AI as Information Asbestos

The CEO of Microsoft, Satya Nedella, warned at Davos last week that unless AI adoption widens dramatically, we may be in a bubble. It was viewed as the latest sign that, yes indeed, we are in an AI bubble and some of the players are getting nervous.

• Talk about a “tell” at the poker table . . .
  • Perhaps the fact that my spellcheck — actually, really, I’m not making this up — corrected his name to “starchy Nutella” might also be a related warning that LLMs aren’t all they’re cracked up to be.

As I’ve documented the numerous failings of LLMs rolled out across the scientific publishing space, it has become clear we’re polluting the scientific record in a new and infernal manner.

Business and editorial firewalls are the normative functional dividing lines between the money side of a publishing business and the editorial side.

In a strong subscription journal, these functions are kept separate so that money doesn’t corrupt editorial decision-making.

• Gold OA journals eliminated the wall in an obvious and corrosive way.

The presumption in the business offices that still hew toward a separation of powers seems to be that LLM licenses and deals are akin to content licensing agreements with aggregators or search companies crawling and indexing a site.

But LLMs are fundamentally different beasts, atomizing, weighting, and regurgitating content in order to not simply represent existing editorial content in a different platform context but to generate new editorial content in stochastic and unpredictable ways.

This difference has significant implications as to who and how these agreements are made and managed, I’d assert.

Corey Doctorow compares the current spray of content from LLMs into our lives to the reckless adoption of a cancer-causing agent every homeowner dreads discovering because the abatement work is expensive and intrusive:

AI is the asbestos in the walls of our technological society, stuffed there with wild abandon by a finance sector and tech monopolists run amok. We will be excavating it for a generation or more.

Mixed in with wallboard and cement in foundations, asbestos was everywhere for a time, and now information asbestos is now everywhere as LLMs are being pushed to scale.

Worse, LLMs at top journals are injecting cancerous asbestos fibers into business and editorial firewalls, a new carcinogen neither side may detect or even know to look for given how relentlessly companies are pushing for mindless adoption given the money at stake.

It’s not like medical journals haven’t dabbled in monied carcinogenic products before:

AI may just be today’s version:

Why would journal editors sign off on their journals being used to advertise information carcinogens?

Because they don’t view them as carcinogenic – yet.