YouTube2Text

Keynote Remarks by Mary C. Daly at SIEPR’s Policy Forum — Transcript

by Federal Reserve Bank of San Francisco · 10,393 words · 1,549 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Perfect. I'm going to going to show some
  2. 0:01slides. I can prove that I'm an
  3. 0:03economist when I show slides.
  4. 0:06Uh, so thank you Neil for that warm
  5. 0:08introduction and thank it's really
  6. 0:10lovely to be back at CPER. It's one of
  7. 0:11my favorite places to speak. It's right
  8. 0:13here uh close by San Francisco Fed, but
  9. 0:16also the home of so many bright and
  10. 0:18young people in particular and I always
  11. 0:20enjoy coming to the Stanford campus. So
  12. 0:23today I want to talk about AI in the
  13. 0:25economy and I want to focus on the
  14. 0:27expectations which Neil gave us a little
  15. 0:29preview of, the reality and I mean the
  16. 0:32realities today and maybe the realities
  17. 0:33tomorrow which could be different and
  18. 0:35how monetary policy makers in particular
  19. 0:37have to think through this because it's
  20. 0:39not you know something that we can just
  21. 0:41imagine will be great. We have to
  22. 0:43actually investigate what's happening
  23. 0:44today. So let me start with the promises
  24. 0:48of AI, the AI expectations. So this is a
  25. 0:51simple chart of or graph of labor
  26. 0:53productivity growth, 10-year moving
  27. 0:55average over the last you know century
  28. 0:58or more and it really shows that
  29. 1:01electrification was a big boom to
  30. 1:04productivity growth for a long time. The
  31. 1:06internet was a also a durable boom, but
  32. 1:09not quite as large as electrification
  33. 1:11and if you listen to most of the
  34. 1:12enthusiasts you know even three or four
  35. 1:14years ago, they were saying AI's impact
  36. 1:17is going to supersede or be greater than
  37. 1:19electrification and right now it doesn't
  38. 1:22look quite there. So the question is
  39. 1:24where are we going and Neil put this
  40. 1:26well. We all expect this to be terrific
  41. 1:30and transformative and disruptive and
  42. 1:32all the things in that category, but we
  43. 1:34don't know exactly when this will
  44. 1:35happen. So what I want to do is bring
  45. 1:38that vast question into a narrow window
  46. 1:41where monetary policy must look. So let
  47. 1:43me start with how we think about this,
  48. 1:46how I think about it as a monetary
  49. 1:47policy maker, also as an economist I
  50. 1:49think is that we have to focus on
  51. 1:51economic growth. There's four lenses,
  52. 1:53economic growth, productivity,
  53. 1:56inflation and the labor market and of
  54. 1:58course inflation and labor market's
  55. 2:00important for the economy, but it's also
  56. 2:02our dual mandate responsibilities at the
  57. 2:05Federal Reserve. So now I'm going to
  58. 2:06look through those four lenses and I'm
  59. 2:08going to think about what's happening
  60. 2:10today. What's What do we have so far? So
  61. 2:13I want to start with growth. These are
  62. 2:15I'm going to put two charts together
  63. 2:17here to to say one basic point. We have
  64. 2:20a technology-driven growth boom.
  65. 2:23The left-hand side of the charts shows
  66. 2:25real GDP growth, four-quarter change,
  67. 2:28and it shows that the blue line that we
  68. 2:30are growing 2.1% over that 12-month
  69. 2:33change. That is a really good growth
  70. 2:35rate right now, but if you took
  71. 2:36technology goods out of that equation,
  72. 2:39we'd be growing at 0.6%. So when you
  73. 2:41hear people say this is a technology-led
  74. 2:43boom, they're absolutely right. These
  75. 2:45are the the points for it. So then
  76. 2:47people get very nervous, and of course,
  77. 2:49we all want to think about this, about
  78. 2:51how long can this last? What really is
  79. 2:54going on? And especially when there's
  80. 2:55been some
  81. 2:57market volatility on AI valuations, the
  82. 2:59natural question is, is this something
  83. 3:01we can expect to continue? So I There's
  84. 3:03many ways I could show this, but I like
  85. 3:05to show this data center chart. And the
  86. 3:07reason I like to show the data center
  87. 3:09chart is it shows that you're having a a
  88. 3:11general upward trajectory. We have
  89. 3:13things already in play. They're under
  90. 3:15construction. That's the green bars in
  91. 3:1726, and then we have lots of things
  92. 3:19planned in the pipeline. So even if we
  93. 3:21get a pullback of data center enthusiasm
  94. 3:24or communities say, wait a minute, let's
  95. 3:26slow this train down, you still see
  96. 3:28there's so much in the pipeline that you
  97. 3:30can expect this to continue.
