Note: This post is also shared on LinkedIn.

The “AI” label: Constantly shifting target or recently overblown? My perspective on this, engaging with the insightful post by Dr. Sebastian Wernicke: lnkd.in/etaw_-fi

The original post argues “AI” was first overused in marketing statistical tools. Now, with advanced generative AI (LLMs), it says a new error is treating them as mere upgrades for traditional ML, not distinct tools for unstructured data/generative tasks, thus misjudging their strengths.

I partly disagree.

First, what is AI?

Cédric Villani reportedly called AI “unfinished computer developments brimming with fantasies and capable of generating buzz” (lnkd.in/ezQH7BWc).

Mustafa Suleyman noted AI is often joked as “‘what computers can’t do.’ Once they can, it’s just software” (The Coming Wave, 2023).

Academically, Berente et al. defined AI as “the frontier of computational advancements that references human intelligence in addressing ever more complex decision-making problems” (MIS Quarterly, 2023).

Common theme: AI is dynamic, a moving target redefining itself as capabilities emerge. Accepting this, it is expected everything in the field gets labeled AI eventually.

The view that LLMs “truly deserve” the AI label over earlier models, while understandable, may overlook this shifting benchmark. LLMs are powerful, but so was ELIZA in 1966 to its users. Breakthroughs come, dust settles, new hype begins.

Popular LLMs use the 2017 Transformer architecture, a deep learning method similar to earlier neural networks, still reliant on gradient descent, a key driver of AI progress.

For a more irreverent take: lnkd.in/eujJF4gb

Second, was the term AI overused for years?

In my view, professionals long used “AI” with parsimony. PAMI, a top journal (the “Nature” of computer science), does not even have “AI” in its title. Terms like data mining, pattern analysis, computer vision, and NLP were common.

Only after ChatGPT’s Nov 2022 release did “AI” become a catch-all, from basic arithmetic/linear regressions to LLMs.

Early 2023 earnings calls chorused “AI.” Firms with perhaps just old linear regressions in supply chains touted AI capabilities, some mentioning it dozens of times, boosting tech stocks by an average of 11.9%.

A notable exception: Apple. Zero “AI” mentions in their Q2 2023 earnings call (once in Q&A), sticking to “machine learning.”

Then came Apple Intelligence, a more Apple-like way of spelling “AI.”

Now everyone is using it, even Apple, and it has come to mean just about anything.