Tag Archives: artificial intelligence

AI investment compared to railway boom

The blog Urbanomics has a comparison of the current AI investment concentration to the 19th century railroad investment boom in England and the United States. In this particular case, the blogger neglected to provide the original source, which he or she normally does. Financial Times and Economist are typical sources. Anyway, here are some stats mentioned:

  • Peak “railway mania” in the UK was around the 1840s, and railroad investment accounted for around half of all investment at that time.
  • Between about 1830 and 1870 in the UK, railroad investment accounted for about 20% of all investment.
  • In the US, episodic railroad investment booms occurred in the 1840s and 1870s. Railroad investment at these times was around 40% of all investment. This accounted for GDP growth of about 6-10%.
  • The brief clip actually doesn’t tell us how much of total US investment in 2025 is directed to AI. But it accounts for GDP growth of around 2%.

These are interesting numbers, but I don’t think comparing 19th century and 21st century US GDP growth is a very good comparison. That is essentially comparing a fast-growing developing country to a slow-growing advanced economy. If I had to pick one or the other to live in, I would probably go with the one that has safe drinking water, antibiotics, vaccinations, relatively painless dentistry, and air conditioning.

what’s next for (incremental improvement of commercial) AI

We normals are hearing in the media that the large language model approach to AI has run its course, that further scaling it up is prohibitive in terms of energy, and that there is an AI-hype-driven financial bubble ready to pop any moment. According to at least one blogger though, the big breakthrough happening right now is having these models “reason” internally before they give an answer.

Two of those leading engineers are: Julian Schrittwieser who helped teach AlphaGo how to play Go at a level never witnessed in human history and is now a lead researcher at Anthropic. And Łukasz Kaiser, who whilst at Google Brain, co‑authored the paper that launched the architecture now driving every major released model on “Attention is all you need”…

Kaiser, for his part, corrects time horizons. The category of work that still belongs unquestioned to humans is shrinking. He states, with a deep belief, that these AI systems will be able to do any labor task currently performed on a computer within a timeframe of five years! …

The question is not whether machines will pass some imagined threshold in the future, but what it means that they have already crossed thresholds we still debate as hypothetical. A society reacts to what it believes is true, not to what is true. When the prevailing public understanding is delayed by years, institutions are, by definition, operating in a prior decade.

We can model technological progress as a series of sequential, overlaid S-curves that have to overlap in just such a way to produce continuous exponential growth. At least some insiders are still thinking in terms of keeping this S-curve going, in a competition between companies and countries. And when we see a new technology break through into widespread public, commercial use, it has already been going in the lab for awhile. That used to be measured in decades, now it is months if these optimist insider voices are to be believed.

https://onepercentrule.substack.com/p/is-ai-on-a-new-trajectory

AI-mediated transportation asset management

This article is called “Cities and states are turning to AI to improve road safety“. Basically the concept is to pay private vehicle owners to install dashboard cameras which take video of street conditions and feed it into a central database. What makes it “AI” seems to be computer-assisted analysis of the videos.

This all makes sense to me, although I wonder if you just put this technology on all the public fleet vehicles out there (buses, police cars, fire trucks, public works vehicles, maybe partner with utility companies) if that would be enough.

I do like the idea of focusing more on the infrastructure itself when it comes to safety, rather than vehicles and their drivers which is essentially blaming the victim. With gradual advent of autonomous vehicles, I see a shift in attitudes towards zero tolerance of deaths and injuries. Early on, my thought was that this was unfair because human-controlled vehicles cause so many deaths and injuries and we tend to think of these as inevitable. But as I have thought about it more, the public has essentially zero tolerance for deaths and injuries on any form of public transportation, whether trains, buses, or planes. It is time we held motor vehicles and the infrastructure they are traveling on to this same standard, and the trend seems to be in that direction.

The other positive trend here is a core principle of asset management itself. We all know infrastructure is expensive and difficult to build and maintain, but it does wear out and need to be repaired and eventually replaced. Each time you do a repair or a replacement, you have a chance to upgrade at low or sometimes no extra cost. Any single piece of infrastructure lasts a long time, but there are always things wearing out here and there throughout the system. So if you have a solid vision of where you want to go and you make those repair/replace/upgrade decisions consistent with it, small changes can add up to big system change over time, and this can be done cost-effectively. We don’t need “AI” to do this necessarily, but if calling it AI helps us get over the psychological hurdle to actually make it happen, let’s go for it!

going to college is still a lot better than not going to college

I hear people “questioning the value of a college degree” in the media. Sure, education is getting more and more expensive at a time when wages seem to be stagnating and there is some uncertainty whether career prospects for today’s graduates will be similar to those of past generations. But the numbers say (paying to study and not work for four years and) getting a degree is still a much better investment than not getting a degree and going right to work after high school. Sure, you could borrow the cost of four years of college and bet it on cryptocurrency or the Super Bowl, and you might come out ahead, but you might also come out living a short life under a bridge somewhere. You could also train as, say, an electrician and probably have a decent income and successful career, but you would still probably do better in the long run as an electrical engineer.