  98. 3:32Now data centers themselves are not
  99. 3:34They're just large buildings. The thing
  100. 3:36that really fuels the AI boom is the
  101. 3:39materials and goods and software that
  102. 3:41goes in the data centers. And so this is
  103. 3:43a small starting point, but you can just
  104. 3:46multiply that and you get something that
  105. 3:48looks like the picture on the left. So
  106. 3:50ultimately, when we think about growth,
  107. 3:52there's not only the growth we see, but
  108. 3:54some general appreciation for this
  109. 3:57growth will continue even if it's not at
  110. 3:59the same pace.
  111. 4:01So then let's look at productivity.
  112. 4:03So, productivity evidence or
  113. 4:05productivity growth evidence is mixed at
  114. 4:07this point. Let me tell you what I mean.
  115. 4:10So, if you think about labor
  116. 4:12productivity in 23 24 25 and 26 which
  117. 4:15are the middle uh set of bars, well then
  118. 4:18you see that it's it's above where we
  119. 4:20were in the previous decade. The
  120. 4:23historical average is roughly rule of
  121. 4:25thumb 1.5 %. That's what we had in the
  122. 4:27previous decade, but if you look at what
  123. 4:29we've been experiencing, it's much
  124. 4:31faster than that. So, there's no doubt
  125. 4:33that labor productivity growth is is is
  126. 4:36quick right now and that's a good thing
  127. 4:37for our economy. It means the economy
  128. 4:39expands faster, people can do uh the
  129. 4:42output grows. Those are all good things.
  130. 4:45But if you really want to think about
  131. 4:46the impact of AI and whether to
  132. 4:48attribute the labor productivity growth
  133. 4:50we've been seeing to AI and whether the
  134. 4:52transformation has begun, you can't just
  135. 4:55look at the blue bars. You have to look
  136. 4:57at what economists call total factor
  137. 4:59productivity. I think what people call
  138. 5:01the magic behind productivity growth,
  139. 5:03right? It's the the part of productivity
  140. 5:05growth that has that sustainable durable
  141. 5:08part. It's the part about
  142. 5:09electrification that it wasn't just you
  143. 5:12could turn on a light. It's that it
  144. 5:13transformed everything we knew about how
  145. 5:15to live and work in the US. So, that's
  146. 5:18the piece that has been less than
  147. 5:20positive. If you look at TFP, it was
  148. 5:22really strong for a little bit in 23 and
  149. 5:2524, but now it's it's at average or a
  150. 5:27little below average and so we haven't
  151. 5:30seen the magic there yet that is what
  152. 5:33the AI tools promise. That's quite
  153. 5:36consistent with the survey of of firms
  154. 5:39on the right. This is a survey by Nick
  155. 5:41Bloom and his team and it just basically
  156. 5:43shows that it over the past 3 years the
  157. 5:46firms that they survey haven't really
  158. 5:49said that they had an impact of AI on
  159. 5:51their productivity growth. There's some
  160. 5:53of course that have had a positive
  161. 5:54impact. If you go to certain industries,
  162. 5:56you'll see a positive impact. You can
  163. 5:57certainly find case studies. You can
  164. 5:59certainly find evidence of green shoots
  165. 6:02forming, but overall in the whole
  166. 6:04economy just haven't seen.
  167. 6:06Right now the big impact that many
  168. 6:09expect, including me.
  169. 6:11So now let's look at the the next thing
  170. 6:13about price pressures. I said I was
  171. 6:14going to look through the lens of the
  172. 6:15dual mandate. Let's start with
  173. 6:17inflation.
  174. 6:18There is no doubt that AI is fueling
  175. 6:21price pressures. You can see this in the
  176. 6:23left hand chart. I think that the left
  177. 6:25hand chart is particularly remarkable
  178. 6:27because it shows that after you know two
  179. 6:30decades, three decades of disinflation,
  180. 6:33deflation in technology goods, we now
  181. 6:36have 15.7% almost 16% inflation in these
  182. 6:40goods. And that is a huge reversal of
  183. 6:44price changes for a an important part of
  184. 6:46our economy. And it really is reflecting
  185. 6:49the demand for those goods relative to
  186. 6:51the supply.
  187. 6:53So then you want to ask if you're a
  188. 6:54policy maker, I think or all of us, how
  189. 6:57much of this relative price shock, this
  190. 6:59change in the relative price of
  191. 7:01technology goods is going to spill over
  192. 7:03widely into the economy. Well, I'm going
  193. 7:05to give you two ways to think about. The
  194. 7:07first one is in this chart. One is how
  195. 7:09much is it directly contributing to
  196. 7:12overall headline inflation. And you can
  197. 7:14see a modest amount, 0.2 percentage
  198. 7:17point onto the total inflation. That
  199. 7:20just tells you that right now headline
  200. 7:22inflation is being driven by other
  201. 7:24variables, not the AI goods.
