tech world and tech capabilities, when those already look very different just a short time later.
If you work in AI at all, read this book quickly. Near future dates the fastest of all SF, and this is an area that is dating like fresh-baked bread. No offense to Feinstein about any of that: it's just the nature of the field right now.
That said: Monitored is an epistolary novel, in a zippy style. It's told through chat logs, surveillance videos, meeting transcripts, email exchanges, and similar media. It reads quickly, and focuses on the people (and bots) and what they say to each other.
UniView is a newish company, apparently already public, headquartered in Manhattan. They create sophisticated AI bots - their Sam is the basis of what's said to be all self-driving cars, though he was at least partially open-sourced, so UniView is not making as much money as it might.
(Actually, maybe I will nitpick. UniView has three active models at the beginning of the book - three products. Unlike the real-world frontier-model makers, each UniView model is discrete; they don't do new versions of old models, a la ChatGPT 5.5-Pro.
Those three models are:
- Sam is presumably the one bringing in revenue, either partially through licensing to other companies or entirely through running the UniView proprietary version.
- Lex runs the internal UniView HR function, and there's no indication it has been licensed - though it could be.
- The third model, Casey, has no specific purpose Feinstein mentions, and is deleted very early in the book, so it clearly is not bringing in revenue.
Frankly, they act like a pre-IPO company most of the time, which could either be Feinstein doing it deliberately or something he didn't think of. And we don't see any of the UniView functions that would actually bring in revenue - marketing, sales, partnerships - which is either a bug or a feature, depending on how you look at it.)
UniView has a fourth bot in development: Quinn. Quinn will be an adtech engine, designed to create broadcast-quality video ads in real time, individualized for particular viewers. (I did wonder how Quinn would do this all that quickly from the server side, without any lag, and how it would get that data on all of the viewers, but those are minor quibbles.)
So, my marketing brain thinks: OK, so UniView will need to have an alpha customer or two - probably some big ad agency, but maybe a media outlet - to vet Quinn's output, provide glowing quotes, and work with the Quinn dev team under an NDA. Nope, not at all. Nothing like that. In fact, Feinstein has UniView fire its entire salesforce midway through the novel, right before the launch, and seems to be expecting customers to come to them after launch. This is a vaguely plausible motion in a consumer space - though you still need to have distribution at scale ready - but makes no sense for B2B. The customers for a ad-creation engine are ad agencies (at least in the short term, until they're actively routed around and marginalized) and large, mostly consumer-facing companies. Those companies are not going to come up to your door.
There's also no mention of metrics to feed back to customers, and I really wanted to know the pricing model for this thing, but now I'm getting on my hobbyhorses, so I should get back to the plot.
(Well, except for a quick second nitpick. UniView's AIs are presumably agents, but Feinstein doesn't use the word "agentic" at any time in this novel. I think this is on purpose; he also doesn't use "GenAI." Since AI is moving so quickly, Feinstein did what he could to future-proof the novel, and uses mostly base-level, long-term terminology. Readers may think of them as LLMs, but be aware that Feinstein is not being that specific - so the ways his characters interact with his bots are not "wrong" because they don't match current practice.)
UniView needs to train Quinn how to target individuals - teach it the basics of advertising. Training bots at UniView mostly happens through text-based chat - occasionally voice, but not much - which is slightly quaint. They decide to hire an advertising expert to do this, and grab Noah Ross, a fortyish veteran looking for work after a vaguely-mentioned (and somewhat clichéd) personal and professional breakdown.
Noah goes to work on the Quinn team, under project leader Simon Chung and working closely with ethicist Haley Ellis. (Feinstein sets up for a potential romance between Noah and Haley, but the format of the novel - also some plot elements - leaves that mostly as potential rather than actuality.) He chats with Quinn about how advertising works, how to target individuals, what people care about - all of the emotional grounding of the ad biz. Quinn generates ads for particular profiles, which are gauged on a standard scale; in the best Silicon Valley fashion, Noah's sole purpose is to make Number Go Up, as quickly as possible for the upcoming launch.
All that is pretty straightforward, even if the bot training is a bit more touchy-feely and less tinkering-with-weights than I expected. As is usual with SF novels about AI, the bots are supposedly non-sentient...but they act like people and talk like people and the other characters interact with them like people, which leads to the Big Ending.
But, meanwhile, the CEO and founder of UniView, Ian Lindell, is coming up on a milestone birthday. He had promised himself that he would do a whole bunch of things by that birthday, and the one missing thing on that list is "become a billionaire." So this launch has to make him a fortune.
Again, that feels very pre-IPO, and is a level of ambition and arrogance that would be unremarkable for any founder/CEO. Since UniView is already public, and Ian doesn't seem to have Musk-level abilities to cloud the minds of his board [1] to vote him massive unearned bonuses, he has to go in a somewhat different direction than the usual "launch product, disrupt everything, get all the money" playbook.
Ian's choice is frankly that of a cartoon villain; it happens very early in the novel and I found it weird, unlikely, and bizarre. It's not that I don't think CEOs aren't sociopaths who would take any action that benefited themselves; it's just that it's so criminal, and would be massively obvious in retrospect, that I don't believe he would be stupid enough to believe he could get away with it as depicted.
But he does the stupid illegal thing, and he covers over it, and he is mildly paranoid that a competitor will beat them to market, so he both makes things more difficult for the Quinn team (mostly inadvertently) and speeds up the launch radically.
Are these bots people? Do they have free will? Will they make choices?
This is a SF novel, not the real world, so the answers should be obvious. Also, though nobody actually says so during the novel, Feinstein clearly expects we will all think of Quinn's outputs (ultra-targeted video ads that know everything about you, manipulate your desires, and are really great at getting you to buy things you don't need) as horrible.
It does end well, and the speed imparted by Feinstein's epistolary style is a real strength - as I said up top, this novel is zippy and modern-feeling, which goes a long way to making it work well. I do have a lot of nitpicks - perhaps my background does not actually make me the perfect reviewer for this book, but instead a guy who knows too much for his own good - but they're mostly on the worldbuilding end, and mostly (I expect) artifacts of what changed in the world between when Feinstein started writing this book and I started reading it.
I do think this will date quickly; it's already dating quickly. But that just means it's of a particular time, and that can be a great advantage in SF.
[1] He presumably has a board, but they don't figure in the novel at all. UniView is stated to be a public company, but there's nothing about governance or financial reporting in the novel. As far as we can tell, by the end of the book UniView consists of a CEO, an automated HR bot, and a few product teams - no one else.
No comments:
Post a Comment