Anyway, this is from the Financial Times, which I still seem to have residual access to from my own recent student career.

To determine whether recent graduates are having an especially tough time in 2025’s low-hiring environment, the comparison we should make instead is with others who recently entered the labour market for the first time, regardless of age. A newly job-seeking graduate might be in their mid-twenties, but someone entering the world of work straight from high school will be several years younger.

Once we do this, it turns out that those without a degree are actually having a much harder time of it. In the US, unemployment among recent college graduates is up 1.3 percentage points from its mid-2022 low, but by almost double that among recent labour market entrants without a degree, who have seen a 2.4 point rise. This is very different to the much more modest 0.7 point rise among the frequently — but inappropriately — cited group of non-grads in their mid-twenties who are sheltered from today’s harsh hiring conditions.

But evidence for the kind of large-scale AI-driven displacement of early-career knowledge-sector jobs that would explain broad-based graduate malaise remains conspicuous by its absence….When viewed instead as a broader cooling of the labour market, in which inexperienced workers of all stripes bear the brunt (and especially those with the least skills) we don’t need to reach for such exotic explanations. The unwinding of extremely tight post-pandemic labour markets, rising input costs from inflation, tax changes and tariffs, plus the broader economic uncertainty during Donald Trump’s second term, are sufficient to explain what we’re seeing.

AI-related changes to the job market and wider economy are almost certainly coming, in my view, but we may be perceiving a causation between today’s technology and economic/political headlines that is not quite happening in real time.

Is the AI bubble bursting?

Apparently trying to answer this question is consuming a lot of bandwidth in the financial, tech, and even geopolitical arenas right now. Here is one answer from Larry Johnson, whose politics and past statements I do not necessarily endorse. Just to very briefly summarize his article: YES.

A few insights of my own:

  • The AI “hype bubble” has almost certainly reached a commanding height, and will pop at some point. This will probably be felt in stock market index valuations, which are dominated by a handful of large tech companies at the moment. In my lifetime now covering half a century, we have seen this cycle first with the personal computer itself and then with the internet. In both cases, the expectation that these technologies would super-charge economic growth in a few years did not happen, and led to financial market declines. Both technologies have in fact transformed the economy drastically, it just took a few decades rather than years. Things do seem to be happening faster this time around, I admit.
  • When it comes to stock market crashes, there is usually some precipitating event like the Asian financial crisis in 1997 or U.S. derivative bubble in 2007. The combination of technology bubble bursting and external financial shock seems to be particularly powerful. In fact, when I look back, I think I can argue the forward progress of the U.S. halted around that 1997 (financial crisis) to 2000 (Bush v. Gore) to 2001 to 2003 (9/11 attacks and Iraq invasion) period, and went into outright decline between the 2007 financial crisis and 2020 Covid crisis.
  • Apparently some in Silicon Valley thought the artificial general intelligence singularity was so near when the LLMs first came out, and that US tech companies were so far ahead of international peers, that it justified huge short-term investments in order to gain a first mover advantage that would then be insurmountable. This particular bubble seems to be popping at the moment, with AGI clearly not here right now, and perhaps a loose, emerging consensus that LLMs are a useful technology but not a likely path to AGI. So companies may have over-invested in infrastructure that will hurt some of them badly in the short term, while possibly benefitting us all in the longer term (think about 19th century railroads for a fairly obvious analogy).

So there is somewhat of a race here – will we start to see significant economic benefits of these new technologies before some external shocker hits us? This is the luck of the draw. It seems luck has not been on our side for the last 25 years or so. Perhaps we’re due.

mayors, governors, and senators

In a random AI experiment (Microsoft’s Copilot in this case), I have generated a list of 2028 US presidential candidates. Here were my criteria:

  • Current or previous mayors of the largest 100 US cities, re-elected at least once. Alive and under 70.
  • Current or previous governors of US states, re-elected at least once. Alive and under 70.
  • Current or previous US Senators, re-elected at least once. Alive and under 70.

