Machine learning may become a ubiquitous and fundamental enabling layer, similar to how relational databases impacted society in the eighties
We're now four or five years into the current explosion of machine learning, and pretty much everyone has heard of it. Tweets: @ryanspoon , @stevenjluck , @graph_zahl , @ethanhays , @abhijitbhaduri , and @valaafshar Tweets: Ryan Spoon / @ryanspoon : “one of the challenges in talking about machine learning is to find the middle ground between a mechanistic explanation of the mathematics on one hand and fantasies about general AI on the other” - @BenedictEvans http://www.ben-evans.com/... Steve Luck / @stevenjluck : Here's the key point: machine learning gives us automation, not intelligence. http://twitter.com/... Moritz Zajonz / @graph_zahl : How to think about #machinelearning: “[ML] gives you infinite interns, or, perhaps, infinite ten year olds.” (For the most part.) http://www.ben-evans.com/... Ethan Hays / @ethanhays : “As soon as we say ‘AI’, it's as though the black monolith from the beginning of 2001 has appeared, and we all become apes screaming at it and shaking our fists. You can't analyze ‘AI’. ” http://twitter.com/... Abhijit Bhaduri / @abhijitbhaduri : “Excel didn't give us artificial accountants, Photoshop and Indesign didn't give us artificial graphic designers and indeed steam engines didn't give us artificial horses.” What a terrific way to explain tech. #DigitalTsunami http://twitter.com/... Vala Afshar / @valaafshar : Unhelpful ways of talking about current developments in machine learning: —Data is the new oil —AI will take all the jobs —And, of course, saying #AI itself More useful things to talk about ML: —Automation —Enabling technology layers —Relational databases —@BenedictEvans http://twitter.com/...
Context & Ripple Effects
Writing in 2018, four or five years into the ML explosion, Benedict Evans staked out the middle ground Ryan Spoon described — between mechanistic mathematics and general-AI fantasies — with Steve Luck's compression of the thesis: machine learning gives us automation, not intelligence. The relational-database analogy matters because it predicts where value lands: in applications built on top of the layer, the way Excel's capabilities later spawned a wave of task-specific unbundling tools and thousands of API-siloed startups.
The subsequent coverage reads as a running test of that thesis. Stratechery argues AI's first successful wave looks like the first wave of computing — enterprise installations that cut jobs — while the Knight First Amendment Institute pushes the same logic into policy as AI as a normal technology rather than humanlike intelligence, and Emily Bender's work on what LLMs can and cannot do marks the boundary of what the layer actually automates.
First-order effects
- Companies that accept the 'automation, not intelligence' framing can ship ML as a feature inside existing products — the pattern by which Excel's capabilities produced task-specific unbundling tools rather than replacement platforms — instead of waiting on general AI.
- The framing debate lands hardest on the people building the layer: the strain and burnout ML scientists describe after ChatGPT shows how the industry's self-narrative bears directly on its workforce.
Second-order effects
- If adoption follows the enterprise-first pattern — installations that cut jobs before consumer-facing products — the near-term beneficiaries are corporate cost centers, and the hidden human labor beneath the layer, like the tasker annotation underclass hired via Scale AI, scales with it.
- Competing framings force a strategic fork for vendors: those selling ML as mundane infrastructure compete on distribution and integration, while those selling it as intelligence compete on capability claims that critics like Bender are positioned to test.
Third-order effects
- The Knight First Amendment Institute's 'normal technology' argument gives the relational-database view a policy corollary: governing deployment and diffusion beats building public policy around controlling superintelligence, which it warns may make things worse.
- If the pattern holds, ML settles into the commodity-layer role relational databases occupied — with value and differentiation migrating to the applications and workflows built on top, and the layer itself fading into assumed infrastructure.
The trend: Machine learning is following the relational-database path — a ubiquitous enabling layer whose value accrues to the applications built on it, with enterprise adoption first and the automation-versus-intelligence framing setting both product and policy stakes.