A weekly letter from the founding editor of The Browser. Correspondence and criticism gratefully received and always read: firstname.lastname@example.org
If you are not a paying subscriber to The Browser, and enjoy this letter, please do become a paying subscriber to The Browser, because that is how I earn the money to write this letter.
This week: AI overload, text-to-voice, notable newsletters, and books I have been reading.
WE HAVE been recommending relatively few articles about artificial intelligence and GPT4 on The Browser of late — relative, at any rate, to the number being published in newspapers and newsletters in the months since ChatGPT went public. For myself, I find people's accounts of their conversations with AI somewhat similar to people's accounts of their dreams: Both subjects can seem fascinating, even revelatory, to the person involved, but neither travels well into writing or conversation.
I do feel that something momentous is happening when machines master language, but I do not know what that something will amount to over time, and nor, I think, does anyone else. At most I am confident that we will soon understand a lot more about how language works, thanks to AI, an advance which is excitement enough for me; and that we will come to a better understanding of what we mean by "intelligence".
I recommended Tyler Cowen's conversation with "Jonathan Swift" on The Browser last week because I saw it as useful evidence of just how far AI has already moved beyond the Turing Test. Not only can GPT4 pass for a human being, it can assume the history and to some very limited extent the personality of a particular person. Likewise, I recommended Stephen Wolfram's recent paper on AIs and neural networks because I thought it the most detailed and concrete explanation likely to be accessible to a general reader such as myself.
As to the social and political implications of AI, I recommend The Moral Economy Of High-Tech Modernism, by Henry Farrell and Marion Fourcade. The paper is primarily concerned with the replacement of bureaucratic choice by algorithmic choice, but the interests and factors at work in that shift are similar to those at work in the rise of AI.
Most of all I recommend exploring AI at first-hand. It is readily available in the form of ChatGPT. Play with it, ask it questions, give it commands, draw it into conversation. A minute with ChatGPT is worth a thousand words of commentary.
An email from Berit Andreone reminds me of something that used to be very much on my mind until I was distracted by AI. Are we anywhere near the point at which Browser subscribers can listen with pleasure to recommended stories as well as reading them?
We did suggest a while back that Browser subscribers who wanted audio might try linking their accounts to Instapaper Premium, which includes a text-to-speech feature. Saved articles are read out loud by a computer-generated voice. But while computer speech is good enough when one merely wants information, as with GPS, there is little pleasure to be had, at least for me, in listening to a computer's reading of long and nuanced essays. The Instapaper feature, while ingenious, is not one that I have been using.
I trust that computer-generated voices will soon improve to the point at which they are acceptable for longer listening, and that in a year or two subscribers may be able to pair The Browser easily with some text-to-voice platform that does near-human speech. Uri points me towards AD Auris, a text-to-speech service which allows one to make Spotify playlists of articles read aloud by computerised voices, and which may well be the best option for the time being.
I have added two newsletters to my feed-reader this week. They are hosted on Substack, they do similar jobs for contrasting publications, and they suggest an interesting trend towards summarisation.
As its name suggests, Last Week's New Yorker, by Sam Circle, summarises and evaluates a selection of articles from the latest print issue of the New Yorker. I find it well-written, and I generally agree with Sam's judgements.
We very rarely recommend New Yorker articles on The Browser, despite their general excellence, only because we think it highly likely that Browser subscribers will have seen them anyway.
So, on the face of it, if you want to be alerted to the best things in the New Yorker, without subscribing to the New Yorker, then Last Week's New Yorker might be the answer. The counter-argument is that, whereas an annual subscription to Last Week's New Yorker costs $40, a digital subscription to the New Yorker itself is today being advertised at just $49.99, at least for the first year — a limited-offer discount which seems to me an extraordinary bargain.
Tracking People's Daily, by Manoj Kewalramani, provides translations, summaries and analysis of articles appearing in People’s Daily, the main newspaper of the Chinese Communist Party. The suggested yearly subscription is $40 (in rupees), or you can read for free if you so choose.
The hitch here is that People's Daily is not generally interesting in any conventional sense. You will find in it exactly what you would expect to find in the flagship newspaper of a totalitarian state: Leaders' speeches; meetings between Chinese and visiting officials; protestations of China's benevolent and selfless intentions towards the world at large; plus lots and lots of statistics about exports, bank loans, and industrial production.
The trick is to read People's Daily regularly and carefully until you recognise all the boilerplate, at which point you can start spotting the anomalies, especially the innovations and omissions, which is where the real news is hiding. Has a new adjective been applied to Xi Jinping Thought? Is somebody's name missing from a list of attendees? Is some event of significance being unaccountably ignored?
If you need to know about China in detail, Tracking People's Daily is fascinating stuff. If not, I still recommend a salutary glance at it every now and then, if only to remind yourself how lucky we are to have the New York Times.
