It's #Pride month, and #PatRobertson kicked it off with a bang by...kicking off. What better way to start Pride than with a piece of fanfic by #WilWheaton about Robertson's arrival in Hell?

https://wilwheaton.net/2023/06/the-wait/

Once you've had your fill of #Schadenfreude, get your Pride/#Trekkie (or, if you prefer, #Trekker) gear on with Wil's Acting Ensign Pride collection:

https://shopstands.com/collections/wil-wheaton

2/

Seguing smoothly into science fiction by way of Trek, let's turn to #TedChiang, a titan of the field, who is also a wicked-sharp critic of #AIHype. Ted's been at this since at least 2017, when he identified tech CEOs' fears of #AI as a form of transference of their fear of corporations, which are, after all, autonomous artificial life-forms that are rapidly devouring the human race:

https://www.buzzfeednews.com/article/tedchiang/the-real-danger-to-civilization-isnt-ai-its-runaway

3/

When it comes to inventing devastating metaphors for AI, this is Ted's busy season. "AI is a blurry JPEG of the web":

https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web

It's also "the new McKinsey":

https://www.newyorker.com/science/annals-of-artificial-intelligence/will-ai-become-the-new-mckinsey

And now, in the *#FinancialTimes*, Ted sits down for lunch with Madhumita Murgia to talk about the linguistic game we play when we describe a statistical inference tool as "artificial intelligence," given that it is neither "artificial," nor "intelligent":

https://www.ft.com/content/c1f6d948-3dde-405f-924c-09cc0dcf8c84

4/

Start with whether machines "learn": "#MachineLearning" is just adjusting weights in a statistical model. When you teach a child something, you're not adjusting weights! "Machine learning" is a useful metaphor for thinking about a subset of applied stats, but it's also a trap, tricking us into unconsciously anthropomorphizing an intelligence behind plausible sentence generators. We talk about AI models "hallucinating" - another linguistic trap.

5/

But the real "AI hallucination" is when we wet human people hallucinate a dry, digital intelligence behind the plausible sentences.

Calling it AI, saying that it learns or hallucinates or knows or understands - these are hallucinatory traps for wet squishy humans. Even the fact that chatbots use the pronoun "I" is a slippery slope into imagining an intelligence on the other side of the keyboard.

What should we call this discipline, if not "AI"? Ted says, "applied statistics."

6/

Applied statistics can automate a lot of work away, but what it's best at is automating the #BullshitJobs that #DavidGraeber (#RestInPower) described in his brilliant 2018 book:

https://memex.craphound.com/2018/06/20/david-graebers-bullshit-jobs-why-does-the-economy-sustain-jobs-that-no-one-values/

An academic friend tells me that they use #LLMs to write recommendation letters for grad students, and that grad students use LLMs to take minutes on department meetings.

7/

Presumably, someone else is using LLMs to ingest and summarize these recommendation letters and departmental meetings. LLMs can inflate a few bullet points into several florid paragraphs, and deflate them back into bullet points, with significant semantic losses on the way.

Is there a non-bullshit use for these? Maybe. I have spent a lot of my activist career as an anti-bullshit actor.

8/

For example, I was among the first public interest delegates to #WIPO, the most industry captured UN specialized agency, which has the same relationship to terrible copyright proposals that Mordor has to evil - a kind of infinite wellspring of bad ideas.

These ideas emerge out of an extremely bullshit process.

9/

The Standing Committee on Copyright and Related Rights meets periodically in Geneva for a highly stylized, multi-day session in which national delegates rise and deliver cryptic, stilted remarks about various proposed clauses to awful treaties like the #BroadcastTreaty.

These remarks are recorded by the Secretariat, who then offers each delegation the opportunity to redact or alter the official record of their remarks.

10/

Then, six months later, the Secretariat publishes this revisionist version of the session, which would be impossibly dull and cryptic even without all the revisions. This is the Shield of Boringness (h/t @danacorn) in action.

One day at one of these meetings, @wseltzer had a brilliant idea: let's make our *own* transcript.

11/

There was no public internet at WIPO back then, so we set up an ad-hoc network off one of our laptops (thankfully, they'd just installed electrical outlets at some of the NGO delegation seats), then used #Etherpad to create a shared document. We traded off transcribing the remarks with correcting typos and then annotating the text with plain-language descriptions of what was really going on.

12/

Then we published: twice a day, at the lunch and dinner breaks, unplugging the Ethernet cables from one of the shared PCs in the mezzanine and uploading our transcripts. These hit the nerd sites of the day, like @slashdot, and created realtime pressure on national delegations that had caved to industry demands at the expense of their populations.

13/

They started to get calls from their capitols demanding explanations - andthe delegates who'd taken brave stands for the public interest were praised (for the first time ever!) by their bosses in distant ministries.

It worked so well that other NGOs started sending delegates to these meetings, and I started schlepping a wifi access point and a power strip to the meetings, so that several of us could collaborate on even more detailed realtime transcripts and annotations.

14/

We also collaborated on coordinated remarks, because each NGO was typically only allowed to speak for 1-2 minutes at then end of a 1-3 day meeting, so we drafted a unified set of comments that we delivered as a serial when the chair called on us.

These unofficial transcripts became the de facto record of the WIPO meetings.

15/

I used to run into national delegates who'd been rotated in after a ministerial change in their home countries, who thanked me for our work and said it was the only way they could follow (and thus participate in) the proceedings.

We were setting the agenda, in other words. It was pretty cool, and it made the WIPO establishment furious.

16/

The secretariat - a veteran of the US Trade Rep's office who made her bones cramming brutal, human-rights-abusing trade terms on the textile workers of south Asia - threatened to expel us.

