Stop Deleting Your Data: Physical AI and the Grid | Grid Mod Pod Ep. 92 with Sean Murphy

Value of the Information & Strength of the Argument

The episode’s informational value lies in its clear articulation of a structural mismatch between the grid’s physical timescales and its data-handling practices. Murphy provides a coherent conceptual framework: physical AI as AI for machines about machines, trained on time-series measurements from physical infrastructure rather than human-generated digital exhaust. He identifies concrete fast phenomena — inverter-based oscillations documented in Europe and the U.S., data center load trips, and 10-millisecond wildfire ignition events — that fall below SCADA’s temporal resolution. The argument that the expensive part of sensing (hardware and truck rolls) is already done, and that only the decision to retain data is missing, is a strong and actionable point. The smartphone double standard and the AMI example are effective rhetorical devices that expose an inconsistency in how utilities evaluate ROI. However, the argumentation remains largely qualitative. The claim that two use cases pay for all storage is asserted rather than demonstrated with figures. The ‘60-plus AMI use cases’ figure is repeated without attribution. The Jevons paradox reference is invoked briefly but not developed. The discussion of why vendors sell silos is plausible but presented as opinion. Overall, the reasoning is internally consistent and grounded in industry experience, but it relies on persuasive framing more than on quantitative evidence or systematic analysis.

Scientific Rigor, Source Quality, Title Accuracy

Scientific rigor is moderate. The guest is a credible practitioner with an academic background and multiple data-focused companies, and the host holds a technical strategy role at AEIC. The technical claims about SCADA sampling intervals, synchrophasor capabilities, and inverter-based resource behavior are consistent with established power systems literature, but no specific studies, reports, or datasets are cited during the episode. The description links only to AEIC’s website, LinkedIn, and Spotify — no external references or papers are provided. The title ‘Stop Deleting Your Data: Physical AI and the Grid’ accurately captures the episode’s central thesis and is not misleading. The chapter markers align with the content. The episode is transparent about its advocacy position: it is a call to action for utility leaders, not a neutral survey. This is appropriate for a podcast format but means listeners should treat the quantitative claims as illustrative rather than verified.

The title accurately reflects the core thesis of the episode: utilities should stop deleting grid data to enable physical AI. The subtitle and chapter markers align well with the content discussed.

7/10

The episode features a credible industry expert (Sean Murphy, founder/CEO of PingThings) and is hosted by a technical strategy VP at AEIC, a long-established utility industry association. Claims about sensor deployment, SCADA sampling rates, inverter-based oscillations, and AMI use cases are consistent with known industry literature. However, the discussion is largely qualitative and anecdotal; no specific studies, datasets, or peer-reviewed references are cited on air. The '60+ AMI use cases' and '10,000 measurements per second downsampled to 4 readings/hour' figures are plausible but not sourced in the transcript. The framing is advocacy-oriented (stop deleting data) rather than neutral analysis.

• 00:00Stop deleting the data
• 00:37Intro
• 01:39Sean Murphy, PingThings, and a recovering data scientist
• 02:18The cheat code: a grid run on physics equations
• 05:04What physical AI actually means
• 06:59Digital exhaust, and fifty years of grid sensing
• 09:39The sensors are already out there, switched off or downsampled
• 11:37Microseconds to hours: every timescale the grid runs on
• 14:01Why the grid moves slower than tech
• 15:50Proving value before you can unlock it
• 17:15The horsemen of the grid apocalypse
• 22:28What it really costs to keep the data
• 25:35The smartphone double standard
• 26:46AMI: sixty use cases nobody counted
• 30:06Change the rules, or have this conversation again in 2050
• 31:53"Tell me what to do with data I've never seen"
• 33:49Jevons paradox: make use cases easy to build
• 34:22AEIC's Data Framework Working Group
• 34:54Why vendors sell silos
• 37:41What to stop doing: stop deleting your data
• 38:29It exists today, at a few forward-thinking utilities

• AEIC Website — Association of Edison Illuminating Companies, the organization hosting the podcast and its Data Framework Working Group.
• AEIC LinkedIn — Professional network page for AEIC, referenced in the episode description.
• Grid Mod Pod on Spotify — Podcast distribution channel for the Grid Mod Pod series.

• U.S. Department of Energy — Synchrophasor Technologies and their Deployment in the Recovery Act Smart Grid Programs — DOE report documenting the deployment of phasor measurement units and the value of high-resolution grid data, consistent with Murphy's argument.
• NERC — Inverter-Based Resource Performance — NERC's inverter-based resource disturbance reports document oscillations and tripping events similar to those described in the episode.
• EPRI — Advanced Metering Infrastructure — EPRI research on AMI use cases beyond billing, supporting the claim that smart meters enable dozens of applications.