Books I have been reading
Nineteen Eighty-Four, by George Orwell (1949).
I am a timid, tree-hugging liberal. My preferred way of life is a house and garden connected by public transport to a walkable town centre where I can do as I please. I doubt I harbour any unconscious desires to live under an totalitarian dictatorship. But as I re-read Nineteen Eighty-Four, almost seventy-five years after it was written, and many decades since I read it last, I cannot warm to Winston Smith, nor can I quite believe in the world that Orwell imagines for him.
Yes, the life described there seems horrible to me — but why are Winston and Julia the only people who seem to hate it? Why is Smith the only person whose memory seems to work? How can an all-seeing totalitarian state have placed Smith's home-surveillance camera with a blind spot just where Smith has his desk and his diary?
I start to imagine a revisionist reading of Nineteen Eighty-Four in which Smith is revealed as an unreliable narrator, and the horrors of Airstrip One are revealed as the product of his paranoia — something along the lines of Dennis Lehane's Shutter Island. Not that Orwell intended such a reading, but all credit to him if his novel would support it, since a classic novel is, almost by definition, one that can be read at different times in different ways.
Some have called Nineteen Eighty-Four the greatest novel of the 20th century. It would deserve that acclaim if its impact had helped prevent the worst of its prophecies from being realised. At this safe distance in time, I doubt that was the case. If Orwell had died a couple of years earlier with Nineteen Eighty-Four unfinished, I doubt Britain would now be another North Korea. But I respect the argument and it has some merit.
If Nineteen Eighty-Four were published today, I suspect it would pass little-noticed. It would be classed as a work of dystopian science fiction and compared unfavourably by critics with the work of George Saunders, Ted Chiang, or Margaret Attwood. I do not mean by this to diminish Orwell's achievement, but rather to salute the flourishing of science fiction and speculative fiction in the decades since Orwell's death. Even our dystopias have got better.
The Fourth Man, by Robert Baer (2022)
I mentioned a couple of weeks ago how much I enjoyed The Company, by Robert Littell, a book about spies which was offered as a work of fiction. I have just now finished reading The Fourth Man, by Robert Baer, a book about spies which is offered as a work of non-fiction. Given the impossibility of checking what is fact and what is fiction in the world of espionage, even for those involved, I wonder if the distinction matters. If The Fourth Man had been published as a novel I would have been none the wiser.
The Fourth Man covers much of the same ground that The Company does (indeed, as many spy novels do), namely, the pursuit of of a double-agent in the upper reaches of CIA. The conclusions to Littell's fictional hunt and to Baer's supposed factual hunt are eerily similar: The answer has been hiding in plain sight.
As far as I know, Littell never worked for CIA. Baer spent 30 years there. Yet the minutiae of Littell's fictional CIA — the jargon, the daily habits, the personality types — are exactly those of Robert Baer's factual CIA. I see three ways to explain this coincidence:
— Littell, sitting at his desk, intuited brilliantly how CIA operates day-to-day;
— Novelists find it much easier to obtain detailed information from and about CIA than one might have imagined;
— The culture of CIA is shaped by its depiction in spy novels. Life imitates art.
The third of those possibilities is the most pleasing. John Le Carré believed it to be true of the British Secret Intelligence Service that young spies were learning their habits from his books. It also echoes the more readily documented ways in which the culture and conversation of the American Mafia are said to have reinvented themselves in the light of The Godfather. According to the New York Times:
Generations of mobsters have looked to The Godfather for inspiration, validation and as a playbook for how to speak and act and dress. Federal and local investigators on surveillance duty saw and heard made men and wannabes imitating the mannerisms and language of the screen gangsters.
Diego Gambetta captures the reasons for this mimicry in his book, Codes Of The Underworld:
How a real mobster should behave, dress, and speak are questions for which there is no optimal technical solution that presents itself independently of what others do and perceive as the meaning of their action. While criminals need conventional signals to communicate with each other and with the outside world, they are also hard put to agree on what these signals are and how to establish them credibly. They lack a coordinating and standardizing authority, and have to operate in secrecy. Movies can accidentally offer some solutions to those problems. What they offer is 'common knowledge', the foundation of coordination in the absence of a central authority. Coordination occurs when everyone knows that everyone else knows that s means k or that people like us always do j.
Of course, one thing that "people like us" do not generally do in CIA is to publish books accusing other CIA officers of being double agents.
So while I enjoyed Baer's Fourth Man, just as I enjoyed his earlier memoir, and his book about Saudi Arabia, I can see why others found it repugnant for Baer to imply on the basis of circumstantial evidence alone that a well-respected CIA veteran was a double-agent working for the KGB.
In fact, I found it amazing, accustomed as I am to strict British libel laws, that such a book could be published at all. I know things are different in America. But are CIA officers, of all people, "public figures"?