The annual meeting of Fortune’s CEO Initiative, a gathering of leaders who want their companies to be a positive force in the world, concluded Tuesday in New York City. CEOs generally are aware there’s a problem. Their companies don’t create enough jobs, their employees don’t have the right skills, the technology that made business so efficient is alienating its users, and there’s no consensus on how to make commerce more transparent and trustworthy.

These are the broad strokes of an exhausting and exhilarating day-plus of conversations among well-meaning business leaders, more or less aware of their problems and more rather than less interested in fixing them.

I moderated three plenary sessions Tuesday:

Raghuram Rajan, a University of Chicago economist, posits that a bottoms-up focus on community can right the imbalance that’s been created in the developed world from too much emphasis on centralized government and markets. (Watch our interview here.)

A “town hall” format lunch discussion surfaced few executives willing to own up to capitalism’s breach of trust with its less-well-off members. (Exceptions: Investor Tim Collins, who thinks another wave of political change is the answer, and Allstate CEO Tom Wilson, who suggests companies should publicly quantify their commitment to job creation.)

Journalist David Brooks earned a quiet-enough-to-hear-a-pin-drop reception and a standing ovation for his passionate paean to a life of meaning, which he dubs the “second mountain.” Brooks contends there are “first mountain” people and companies, those reaching hard for the brass ring and trying to win at any cost, and their second-mountain counterparts, those trying to be moral, good, and fulfilled, even at the expense of maximizing profits.

NEWSWORTHY

Zuck gets deep-faked: British artists published a video of Facebook king Mark Zuckerberg gloating like a James Bond villain about all the data he controls. The video is an apparent retort to Facebook's decision not to remove a recent viral clip doctored to make Nancy Pelosi appear drunk. The company issued a limp statement that it would tell its recommendation engines not to promote the Zuck clip if fact checkers marked it as false.

State of the Internet 2019: It's that time of year when VC and Internet maven Mary Meeker unveils her famous slide deck of web trends and trouble spots. Highlights from the 330 deck: tech firms are focusing on privacy; more people are using images to communicate; Twitter and Amazon are making small in-roads into the Facebook-Google ad duopoly; 51% of the world's population is now online.

Uncle Sam loses your data (again): The recent theft of license plate and photo data from the U.S. Customs and Border Patrol has some familiar and depressing lessons. Namely, compiling large amounts of data creates an inviting target. Also, third party contractors are the weakest link (in this case, a Tennessee vendor that makes license-plate scanning software appears to have been compromised).

DOJ gunning for Google: The Justice Department’s antitrust chief made remarks that should cause the Big Tech companies, especially Google, to start sweating. He invoked now-busted trust Standard Oil and brushed off claims the DOJ should limit its antitrust inquiry to consumer prices.

Google fires its lobbyists: In a counterintuitive move, the search giant is paring back its legions of lobbyists and influence peddlers. Insiders say the overhaul reflects how its existing operation—which saw the number of lobbying firms on retainer expand from four to 30 over the years—has become outdated.

FOOD FOR THOUGHT

Machine learning is an environmental nightmare:  It turns out training all those algorithms is a lot more energy intensive than one might think. MIT researchers measured the environmental impact involved with building natural language processing models, and discovered each of them produced the equivalent of the lifetime emissions of five cars. Most of the energy expended, however, came in a final process of fine-tuning that did little to improve the models:

They found that the computational and environmental costs of training grew proportionally to model size and then exploded when additional tuning steps were used to increase the model’s final accuracy. In particular, they found that a tuning process known as neural architecture search, which tries to optimize a model by incrementally tweaking a neural network’s design through exhaustive trial and error, had extraordinarily high associated costs for little performance benefit. Without it, the most costly model, BERT, had a carbon footprint of roughly 1,400 pounds of carbon dioxide equivalent, close to a round-trip trans-America flight for one person.

IN CASE YOU MISSED IT

KKR Mints a New Cybersecurity Unicorn By Rey Mashayekhi

How to Avoid Winding Up As Pets for A.I. Robots By Robert Hackett

How Facial Recognition Will Change Driving By Jeff John Roberts

IMF Chief Economist: How Amazon Bungled H2Q By Lucinda Shen

Facebook Debuts a New Tracking App Just Months After Controversy Over a Similar App By Danielle Abril

Salesforce CEO Marc Benioff Swears He Is 'Not an Activist' By Robert Hackett

BEFORE YOU GO

Even as men do more housework than in the past, it is still women who are blamed for a messy house, new studies suggest: "When participants were told that a woman occupied the clean room, it was judged as less clean than when a man occupied it, and she was thought to be less likely to be viewed positively by visitors and less comfortable with visitors."

This edition of Data Sheet was curated by Jeff John Roberts. Find past issues, and sign up for other Fortune newsletters.