  202. 7:26But we're also starting to hear and I'm
  203. 7:28sure you are and you probably
  204. 7:30experienced this if you bought a
  205. 7:31computer or a phone etc. The price of
  206. 7:33technology equipment is not just
  207. 7:35narrowly focused on those going up for
  208. 7:37data centers. It's going up everywhere.
  209. 7:40And so certainly firms are feeling that
  210. 7:42cost. The question is is are those
  211. 7:44indirect costs rising so sufficiently
  212. 7:47that we're going to see it in overall
  213. 7:48inflation? And so far we're not hearing
  214. 7:50that.
  215. 7:51Really hearing it's it's more directly
  216. 7:53associated with
  217. 7:55just that sector. Let me get my water
  218. 7:57just in case I need it.
  219. 8:00Okay. So then let's go to the labor
  220. 8:01market, which I think is really an
  221. 8:03important question that people have been
  222. 8:05asking.
  223. 8:06And so far we've seen a I would call a
  224. 8:08limited impact on the labor market. Now,
  225. 8:10that is not to say that there aren't
  226. 8:12impacted individuals.
  227. 8:14And we'd be uh remiss if we didn't say
  228. 8:17that there's certainly individuals who
  229. 8:19are experiencing that. So that is the
  230. 8:21case. But when you look through the
  231. 8:23aggregate economy, we are seeing a
  232. 8:25limited impact of AI at scale. You can
  233. 8:28see this in two ways. I like to show
  234. 8:30this hiring rate and separation rate,
  235. 8:32layoffs and and and quits. You just see
  236. 8:35that right now we're in a low hiring,
  237. 8:37low firing environment. You've probably
  238. 8:39heard that that phrase before. Low
  239. 8:41hiring, low firing. Firms aren't hiring
  240. 8:43a lot of people, but they're also not
  241. 8:45letting go of a lot of people. That
  242. 8:47doesn't mean certain firms aren't laying
  243. 8:48other off, but it does mean that there's
  244. 8:50a little bit of hiring, there's a little
  245. 8:51bit of separation, but mostly you're
  246. 8:54just in a low hiring, low firing
  247. 8:55environment.
  248. 8:57That means that if you look to the
  249. 8:58right-hand side of the chart, that the
  250. 9:00unemployment rate has remained really
  251. 9:02stable and at a level that's consistent
  252. 9:04with the natural rate of of unemployment
  253. 9:06that we expect, the one that just
  254. 9:09incorporates churning and other things,
  255. 9:10job search, people taking time to get
  256. 9:12from one place to another. But right
  257. 9:14now, this is labor market that's fairly
  258. 9:16stable and you haven't seen a big impact
  259. 9:19of AI on the labor market.
  260. 9:22So when I put all of this together, the
  261. 9:24the the point for monetary policy is
  262. 9:27we've got an inflation problem.
  263. 9:29Uh we have a labor market that's stable,
  264. 9:31growth that's good, but we definitely
  265. 9:33have an inflation problem. Right now, we
  266. 9:35don't have AI doing most of the driving
  267. 9:38of inflation, although there is a risk
  268. 9:39that it could contribute more. And we
  269. 9:42also don't have productivity growth in
  270. 9:44the TFP world, that magic part, in a way
  271. 9:47that's picking up steam. But you
  272. 9:48definitely see labor productivity
  273. 9:50helping us out.
  274. 9:52So then, what do I think about for the
  275. 9:54next several years that maybe the one to
  276. 9:57three uh impact? So now I'm going to
  277. 9:59look one to three years out, not that's
  278. 10:01the review of today, the health check on
  279. 10:03the economy. Now, what about one to
  280. 10:05three years out? What's possible? You
  281. 10:08can think of these as risks, you can
  282. 10:09think of these as scenarios, but
  283. 10:11basically I'm going to look through the
  284. 10:12four same four lenses, economic growth,
  285. 10:15productivity, inflation, labor market,
  286. 10:17and ask what's what's coming.
  287. 10:21So let's start with AI valuation
  288. 10:23movement. I think most of you read the
  289. 10:25news and you know that AI valuations
  290. 10:27have been moving around a lot. And I
  291. 10:30this chart on the left is really the
  292. 10:32meant to show that. The Magnificent
  293. 10:34Seven have dropped uh
  294. 10:37a lot
  295. 10:38relative to the S&P 493 and that's
  296. 10:42something that everybody's noticing. And
  297. 10:44there's a lot of articles out there
  298. 10:46about is this the the beginnings of the
  299. 10:48of the end, the unwinding of AI. And I
  300. 10:51wanted to put some perspective on it
  301. 10:53because I think of it
  302. 10:55I think context matters. So if you look
  303. 10:57next door on the right hand side of this
  304. 10:59chart, you see that yes, there has been
  305. 11:01some repricing, but it hasn't eroded the
  306. 11:05considerable gains that have been going
  307. 11:07on for the last three years. So yes,
  308. 11:10there's a repricing. No, it doesn't seem
  309. 11:13to be the unwinding of the technology,
  310. 11:16interest and enthusiasm and investment
  311. 11:19that we've been seeing.