All that to say: we could have done even more if we'd had a reliable automated transcription tool.

17/

The valuable part of the work was the annotations, not the transcription, and we were always shorthanded, and automated transcription would have freed up a set of hands - and a mind - to make sense of the delegates' remarks and explain them to others.

On the subject of tech regulation and AI: if you're not reading #SayashKapoor and #ArvindNarayanan's AI Snake Oil newsletter, allow me to gently suggest that you consider it:

https://aisnakeoil.substack.com/

18/

The latest edition is "Licensing is neither feasible nor effective for addressing AI risks," and it does exactly what it says on the tin:

https://aisnakeoil.substack.com/p/licensing-is-neither-feasible-nor

19/

The issue here is #SamAltman - CEO of OpenAI, a company that is not "open" and whose products are neither "artificial" nor "intelligent" - demanding that Congress create an international licensing regime for products that compete with his own #ChatGPT, which loses a large amount of money on every query and can only be profitable if it has no competition to get in the way of jacking up prices once other businesses are thoroughly dependent on its services.

20/

As Kapoor and Narayanan explain, the licensing regime that Altman demands would:

* Produce a dangerous monoculture in which every app would have a shared set of vulnerabilities that could be exploited by bad actors;

* Homogenize the products of "AI" tools - e.g. every resume-sorting bot would have the same blind-spots and irrational exuberances;

21/

* Give a small number of firms control over the Overton window, letting their products define which ideas and sentiments get sorted to the top of your inbox or social media feeds;

* Centralize control over opinion-formation, with licensed companies controlling how complex ideas are summarized ("those aren't bald spots, they're solar panels for sex machines!");

22/

* Lead to regulatory capture: when an industry is dominated by a handful of large firms, it's much easier for them to converge on a set of lobbying priorities, and their cozy oligopoly lets them extract sufficient profits that there's plenty of cash to spend on making those lobbying priorities into policy reality.

23/

Kapoor and Narayanan favor "development and evaluation of state-of-the-art models by a diverse group of academics, companies, and NGOs" and promise future work on risk assessment and guardrail development.

A key element of any AI policy framework is data acquisition, processing and utilization.

24/

The #AdaLovelaceInstitute's giant "Rethinking data and rebalancing digital power" report is a *banger* on this subject, covering interoperability, privacy, equity, information security and more, with superb contributions from @1br0wn and #JathanSadowski:

https://www.adalovelaceinstitute.org/report/rethinking-data/

25/

The privacy debate changed forever a decade ago, when #EdwardSnowden handed a group of journalists a trove of #NSA documents detailing a massive, lawless global surveillance campaign. The tenth snowdenversary has prompted a lot of commentary. I really liked #AlanRusbridger's retrospective:

https://www.theguardian.com/commentisfree/2023/jun/06/edward-snowden-state-spying-guardian-alan-rusbridger

26/

Rusbridger was the editor-in-chief of *#TheGuardian* during the Snowden publications, who reminds us that whistleblowers continue to meet with cruel treatment and punishment, rather than the celebration they're due. He also reminds us that editorial independence is key to the brave reporting that whistleblowers rely on: when he was publishing the Snowden revelations, his bosses were *incapable* of ordering him to stop.

27/

They could have fired him, but they were not permitted to override his editorial judgments.

Another good Snowden take comes from #EwenMacAskill, the former *Guardian* defense and security correspondent, who was one of the original Snowden reporters:

https://www.theatlantic.com/ideas/archive/2023/06/edward-snowden-nsa-mass-surveillance/674315/

MacAskill tells us that neither he nor Snowden have any regrets about their decision.

28/

More to the point, he quotes the #ACLU's #BenWizner, Snowden's lawyer, who reminds us that as dismal as Snowden's exile in Russia is, it's far better than what everyone expected at the time - lifetime confinement to a Gitmo-style American gulag or worse, a firing squad.

29/

(Snowden wasn't trying to get to Russia - he was aiming for Ecuador, but Secretary of State #JohnKerry canceled his passport after his flight took off from Hong Kong, which gave the Russians the pretext they needed to detain and effectively kidnap him.)

The Snowden leaks ushered in an era of mass encryption, ending the age in which most data was sent or stored "in the clear."

30/

From the mass storage on your phone to the web sessions your browser initiates to the instant messages you send and receive, the age of cleartext is over.

Meanwhile, the claims by the spy agencies - who have proved time and again that they will lie to the public and their democratically elected overseers about their illegal surveillance - that Snowden did untold harms to national security remain as empty as they were a decade ago.

31/

As Snowden told MacAskill: "Disruption? Sure, that is plausible. But it is hard to claim 'damage' if, despite 10 years of hysterics, the sky never fell in."

It took a brave, independent press to publish the Snowden revelations, but even a decade ago, the press was ailing. Big Tech's chokehold over subscription payments, ads, and delivery of content has allowed it to steal a fortune in cash from the news business.

32/

Unfortunately, the media - and its friends in government - have decided that the real problem is that tech is stealing "content," not money, leading to proposals to restrict who can link to, quote and discuss the news. For the past month, I've been working with @eff on a series describing how tech steals money (not content) from the news, and what to do about it:

https://www.eff.org/deeplinks/2023/04/saving-news-big-tech

33/

Though I disagree with people who say tech is stealing news content, I firmly agree that the collapse of the news industry is bad news for society. Indeed, one of the reasons we desperately need an independent press is to ensure critical investigations of the tech industry - something we're more likely to get if the news isn't "partnered" with tech for its survival.

34/