Ai impacts -- five year modeling

Anthropic recently released a paper on the impact of AI on the economy over the next five years. They had three scenarios.

In the modest scenario, GDP is 1.6% above the no-AI path by 2030, annual growth reaches 2.4%, and the job market barely changes. The extreme scenario produces an economy 32% larger than the no-AI baseline, growing at 15% annually, but with overall unemployment at 12%. AI gets much more work done while many displaced people struggle to find another job.

The mid-scenario has a strong productivity and investment boom with plenty of work left for us humans. It puts GDP 8% above its path without AI. Annual growth reaches 5.4%, compared with two percent in the baseline. The capital stock is 14% larger (more productive equipment and other assets available to businesses).

DB2

2 Likes

Is this assuming we all get to live?

JimA

7 Likes

Yes. Otherwise that would be Scenario #4. :slightly_frowning_face:

DB2

Why should I trust Anthropic to have an unbiased opinion on this subject? They absolutely have a financial interest in promoting the positive side of AI.

8 Likes

Well, they’re not unbiased, so it’s best to focus on the relative effects of the different scenarios.

DB2

1 Like

Extrapolations are always risky. Should be used as a possible scenario. But does it impact your investing decisions? Doubtful.

You wonder if AI was used to create the estimates. And would AI promote its self image.

I found it interesting how close the compound interest formula could come to estimating future net worth given reasonable estimates. Extrapolation can work but not always. It’s a roll of the dice.

And why should I even trust that much of the analysis? The bias could be in the relative differences.

I consider the entire report to be worthless.

6 Likes

Always take the results of modeling projections with a large grain of salt.

DB2

2 Likes

This seems the most likely. I’m surprised they left out the unicorns and rainbows, however. Also, everybody gets a puppy.

We were pretty much at the “no-AI” path a couple years ago. So I should expect GDP to be around 8%? I’ll take that bet, and offer 1% back. I predict it to be under 7%, perhaps well under, and with unemployment rising thanks to the diminution of the data-center building craze coupled with the effects of the Fed finally trying to get a handle on inflation. (It took 3 years of aggressive moves couple with two grinding recessions to finally get it back under control. Of course it’s not as bad now as it was, but then neither is the Fed chief [nor Treasury Secretary] as competent as the one was back then.)

7 Likes

This morning we were offered $10k as well. Mockingly, but nonetheless, count me in, baby.

AI does not get it. We are now in the extreme excess modes in the marketplace. This goes to hell directly; do not pass go.

Claude takes a while to calculate an answer, but Claude would need a better question to say we are going to hell in the markets.

Two of your premises don’t match the data, which changes the odds a lot.

Money supply isn’t shrinking. M2 stands at $23.2T in the July 2026 report, growing 5.4% a year, a record high. May’s 5.6% year-over-year growth was the fastest since July 2022. The contraction was 2022–23; that’s over.

Inflation is elevated, not soaring. CPI rose 0.4% in August, 3.4% year-over-year, with core at 2.4%. Gasoline rose 3.9% and accounted for more than a third of the monthly gain — that’s energy/Iran-war driven, not a broad wage-price spiral. Pipeline pressure is real though: PPI is at 5.4% annually.

Labor market is holding. Payrolls rose 162,000 in August, well ahead of the 53,000 consensus, with unemployment at 4.1%. The 2026 layoff pace is the slowest in four years. Caveat: the prior 12-month average was only 31,000/month, so it’s stable, not strong.

The policy risk is the Fed, not fiscal. After holding all year, the Fed is expected to hike a quarter point next week, and Hammack has said multiple hikes may be needed. Tightening into a 31k/month job trend is where your recession scenario gets its best odds.

Rough odds, my read, not a forecast:

  • Inflation stays 3–4% through mid-2027: ~55%
  • Recession within 12 months: ~30–35% (Fed overtightening + tariffs + energy shock is the path)
  • Outright deflation: <10%. Deflation needs monetary contraction or a credit collapse. With M2 growing 5%+, mild disinflation is the realistic downside, not falling prices.

The full chain you describe (contraction → mass layoffs → collapse → deflation) requires the first link, and it isn’t there. Not financial advice — I’m not an advisor.

Excuse me but the treasury is buying the long end of the curve and the FED is raising rate, or more over the yields are rising. Inflation is about to get a lot worse because the contracts for shipping had not ended and diesel fuel was not a factor, but now is going forward for inflation worries. You do not know much. Or you just are wrong. Workers are hurt in a big way. Businesses that are weak are hurt in a big way. You have not figured it out. Think again.

You’re right on diesel and I underweighted it. Correction, then where I still disagree.

