AI Management: Excellence in Distancing Decisions from Consequences
July 2, 2025
![Dino 5 18 25_thumb[3]](https://www.arnoldit.com/wordpress/wp-content/uploads/2025/06/Dino-5-18-25_thumb3-1.gif "Dino 5 18 25_thumb[3]")
Smart software involved in the graphic, otherwise just an addled dinobaby.
This write up “Exclusive: Scale AI’s Spam, Security Woes Plagued the Company While Serving Google” raises two minor issues and one that is not called out in the headline or the subtitle:
$14 billion investment from Meta struggled to contain ‘spammy behavior’ from unqualified contributors as it trained Gemini.
Who can get excited about a workflow and editorial quality issue. What is “quality”? In one of my Google monographs I pointed out that Google used at one time a number of numerical recipes to figure out “quality.” Did that work? Well, it was good enough to help get the Yahoo-inspired Google advertising program off the ground. Then quality became like those good brownies from 1953: Stuffed with ingredients no self-respecting Stanford computer science graduate would eat for lunch.
I believe some caution is required when trying to understand a very large and profitable company from someone who is no longer working at the company. Nevertheless, the article presents a couple of interesting assertions and dodges what I consider the big issue.
Consider this statement in the article:
In a statement to Inc., Scale AI spokesperson Joe Osborne said: “This story is filled with so many inaccuracies, it’s hard to keep track. What these documents show, and what we explained to Inc ahead of publishing, is that we had clear safeguards in place to detect and remove spam before anything goes to customers.” [Editor’s Note: “this” means the rumor that Scale cut corners.]
The story is that a process included data that would screw up the neural network.
And the security issue? I noted this passage:
The [spam] episode raises the question of whether or not Google at one point had vital data muddied by workers who lacked the credentials required by the Bulba program. It also calls into question Scale AI’s security and vetting protocols. “It was a mess. They had no authentication at the beginning,” says the former contributor. [Editor’s Note: Bulba means “Bard.”]
A person reading the article might conclude that Scale AI was a corner cutting outfit. I don’t know. But when big money starts to flow and more can be turned on, some companies just do what’s expedient. The signals in this Scale example are the put the pedal to the metal approach to process and the information that people knew that bad data was getting pumped into Googzilla.
But what’s the big point that’s missing from the write up? In my opinion, Google management made a decision to rely on Scale. Then Google management distanced itself from the operation. In the good old days of US business, blue-suited informed middle managers pursued quality, some companies would have spotted the problems and ridden herd on the subcontractor.
Google did not do this in an effective manner.
Now Scale AI is beavering away for Meta which may be an unexpected win for the Google. Will Meta’s smart software begin to make recommendations like “glue your cheese on the pizza”? My personal view is that I now know why Google’s smart software has been more about public relations and marketing, not about delivering something that is crystal clear about its product line up, output reliability, and hallucinatory behaviors.
At least Google management can rely on Deepseek to revolutionize understanding the human genome. Will the company manage in as effective a manner as its marketing department touts its achievements?
Stephen E Arnold, July 2, 2025
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
