AI: Somewhere Between Magic and Doom

Happy Friday,

Last issue I gave you a recap of my speaking engagement in Vegas and where I think the conversation around artificial intelligence is headed. This is the first issue of October, so I want to pick up the thread and talk about labor (especially hot recently).

Earlier this week I watched a clip of Elon Musk standing outside the West Wing next to President Trump. He started to say "AI," corrected himself to "SI, pardon me" (Trump signed an order that day directing federal agencies to call it "superintelligence"), and then made a prediction: with robots and superintelligence, everyone in the world can have better medical care than anyone standing there, "including me." Musk happens to be on track to be the first trillionaire, so that is a bold promise from the man with the most to spend on a doctor. A reporter shouted back, "Not if they haven't got a job." Musk answered, "Well, jobs are going to change. Jobs have always changed."

Now, I am not here to hide behind a screen and pretend to dunk on him in true naive 21 year old fashion. I am here because that exchange shows the two extremes in one frame: abundance for everyone on one side, a jobless underclass on the other.

Our trust economy is broken, and the algorithms that feed you were built to surface whatever gets a reaction. Utopia gets a reaction. Doom gets a reaction. A careful sentence about "it depends on the work and the employer" gets scrolled past. With that being said, the middle is where most of the real value sits, and it happens to be where the evidence sits too.

Two economists at MIT's FutureTech project, Guy Ben-Ishai and Neil Thompson, published a review of the economic literature through Brookings on September 15. 2 weeks ago. Their read: the debate is usually framed as mass unemployment versus widespread human augmentation, and the evidence supports neither. What it supports is sharply divergent outcomes, with some work displaced quickly and other work augmented slowly, varying by occupation, industry, and even employer. Read that last part again, because it matters for what I tell you to do below. Even employer.

The Musk side has real counterweights, too. The IMF has estimated that AI will touch almost 40% of jobs worldwide and that, in most scenarios, it will likely worsen inequality, since the gains flow to whoever owns the models, chips, and data centers. Meanwhile Brookings noted this month that Dario Amodei's forecast of 20% unemployment has little evidence behind it so far. Neither prophet has the receipts yet. You and I get to live in the gap between them.

Much of my coursework revolves around recursive system intelligence, so let me give you the idea that scares people most. The scary version: machines begin teaching and learning from themselves to the point where we are no longer in the loop.

This is not science fiction as a concept. In 2017, researchers at DeepMind including David Silver and Demis Hassabis published AlphaZero, a system given only the rules of chess, shogi, and Go that taught itself by playing against itself, and it went on to beat the best existing programs in all three. No human games to study. No coach. Fast forward to this September: a Weco AI team posted a paper where a research agent rewrote its own code, tested the new version of itself, and kept what worked, finding seven successive improvements in an autonomous eight-day run. IBM covered the wider conversation on September 16, and the honest summary there is that OpenAI says fully autonomous recursive self-improvement is not happening today, Anthropic says its systems cannot yet build their own successors, and a Brown University computer scientist, Michael Littman, doubts the concept is even logically coherent, let alone imminent. IBM's Gabe Goodhart noted that many of today's improvement loops still have humans in the loop evaluating what the model suggests.

So: real, early, argued over by serious people. Scary? I understand why. But I do not want to scare you. I want you to steal the posture. A system that gets better by testing itself against reality, keeping what works, and throwing out what does not is a pretty good description of how humans have always grown. We are pretty damn awesome. The question for us is the same one we have always had: how do we keep teaching ourselves so we keep getting better?

Two quick examples. When you take a wrong turn and Google Maps reroutes you, that is artificial intelligence at work. We do not call it that. We call it Google Maps rerouting. When you type in Word and spell check fixes your grammar, that is also AI, and in fact some of its earliest forms. The second we demystify something and give it a plain name, it gets smaller in our minds. Fear loves a vague noun. "AI" is about as vague as nouns get.

Here is a part I think many of you will quietly agree with: AI is making real skills harder to evaluate. People can bullshit a lot more now. A polished cover letter, a clean portfolio, a confident interview answer can all be generated in minutes.

The numbers back it up. Western Governors University surveyed 3,128 US hiring professionals for its Workforce Decoded report, released September 30, and 60% said AI is making it harder to evaluate candidates' real skills. Nearly one in five said they struggle to confirm they are even interviewing a human. LinkedIn launched Hiring Assistant 2 on September 29, and its CEO told Fortune that job seekers are sending roughly 30% more applications than before the pandemic, which makes it harder to know who can do the job. When everyone can look excellent on paper, paper stops meaning much.

Here is the part I find encouraging. Inside that same WGU report, employers are looking to build talent, not just buy it: 42% are closing skills gaps by upskilling their current people, compared with 22% hiring new ones. Many companies are reinvesting in their people.

This is where I want to bring it down to ground level. Do not get caught up in the lofty ideology of what will happen to "jobs" in the abstract. Ben-Ishai and Thompson just told you outcomes vary by employer. So look at yours. Where are you working, and who are you working for? If you own your own business, the decision is in your hands. Most of you reading this do not, so critically evaluate the company you are in. How are they adopting AI? How are they choosing to invest in you? How are they upskilling you so that you stay in your job, and not just so that the job gets done by a cheaper tool?

Continuously thinking about how we develop our skills, and how we position our own life experience and knowledge to stay ahead of the machines, is ultimately what matters most, because it keeps the power and control with us. I will be honest that I do not know which of the two extremes is closer. Probably neither. What I do know is that I would rather bet on a person who keeps testing and rebuilding themselves than on any forecast, mine included.

So here is my question for you: if applicable, what is one thing your employer has done in the last six months to invest in you as a person, not just as output? If you cannot name one, that is information. Reply and tell me; I read every one. And if this sparked something, forward it to one friend who is stuck between the hype and the fear.

See you in two weeks,
Jaron