Human in the Loop

The Sky’s Not Falling. Probably.

• By Howard M. Cohen
• 11/04/2026

“In God we trust, others must provide data.”
(Pathologist Edwin R. Fisher, quoting a scientific cliché during House testimony in 1978.)

In the AI age, there’s no shortage of predictions. The harder question is whom to believe.

The meteoric growth of AI suggests many people are very enthusiastic about it and aren’t afraid to say so.

Then there are the “doomers,” so numerous they are split into two camps!

The existential doomer fears that AI will escape human control and threaten humanity’s survival. This would, to put it mildly, end badly for everyone. Eliezer Yudkowsky, a co-founder of the Machine Intelligence Research Institute, and Nate Soares are among the most forceful advocates of this view. The title of their 2025 book leaves little room for ambiguity: If Anyone Builds It, Everyone Dies.

Geoffrey Hinton, often called “the Godfather of AI,” and Yoshua Bengio, co-chair of the UN’s Independent International Scientific Panel on AI, also warn about the possibility of losing control of AI. Such concerns are often expressed as “p(doom),” a subjective estimate of the probability that advanced AI causes human extinction or a comparable catastrophe. The “p” stands for probability, and “doom” is the outcome.

The economic doomer fears something nearer and perhaps duller but still life-altering: mass job loss, arriving faster than workers and institutions can adapt. Anthropic CEO Dario Amodei’s entry-level warning and a staff report for Senator Bernie Sanders belong here.

Most everybody has an opinion. The optimists rattle on endlessly about how AI will not take away jobs, but the jobs will change, and we’ll have to train for the new jobs. So far, I haven’t heard much detail about what those new jobs will be.

However, the concept is not new. John Maynard Keynes used the phrase “technological unemployment” in 1930.

People point to advances including the locomotive, the electric light, the telephone, the photocopier, and, of course, the internet as examples of technologies that made some jobs obsolete while creating others.

Valid Arguments
I’m prepared to acknowledge that valid arguments exist on both sides of the debate over AI’s possibly apocalyptic future. So I looked at what technologists, lawmakers, economists, and reporters are saying about the future role of humans in AI.

Surprisingly, one optimistic pattern jumped out at me. Many of these arguments still give people a central role. Much of the disagreement is over where people will ultimately fit in the new Digital World Order.

Leaders of the firms developing AI for all of us have a wide variety of perspectives to share. Here are some of the opinions I found.

AI’s Chief Cheerleader
I will admit that I have trouble listening to Nvidia CEO Jensen Huang speak. As AI’s “optimist-in-chief,” he gives any snake oil salesman a run for his money. He told Axios in July that “AI is creating an enormous number of jobs, not taking it away.” He sells the picks and shovels those jobs will need, so adjust your response accordingly. Axios noted in the same story that AI now shows up in layoff announcements and that hiring of young workers may be taking the hit.

Huang’s favorite proof is radiology. At Davos in January, he argued that AI helps radiologists process more patients, increasing demand for their services. The result, he said, is more radiologists, not fewer.

The Optimist from Redmond
In February, Microsoft AI chief Mustafa Suleyman said most computer-based office tasks would be fully automated within 12 to 18 months. In June, he clarified that he meant tasks, not jobs.

On October 7, he shared Nobel laureate Daron Acemoglu’s estimate that about 5% of U.S. work would be automated within 10 years. That measures economy-wide adoption, a different question from what AI might technically be capable of doing. Acemoglu also pointed to the continued need for human expertise: “Hospitals still need radiologists (in fact more than before).”

Radiologists again. Why does every optimist reach for them? They were among the first professions marked for extinction, and they are still here, billing.

We May Have to Change This Column’s Name
Salesforce’s Paula Goldman argues that “in the loop” is the wrong shape entirely. Nobody will verify 50,000 data points by hand, she told CXOTalk, and oversight built that way defeats the purpose of an agent. Her alternative is what Salesforce calls “human at the helm.” People set the goals, the limits, and the escalation triggers, then watch the audit trail for anomalies. Joe McKendrick reported her argument in Forbes on September 29.