  312. 11:20So how do I read this in terms of
  313. 11:22thinking about growth? I read this as
  314. 11:25the private sector, investors, all of
  315. 11:27the market participants, many of you in
  316. 11:29all likelihood, are actually asking good
  317. 11:32questions like what do we think this do
  318. 11:35we think the investment strategy that
  319. 11:37these companies are using, the borrowing
  320. 11:39strategy, etc., is going to be as
  321. 11:42risk-free as it seemed earlier on? Are
  322. 11:45we going to price in a little more risk?
  323. 11:47That is a good thing. I was on a podcast
  324. 11:49with Steve Davis and I said, you know,
  325. 11:51when the private he and I ended up
  326. 11:52talking about this, when the private
  327. 11:54sector responds, this creates more
  328. 11:56opportunity for there to be an orderly
  329. 11:59valuation to right-size
  330. 12:02uh returns with expectations. And so
  331. 12:04this is this is somewhat good news. But
  332. 12:07the second part of the story is that the
  333. 12:10underlying value
  334. 12:12isn't being eroded and ultimately that
  335. 12:15is consistent with there being some
  336. 12:16there there. You know, as Neil said,
  337. 12:18everybody expects this to be
  338. 12:20transformative. The question is how much
  339. 12:23and when? And you put those two things
  340. 12:25together and that single belief that
  341. 12:27this will be a transformative
  342. 12:28technology, it's useful to invest in,
  343. 12:31useful to think about, isn't going away.
  344. 12:33And that is contributing to some of the
  345. 12:36projections you see for growth and some
  346. 12:39of our own sense, my own sense, that
  347. 12:41growth will continue. There's not a risk
  348. 12:43to growth here.
  349. 12:46So what about um productivity?
  350. 12:49So if you stood there, you know, some
  351. 12:51people have been tempted to say, well,
  352. 12:52look, there's nothing happening yet. And
  353. 12:54I think yet is the important piece. We
  354. 12:57can't extrapolate from what's happened
  355. 12:58so far to talk about what's going to
  356. 13:01happen in the next 1 to 3 years. And one
  357. 13:03of the reasons is that the share of
  358. 13:05firms even using AI, just becoming
  359. 13:08familiar with it, employing it, thinking
  360. 13:11about it is growing. And I suspect that
  361. 13:14if firms fully understood what they were
  362. 13:16doing when they bought, you know, one of
  363. 13:17the big vendor programs and then AI is
  364. 13:19empowering a lot of this, you might even
  365. 13:21get a higher number. You know, people
  366. 13:22are turning on the tools, the AI tools
  367. 13:24in normal vendor programs they use for
  368. 13:27HR and payroll, etc., but in accounting,
  369. 13:30but they're also just getting instances
  370. 13:32of these models themselves and thinking
  371. 13:34about what they could do with it. So
  372. 13:35it's growing.
  373. 13:37Importantly, if we return to the survey
  374. 13:38that Nick Bloom runs, we see that if
  375. 13:41they look out, not past what happened in
  376. 13:43the last 3 years, but what do they
  377. 13:44expect to happen in the next 3 years,
  378. 13:46firms are much more enthusiastic about
  379. 13:49how it can have a positive impact on
  380. 13:51their um on their businesses. We had a
  381. 13:54round table
  382. 13:55recently of businesses. It was uh you
  383. 13:58know, just a panel of businesses, big
  384. 14:00businesses, global businesses. And we I
  385. 14:02asked them the speed question at the
  386. 14:04end, which is, do you expect
  387. 14:06productivity gains to be start to see
  388. 14:09transformation in 1 to 3 years or more
  389. 14:12like 5 to 10 years? And each person said
  390. 14:151 to
  391. 14:16because we've done all the hard work of
  392. 14:18investing, but now we're driving for
  393. 14:20ROI. And, you know, as one of the
  394. 14:22contacts said, when firms start driving
  395. 14:24for ROI, things happen fast. So I think
  396. 14:27this is a positive momentum for
  397. 14:29productivity growth. I can't say we're
  398. 14:30going to have it in 1 to 3 years, but I
  399. 14:33can say there's a lot of uh work being
  400. 14:35done to get it in 1 to 3 years.