No, Donald Trump would never have passed this screen, and nor does J.D. Vance because he has not been re-elected to any office so far. But my reasoning is these are people who showed they have what it takes to win high-stakes elections, then perform well enough in the eyes of voters and donors to get re-elected. Sorry to the 70 and up crowd, but for the Democrats in particular it is just time for the older generation to turn over the reigns.

A few familiar names: One person who is familiar, Barrack Obama, would not be eligible. People who have run before and not done all that well (Pete Buttigieg, Cory Booker, Tim Walz, Nikki Hailey, Chris Christie, Bobby Jindal, I’m looking squarely at all of you) should step aside and give others a shot. Marco Rubio and Rahm Emanuel are a couple household names that jump out at me from this list. Does anyone on this list actually excite me? Michael Nutter, best ever (and only really good) mayor of Philadelphia, your country needs you!

I also asked Copilot to help me encode the table as HTML, which it was able to do. There are undoubtedly better ways to add tables in WordPress, which maybe I will be smart enough to learn about some day. So without further ado, here is the list sorted from youngest to my mandatory retirement age of 69:

Name City/State Party Years in Office Estimated Age Role
Quinton LucasKansas City, MODemocrat2019–present40Mayor
Kate GallegoPhoenix, AZDemocrat2019–present43Mayor
Jacob FreyMinneapolis, MNDemocrat2018–present43Mayor
Pete ButtigiegSouth Bend, INDemocrat2012–202043Mayor
David HoltOklahoma City, OKRepublican2018–present46Mayor
Todd GloriaSan Diego, CADemocrat2020–present47Mayor
Tim KellerAlbuquerque, NMDemocrat2017–present47Mayor
Andy BeshearKentuckyDemocrat2019–present47Governor
Eric JohnsonDallas, TXRepublican2019–present48Mayor
Tom CottonArkansasRepublican2015–present48Senator
Regina RomeroTucson, AZDemocrat2019–present49Mayor
Andrew GintherColumbus, OHDemocrat2016–present50Mayor
Jared PolisColoradoDemocrat2019–present50Governor
Cory GardnerColoradoRepublican2015–202150Senator
Julian CastroSan Antonio, TXDemocrat2009–201451Mayor
Chris MurphyConnecticutDemocrat2013–present51Senator
Muriel BowserWashington, D.C.Democrat2015–present52Mayor
Kevin StittOklahomaRepublican2019–present52Governor
Brian SchatzHawaiiDemocrat2012–present52Senator
Alex PadillaCaliforniaDemocrat2021–present52Senator
Gretchen WhitmerMichiganDemocrat2019–present53Governor
Nikki HaleySouth CarolinaRepublican2011–201753Governor
Ben SasseNebraskaRepublican2015–202353Senator
Bobby JindalLouisianaRepublican2008–201654Governor
Marco RubioFloridaRepublican2011–present54Senator
Kasim ReedAtlanta, GADemocrat2010–201855Mayor
Cory BookerNewark, NJDemocrat2006–201356Mayor, Senator
Gavin NewsomSan Francisco, CADemocrat2004–201157Mayor, Governor
Scott WalkerWisconsinRepublican2011–201957Governor
Tammy DuckworthIllinoisDemocrat2017–present57Senator
Kelly AyotteNew HampshireRepublican2011–201757Senator
Michael BennetColoradoDemocrat2009–present60Senator
Brian KempGeorgiaRepublican2019–present61Governor
Tim WalzMinnesotaDemocrat2019–present61Governor
Chris CoonsDelawareDemocrat2010–present61Senator
Martin O’MalleyBaltimore, MDDemocrat1999–200762Mayor, Governor
Chris ChristieNew JerseyRepublican2010–201862Governor
Jeff FlakeArizonaRepublican2013–201962Senator
Barack ObamaIllinoisDemocrat2005–200863Senator
Mark BegichAlaskaDemocrat2009–201563Senator
Jane CastorTampa, FLDemocrat2019–present64Mayor
Rahm EmanuelChicago, ILDemocrat2011–201965Mayor
Mitch LandrieuNew Orleans, LADemocrat2010–201865Mayor
Kim ReynoldsIowaRepublican2017–present65Governor
Michelle Lujan GrishamNew MexicoDemocrat2019–present65Governor
Dean HellerNevadaRepublican2011–201965Senator
Mike DugganDetroit, MIIndependent2014–present66Mayor
Phil MurphyNew JerseyDemocrat2018–present67Governor
Andrew CuomoNew YorkDemocrat2011–202167Governor
Joe HogsettIndianapolis, INDemocrat2016–present68Mayor
Michael NutterPhiladelphia, PADemocrat2008–201668Mayor
Deval PatrickMassachusettsDemocrat2007–201568Governor
Terry McAuliffeVirginiaDemocrat2014–201868Governor
Jon TesterMontanaDemocrat2007–present68Senator
Heidi HeitkampNorth DakotaDemocrat2013–201969Senator
Joe DonnellyIndianaDemocrat2013–201969Senator

Is AI speeding up computer programming efficiency?