The Lawmakers
California has put human oversight into law. Governor Gavin Newsom signed SB 947, the No Robo Bosses Act, on September 30. The law bars employers from relying solely on AI to fire or discipline workers and requires human oversight and verification when AI assists in those decisions. The bill’s author, Senator Jerry McNerney, said, “AI must remain a tool controlled by humans, not the other way around.”

In Washington, Senators Ed Markey and Brian Schatz introduced a federal No Robot Bosses Act in June. It would let workers opt out of algorithmic management in favor of a human manager. Schatz summed it up: “Workers should be managed by people”, not robots.

Arizona’s Mark Kelly introduced the Make AI Work for Americans Act, which would make the companies profiting most from AI invest in workers. “This is something we have to tackle immediately,” he said.

Senators Mark Warner and Josh Hawley, one Democrat and one Republican, have pushed the AI Jobs Clarity Act. It would require major employers and federal agencies to report AI-related layoffs to the Department of Labor. A bipartisan group of senators also pressed federal agencies in March for better data on AI’s labor impact.

Notice what the human-oversight measures protect. They protect people from machine decisions about them. They do not guarantee that the job continues to exist.

The Academics
Johns Hopkins professor Ritu Agarwal puts it plainly: “Humans will steer the ship and be needed in different capacities.” Her colleague Rick Smith adds the uncomfortable follow-up. Human judgment remains a safeguard against hallucinations, but AI is swallowing the routine work juniors learn on. Where do tomorrow’s experienced professionals come from?

Harvard Business School’s Raffaella Sadun calls the temptation to automate away investment in people “a terrible trap.” Her reasoning: employees carry tacit, firm-specific knowledge that makes them worth far more than their job descriptions suggest.

Sadun's colleague Joseph Fuller found that 41% of work tasks could be automated or augmented, yet only about a third of company AI experiments succeed. Companies, he says, are not training people formally.

Kennedy School economist Doug Elmendorf offers the sober footnote. He expects significant job losses over the coming decades, though he believes most displaced workers will find new work.

The Warning Label
CBS News reporter Shaun Raviv examined another danger. Humans keep their jobs but stop thinking, reduced to rubber stamps.

Hugging Face scientist Margaret Mitchell’s remedy is “strategic friction,” deliberate steps where a person must engage. Her warning: when people stop practicing a skill, no one may be left who knows how to fix the system when it breaks.

Anthropic CEO Dario Amodei warned in May 2025 that AI could eliminate half of entry-level white-collar jobs within one to five years, as Axios reported.

And an October 2025 staff report for Senator Bernie Sanders warned that AI and automation could destroy nearly 100 million U.S. jobs in a decade.

How Humans Remain in the Loop
So where does this leave you? I have a few suggestions:

• Write down which decisions stay human. Goldman’s question is blunt: what should technology decide, and what must a human answer for? Put the answer in writing before the agent ships.
• Set the controls, not just the checkpoints. Define the agent’s goals, its out-of-bounds list, and its escalation triggers. Then insist on an audit trail that tells you exactly what it did.
• Add friction on purpose. Pick the steps where a person must think instead of click. Mitchell’s point is that smoothness has a price, and your skills pay it.
• Protect the on-ramp. Fuller’s example is worth stealing. One large tech company rewrote junior engineering jobs so those hires still code while getting structured exposure to sales and product management. Ten years from now, those people could be your senior staff.

AI models may give the impression of having a presence, but I don’t take that as evidence of sentience, and neither should you. They have no dog in any hunt, no personal investment, financial or emotional. Itruly doubt they ever will. But keeping humans in charge will take more than confidence that machines don’t care. Humans still need to determine the direction and growth of our world and everyone in it. That means keeping the skills and authority to do so.