  401. 14:40So
  402. 14:41you might worry that this inflation will
  403. 14:42never be bridled, that we just can't
  404. 14:44catch up fast enough and that this is
  405. 14:47something that the Federal Reserve and
  406. 14:49central banks across the globe are
  407. 14:50really going to have to deal with. And
  408. 14:52definitely we have to think about this
  409. 14:54as a risk and we have to deal with it.
  410. 14:55But the Federal Reserve, central banks
  411. 14:57are not the only game in town. One of
  412. 15:00the things you really rely on is the
  413. 15:01private sector of firms to say,
  414. 15:04wait a minute, maybe things are too
  415. 15:06costly. And so this chart short of
  416. 15:09reassuring in that way. So as new models
  417. 15:12became available, open weight, not just
  418. 15:14the closed weight models, as open weight
  419. 15:16models became more available, firms
  420. 15:18started using those. And this is a share
  421. 15:20of tokens processed
  422. 15:22on open router using open weight models
  423. 15:25versus closed weight models. And you can
  424. 15:27see that open weight models have gained
  425. 15:29a lot of ground. The share has increased
  426. 15:31quite a lot. They're much less
  427. 15:33expensive. And that's actually
  428. 15:36been correlated with the reduction in
  429. 15:38token prices for the OpenAI and
  430. 15:41Anthropic. So the private sector market
  431. 15:44seems to still be working. You have less
  432. 15:46demand for your service and you shift
  433. 15:48gears in terms of your pricing. Another
  434. 15:50thing we're hearing about is that firms
  435. 15:52are actively um think about memory and
  436. 15:55chips, chips, you know, regular chips
  437. 15:57and memory chips. Firms are paying
  438. 16:00higher prices, regular firms, not tech
  439. 16:02firms. They're paying higher prices as
  440. 16:04well for these chips. We're hearing more
  441. 16:06and more. The companies are saying, is
  442. 16:08it really true that the chip depreciates
  443. 16:10every 2 years or can I get more useful
  444. 16:12life out of this? Or maybe I can remodel
  445. 16:15my product line to use fewer chips or
  446. 16:17older chips and and not have to rely on
  447. 16:20this, you know, every two years I have
  448. 16:21to buy new and compete with the highest
  449. 16:23prices. So that's another what I would
  450. 16:25say competitive marketplace doing some
  451. 16:27of the work of disciplining this higher
  452. 16:29inflation. Doesn't mean that the Fed
  453. 16:31isn't can't think doesn't have to think
  454. 16:33about this and it's all going to work
  455. 16:35out beautifully, but I think an
  456. 16:36underappreciated point is that the
  457. 16:39private sector is responding to the ever
  458. 16:41rising price of these items and you're
  459. 16:43seeing prices start to respond.
  460. 16:48The final then thing I want to say
  461. 16:49before I turn to a little bit about what
  462. 16:50this means for policy is that
  463. 16:53the labor market is one of those places
  464. 16:55you can't extrapolate from either. You
  465. 16:57can't say nothing's happened so far, so
  466. 16:59nothing will happen. And I don't think
  467. 17:01we can also say that the worst will
  468. 17:03happen. So it's it's a very human
  469. 17:05characteristic to say to extrapolate
  470. 17:07from what's happening or what you fear
  471. 17:09is happening and say the worst will
  472. 17:10happen or the best will happen.
  473. 17:12Somewhere in the middle is what's likely
  474. 17:14to happen. Right now we see evidence
  475. 17:16that younger workers in particular, but
  476. 17:18mid-career workers as well are worried.
  477. 17:20This is a study out of Anthropic. They
  478. 17:23did a survey. You do see that workers
  479. 17:25are more worried at the younger part of
  480. 17:27the age distribution and earlier in
  481. 17:29career than at the more senior levels.
  482. 17:32You also see that unemployment rates for
  483. 17:34college recent college graduates are
  484. 17:36rising relative to the unemployment rate
  485. 17:38unemployment rate for all workers. And
  486. 17:40if I expanded this
  487. 17:43time frame, you'd see that this is very
  488. 17:44unusual. We haven't seen this in the
  489. 17:46United States for a long, long time. And
  490. 17:49it just speaks to the idea that, you
  491. 17:51know, the AI is changing the dynamic for
  492. 17:54younger workers. Now we spend a lot of
  493. 17:57time asking our our teams our our CEOs,
  494. 18:00our round tables, why is this? And I'll
  495. 18:03give you something that maybe we can we
  496. 18:05can think through during the course of
  497. 18:07the Q&A and that is basically maybe
  498. 18:09what's happening is they're waiting
  499. 18:11until they know who they want to hire.
  500. 18:13But if you're a younger worker, that
  501. 18:15doesn't feel very reassuring. So I think
  502. 18:17that is something we have to think
  503. 18:18about. So if now if I back up and ask,
  504. 18:20what does all this mean for policy?
  505. 18:23Uh, and I'll conclude my remarks there.