What you have right:

  • Diesel hit $6.05, up from $5.85 last week and about $3.70 pre-war. Diesel supply remains at just above 25% of pre-war levels per the IEA — that’s a physical shortage, not a price spike that mean-reverts.
  • The PPI showed diesel costs up more than 24% over the month and trucking up more than 14% year-over-year. August CPI barely captured this. Contract repricing is the lag you’re describing — Q4 CPI prints will be worse.
  • The 30-year hit a 19-year high; Treasury doubled long-end buybacks to $4B+ per operation through Nov 4. Yields kept rising anyway — the 30-year climbed above 5.3% after the $6B operation. Treasury buying long, issuing short, while the Fed hikes the short end is fiscal-monetary conflict.
  • Deficit around 6.3% of GDP, 10-year up nearly 70 bps since the war started, mortgages at 6.75%. Rate-sensitive sectors and thin-margin businesses are getting squeezed now, not later.

Where I hold:

Money supply is not being decreased. M2 growing 5%+ at a record high is a fact. What’s tightening is financial conditions — real yields, credit spreads, refi costs — and that’s a different mechanism. It matters for your deflation call: an energy-supply shock plus tight credit produces stagflation, not deflation. Deflation needs demand collapse severe enough to overwhelm a 5% money growth and a $6 diesel cost push. Possible, but it’s the third act, not the second.

Revised odds:

  • Headline CPI above 4% by Q1 2027: ~55%, up from what I implied before
  • Recession within 12 months: ~40–45%
  • Fed hikes into it and worsens the downturn: likely if they hike next week
  • Deflation: still ~10–15%. Stagflation is the modal outcome.

The 4.1% unemployment and 162k print are real, but they’re lagging. If you want to track the turn, watch weekly claims and the U-6, not payrolls.

U-6 is not showing the turn yet — it’s moving the wrong way for your thesis.

Current reading: 7.7% for August, down from 7.9% in July. That’s the lowest level since June 2025. The drop came from involuntary part-time falling by 414,000 to 4.4 million.

2026 path: 8.0% Jan → 7.9% Feb → 8.0% Mar → 8.2% Apr → 8.1% May → 7.9% Jun → 7.9% Jul → 7.7% Aug. The spring bump to 8.2% (April, as tariffs and the war hit) has unwound. Peak was 8.7% in November 2025.

How to read it:

  1. U-6 minus U-3 is the stress gap. Right now 7.7 − 4.1 = 3.6 points. When unemployment rises, the gap between U-3 and U-6 tends to widen — at the 2010 peak it was 7.4 points. 3.6 is a normal-cycle number, not a pre-recession one.
  2. Pre-COVID baseline was 6.5% in late 2019. So 7.7% is a full point above the last “good” economy. Underemployment is elevated, just not accelerating.
  3. The caveat that supports you: participation is 61.6%, down 0.5 points since January. People leaving the labor force entirely don’t show up in U-6 either. Some of the improvement is people giving up, not getting hired.

What would confirm your view: U-6 back above 8% with U-3 flat. That pattern — hours cut, full-time converted to part-time, before layoffs — is how a diesel/margin squeeze shows up first. September data comes October 2.

Right now the household data says businesses are absorbing costs, not cutting labor yet. That’s the lag. It doesn’t mean the cut isn’t coming; it means it hasn’t started.

My comment a friend who follows commodities is saying the shipping contracts locked in diesel, but that is about to end. Diesel will drive a lot of inflation going forward.

My expectations of deflation come with a collapse.

Is that a robotic puppy?

DB2

If you roughly agree with my opinion that the report is worthless, why post it at all? Or why not add some commentary on it to express an opinion?

Links to articles without comment are among the most worthless of posts on discussion boards.

—Peter

6 Likes

I don’t think it is worthless. For example, it shows the wide range of outcomes with difference starting assumptions. It shows how people can have widely differing views on the future impacts without “blowing smoke”.

I disagree. Information by itself can be quite useful. They can start a discussion such as the dozen or so posts in this thread or provide follow-up info. Also, commentary can influence how the article is approached much as polling results can be influenced by how the questions are asked.

DB2

1 Like

Yes.

2 Likes

Isn’t that already a well established point? Why do we need yet another example of starting assumptions basically determining the outcome?

Ok. But what is the point of this article? What discussion are you trying to start?

When I’m talking about commentary, I’m including the asking of questions.

For example, an uncritical surface reading of your summary might indicate that AI is a good thing under all of the assumptions. After all, every scenario shows an increase in GDP caused by AI. But as soon as you look at the author of the report - an AI company - you realize that there is a strong self-interest in showing AI in a good light. So strong an interest, that it calls the whole report into question.

Circling back, I have to question whether this really IS information, or just self serving propaganda.

Personally, I’m not educated enough in AI to tell the difference. So all I can do it question the report and hope someone better educated than I can weigh in on that topic.

In the mean time, that’s such an important question that it [the report] is not actionable.

—Peter

6 Likes