Yes, by about 25% according to a serious look at the hard evidence by some heavy-weight academics (MIT, etc.)

The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers

This study evaluates the impact of generative AI on software developer productivity via randomized controlled trials at Microsoft, Accenture, and an anonymous Fortune 100 company. These field experiments, run by the companies as part of their ordinary course of business, provided a random subset of developers with access to an AI-based coding assistant suggesting intelligent code completions. Though each experiment is noisy, when data is combined across three experiments and 4,867 developers, our analysis reveals a 26.08% increase (SE: 10.3%) in completed tasks among developers using the AI tool. Notably, less experienced developers had higher adoption rates and greater productivity gains.

“Intelligent code completions” kind of matches my own experience with how I have found AI most helpful so far – as software help. Whether it is helping with obscure code syntax or complicated nests of drop-down menus and check boxes, AI makes it much faster to find the exact thing you are looking for. This should in theory give workers a bit more time for planning and creative thinking, but predictably the market wants us not to do our jobs better, but to do them barely adequately as fast as possible. And what passes for “barely adequately” erodes over time while “as fast as possible” gets faster. Which I suppose is efficiency on paper.

One question is whether this is more like the automated loom, which sharply reduced the demand for textile workers, or the cotton gin, which sharply increased the demand for (involuntary, brutalized) workers by removing a bottleneck in the process. Early signs seem to point to the former, but all this will take time to play out.

offloading thinking to AI

It’s disturbing if professionals and students are trying to use AI to avoid hard thinking, as this duo of articles suggests. Ideally, at least in the near to medium term, we need to be doing the opposite. Using AI to perform mundane, repetitive, or just plain frustrating tasks that take up a lot of our time but don’t require deep thinking. Figure out coding syntax is an example, or which of the 99 drop down windows and dialog boxes in Microsoft Word will fix the frustrating formatting problem. (Actually, these last two things are kind of the same as you think about it, just two different ways of accessing a complicated menu of options and trying to communicate with a computer in its version of logic.) If AI can free us from these time wasters, we can have more time for deep thinking and creative thinking. I’m not saying this is the general trend, but this is my personal goal for how I am using AI. For now, I want it to help me do something I could have done myself faster or better. Asking it to think for me would be like asking another person to eat, exercise, or poop for me – I won’t gain any benefits from that.

I’ve been trying to use CoPilot to help me debug a simple stock and flow model. It can’t. It gives me sophisticated-sounding answers that do not even come close to working in the software I am playing with (Vensim PLE in this case).

what’s new with humanoid robots?

ChatGPT, Co-Pilot, and Claude may have helped me speed up some computer-based work tasks over the past year or so, but they still can’t clean my bathroom. This is what I really want to see to believe the robot era has really arrived. This article reads like a press release, but Nvidia says it has an “open-source, pretrained but customizable foundation model that’s designed to expedite the development and capabilities of humanoid robots”. So that’s something, but I am not hopeful it will be cleaning my bathroom or doing my dishes in 2025.

January 2025 in Review

Well, January was a doozy. Here goes:

Most frightening and/or depressing story: Longreads #1 stories of 2024 – this is a lookback but I posted it in January and it has a ton of interesting stuff. Interesting, frightening, and depressing. The story on Israel’s dispatching of air strikes based on statistical analysis is the single most disturbing article I read last year. Everyone should read this article and decide for yourselves where you stand. Another one is called “When the Arctic Melts”. Even as the shadow of fossil fuel propaganda once again overspreads the land, I am afraid the globe could be approaching an irreversible tipping point into runaway warming and sea level rise. Let’s hope the world can afford another four-year round of U.S. backsliding and then pick up the pieces, but I am not sure.

Most hopeful story: I noted that congestion pricing in New York City could provide a glimmer of hope that transportation in the United States could begin to implement 21st century international best practices. (Yes, I am aware the century is a quarter over already – one more indicator of the U.S. slipping towards the bottom of the world’s more advanced nations.) Unfortunately, as I write this on February 13 we see the President himself actively interfering in this state and local matter. “States’ rights” for thee, not for me (i.e. only when it’s convenient to some disingenuous argument).

Most interesting story, that was not particularly frightening or hopeful, or perhaps was a mixture of both: AI agents – coming soon to a computer near you.