  506. 18:26So for monetary policy, we have to look
  507. 18:28through maybe really a one to three-year
  508. 18:30window. So right now, what we have, and
  509. 18:33I said this a moment ago, what we have
  510. 18:35is a stable labor market, really
  511. 18:37positive growth, some positive
  512. 18:39productivity growth for labor
  513. 18:40productivity, and an inflation rate
  514. 18:42that's been above target for six years.
  515. 18:44And we have to focus on price stability
  516. 18:47and bring that inflation down. It's not
  517. 18:49primarily driven by AI, but the AI price
  518. 18:53surges don't help our situation. So
  519. 18:55that's where the focus has to be.
  520. 18:57The second part of this though, what
  521. 18:59happens in the next one to three years
  522. 19:01just leads me to say this. Just because
  523. 19:03we have to fight the inflation problem
  524. 19:05now doesn't mean that we should
  525. 19:08completely ignore the possibilities on
  526. 19:10productivity growth, on how that could
  527. 19:12be disinflationary.
  528. 19:14Some people say the inflationary impulse
  529. 19:16will continue. That's a possibility, but
  530. 19:18I think it's also possible, and maybe
  531. 19:20it's because I came to the Fed in '96
  532. 19:22and I saw what computers and the
  533. 19:23internet did to the possibilities in the
  534. 19:26economy and how that added to the
  535. 19:28disinflationary
  536. 19:30impulse in the economy. We have to be
  537. 19:31open-minded. So planning out what
  538. 19:34policymakers will do for the next one to
  539. 19:36three years is a is kind of a fool's
  540. 19:38errand. I can tell you that what
  541. 19:39policymakers should be, what I think I
  542. 19:41will what I know I want to be focused
  543. 19:43on, what we all have to be focused on is
  544. 19:45what can productivity growth do for us?
  545. 19:47How will AI lead that? When do we start
  546. 19:49to see the transformation? And what is
  547. 19:52going to be the impact on the labor
  548. 19:53market, not 10 years from now, but in
  549. 19:56the next one to three years? So like
  550. 19:58others, I'm extremely
  551. 20:01bullish or enthusiastic about what this
  552. 20:02technology will mean eventually, but the
  553. 20:05thing that I've learned through studying
  554. 20:06all types of things, global trade, other
  555. 20:09technologies, all the way back to the
  556. 20:11industrial
  557. 20:12revolution, is that for policymakers the
  558. 20:14transition dynamic matters. And for
  559. 20:17humans living in the country, the
  560. 20:18transition dynamic happens. And that's
  561. 20:20where if we get 10 years from now and we
  562. 20:23find that
  563. 20:24it hasn't worked the way we wanted to, I
  564. 20:27will offer this. It won't be the
  565. 20:28technology's fault. It will be ours to
  566. 20:31the side. So intentional policy making,
  567. 20:33conversations like the one we have here
  568. 20:35going on today, these are what's
  569. 20:37essential to making sure we're proud of
  570. 20:3810 years from now as opposed to
  571. 20:40wondering why this happened.
  572. 20:42With that I'll hope that I'll be glad
  573. 20:44happy to talk to Neil.
  574. 20:45>> [applause]
  575. 20:50>> Amazing. I think this is uh a terrific
  576. 20:53presentation. Slides are really
  577. 20:55handsome. I I love the color scheme. I
  578. 20:57don't know if that's the the Fed's like
  579. 21:00uh
  580. 21:01schema.
  581. 21:02>> Francisco Fed's schema.
  582. 21:04>> It's it's great. Um
  583. 21:06>> We have a We have 12 reserve banks and a
  584. 21:07board of governors. Everybody has their
  585. 21:09own schema. That's ours.
  586. 21:10>> they're all equally good, but this one's
  587. 21:12the best. Um
  588. 21:13>> That's what I like to hear.
  589. 21:15>> So
  590. 21:16you set me up really well. Maybe I'm
  591. 21:18going to ask a couple questions which
  592. 21:20delve into the the topics you teed up.
  593. 21:24And then uh I have questions coming in
  594. 21:27uh on my iPad and so we can turn to
  595. 21:29those as well. And you can use the QR
  596. 21:32code to submit questions. Uh
  597. 21:36Let me
  598. 21:37Uh
  599. 21:40[snorts] So let's start with uh some
  600. 21:43questions you know that
  601. 21:46uh relate to the dual mandate and I
  602. 21:48think especially the inflation side,
  603. 21:50which is uh you know where the concerns
  604. 21:54are most heavy. So
  605. 21:57inflation inflation has been above
  606. 21:59target. You said six years?
  607. 22:02>> Almost six years.
  608. 22:03>> Almost six years. And you know it's
  609. 22:06it's probably worse than moving sideways
  610. 22:07now. It's it's moving in the wrong
  611. 22:09direction.
  612. 22:10Uh
  613. 22:11Our textbook
  614. 22:13sort of approach to monetary policy says
  615. 22:17we should look through a supply shock,
  616. 22:20right? And much of the inflation we've
  617. 22:23seen recently is supply shocks, tariffs,
  618. 22:27uh the war and energy prices.
  619. 22:29Uh
  620. 22:30Is that the way you think about the
  621. 22:33supply shocks we've been facing? That
  622. 22:35it's something that we should be looking
  623. 22:36through because a supply shock weighs on
  624. 22:39both sides of the mandate, right? It
  625. 22:41reduces growth and increases inflation.
  626. 22:43So if your job is to balance the two,
  627. 22:46then uh something that keeps these in
  628. 22:49balance uh is not something we should be
  629. 22:51over responding to. So is is that how
  630. 22:53you think about it?
  631. 22:54>> You know, I think that is the it's
  632. 22:55definitely the conventional wisdom as
  633. 22:57you said that that we would look through
  634. 22:59these types of shocks because it's also
  635. 23:01by the the idea that by the time our
  636. 23:04monetary policy tools would take effect,
  637. 23:06which is, you know, 12 to 18 months, the
  638. 23:08shock could be dissipated and then we
  639. 23:10would have overcorrected something that
  640. 23:12we didn't need to respond to at all or
  641. 23:15couldn't really affect. You can't affect
  642. 23:16the price of oil, but if you raise
  643. 23:18interest rates in response to the price
  644. 23:20of oil, this is the textbook example,
  645. 23:22you end up curbing an economy that's
  646. 23:23already struggling with lower growth.
  647. 23:25Okay, so that's that's the conventional
  648. 23:27wisdom.
  649. 23:28You know, the way I've been thinking
  650. 23:30about it since maybe the middle of the
  651. 23:32summer is a little bit different than
  652. 23:34that, which is that you know, we have
  653. 23:36inflation. We already start with an
  654. 23:37elevated level of inflation. It's
  655. 23:39already above target. Then we have three
  656. 23:41shocks in overlapping succession, right?
  657. 23:44So they're not just behind each other,
  658. 23:46they're actually overlapping. First we
  659. 23:48have tariffs, then we have the war and
  660. 23:50oil prices, energy prices, which is
  661. 23:52affected food prices, and then we have
  662. 23:54the AI shock. And I think it's a
  663. 23:56reasonable question for all all to ask
  664. 23:58and really investigate, do the fact that
  665. 24:00these shocks overlap and that they
  666. 24:02already come at a period of high
  667. 24:03inflation, do they compound on each
  668. 24:06other and start to change the underlying
  669. 24:08rate of inflation that really matters
  670. 24:11for the durability of inflation
  671. 24:12expectations, etc.? Our best check on
  672. 24:14that is inflation expectations. And at
  673. 24:18the medium and longer run, you haven't
  674. 24:20seen inflation expectations of markets,
  675. 24:22consumers, uh businesses drift. They're
  676. 24:24still consistent with 2% inflation, but
  677. 24:27we can't just be complacent and say well
  678. 24:30businesses and everybody's happy, so we
  679. 24:31shouldn't have to do anything. The way
  680. 24:33you keep inflation expectations well
  681. 24:35anchored is to bring inflation down to
  682. 24:37price stability. So for me, I was quite
  683. 24:40comfortable with the September rate
  684. 24:42increase. I think that's appropriate.
  685. 24:44And with, you know, my bias, as I've
  686. 24:46been saying, my bias is that further
  687. 24:48tightening will be needed because you
  688. 24:49just see this consistent pressure on
  689. 24:52inflation that isn't simply in the
  690. 24:55sectors that are most directly affected
  691. 24:57by tariffs or oil. They're actually
  692. 24:59starting to bleed into the services and
  693. 25:02that services inflation is once it gets
  694. 25:05up, it's extremely sticky. It has a hard
  695. 25:07time coming down. So that's how I think
  696. 25:09about it now and I don't think we can
  697. 25:11just look to the conventional wisdom and
  698. 25:13and hope for the best. We have to reckon
  699. 25:15with the realities we face, which is
  700. 25:17that inflation's been high for a while
  701. 25:19and the pressure is up, not down.
  702. 25:21>> Great. And And I think that basically
  703. 25:23covers my my follow-up, which was going
  704. 25:25to be, you know, how do we modify the
  705. 25:27textbook model when we have more
  706. 25:30frequent, more persistent supply side
  707. 25:33shocks? And your response was well we
  708. 25:35focus on inflation expectations
  709. 25:38of both the market-based measures and
  710. 25:41the survey-based measures. And you're
  711. 25:43also thinking about uh you know,
  712. 25:46services inflation ex housing as
  713. 25:48something that I think you're
  714. 25:50particularly concerned about or tracking
  715. 25:52closely in the current environment.
  716. 25:54>> If I may, we also ask, I mean, I think
  717. 25:56this is really important, it may be
  718. 25:57helpful. We also spend a lot of time
  719. 25:59asking businesses, what are you doing in
  720. 26:02the next year? And increasingly we're
  721. 26:03hearing businesses say I I I'm at the
  722. 26:06end of my my line. I'm going to have to
  723. 26:08pass some of these input costs through
  724. 26:10or they're going to eat into margins.
  725. 26:11Ultimately the disciplining device of
  726. 26:13margins is competitors who will say I'll
  727. 26:15take the margin hit and sell my goods
  728. 26:18and consumers who tell you whether
  729. 26:19they're going to buy your things or not.
  730. 26:21And you know, it seems like consumers
  731. 26:23should have run run out of gas a while
  732. 26:24ago, but they haven't. So you know,
  733. 26:26that's the dynamic that I think is
  734. 26:28really critical to watch as we plan for
  735. 26:30the, you know, think about policy
  736. 26:32response to the inflation impulse.
  737. 26:34>> No, that's and that's that's a great
  738. 26:36addition. So, let me
  739. 26:38talk about a another factor here, which
  740. 26:41is, you know, financial conditions. And
  741. 26:44I think for people who follow the bond
  742. 26:45markets, right, the 10 years at, I think
  743. 26:49it's over 5% this morning, in fact.
  744. 26:51>> morning.
  745. 26:51>> Um and, you know, then we there's also a
  746. 26:54term premium. So, I
  747. 26:57I guess the specific question is, you
  748. 26:59know, to what extent do you think the
  749. 27:01long end of the yield curve is doing
  750. 27:04sort of additional work for you in uh
  751. 27:08tightening financial
  752. 27:09conditions and providing a sort of
  753. 27:11headwind, appropriate headwind for the
  754. 27:13economy? How do you think about those
  755. 27:15factors?
  756. 27:15>> So, tighter financial conditions are
  757. 27:17most welcomed, I'd say, if you're
  758. 27:19thinking about bringing inflation down
  759. 27:20because we have that have tighter
  760. 27:22financial conditions to help with that
  761. 27:24that process.
  762. 27:25The the reason I hesitate on this
  763. 27:28question all the time is that I don't
  764. 27:30want to say markets do the work for the
  765. 27:31Fed because the Fed is given the
  766. 27:33responsibility. Congress gave the
  767. 27:35Federal Reserve the responsibility of
  768. 27:37restoring price stability, keeping
  769. 27:38prices low and stable, making sure we
  770. 27:41maintain full employment. And so that's
  771. 27:42not something we just acquiesce. But I
  772. 27:44do look at the market conditions and ask
  773. 27:46a very simple question. Are they out of
  774. 27:49alignment with where I think we need to
  775. 27:51be? And right now I see they're in
  776. 27:53alignment with where we need to be. And
  777. 27:55importantly, they're reacting to
  778. 27:56incoming information in a way that's
  779. 27:58consistent with the reaction function
  780. 28:01that I have. So, when data come in and
  781. 28:03say the pressure on inflation is up, you
  782. 28:05see market pricing for the Fed's uh
  783. 28:08policy response go up. And you also see
  784. 28:11uh yields go up. When you see oil prices
  785. 28:14rise, you know, you see these types of
  786. 28:15things. And that is a good sign that the
  787. 28:17economy, I said at the beginning talking
  788. 28:20about AI, when the private sector
  789. 28:22participates in their role, that's
  790. 28:24actually helpful. When consumers
  791. 28:26understand where we're going, that's
  792. 28:27helpful. So, I'm I'm I think that gives
  793. 28:30us a little more time to make really
  794. 28:33calibrated decisions. You know, we don't
  795. 28:35have a surgical tool, and so you want to
  796. 28:37be thoughtful about it. We don't also
  797. 28:39have a tool where you can go up and
  798. 28:40down, up and down. You know, it's hard
  799. 28:42on on businesses, consumers, etc. So
  800. 28:46what markets can adjust tomorrow? If the
  801. 28:48world changes, they can just reprice.
  802. 28:50They can move hour by hour, certainly
  803. 28:52day by day or week by week. The Federal
  804. 28:54Reserve has You know, we make policy and
  805. 28:56it stays for a while or it usually stays
  806. 28:59for a while, and importantly people make
  807. 29:01decisions based on it. So I think the
  808. 29:03markets are assisting in making sure
  809. 29:06that we're bringing inflation down, but
  810. 29:09the job is ours. And so, you know,
  811. 29:11that's why I have said that that my bias
  812. 29:13is we'll have to have additional
  813. 29:15tightening, but exactly how much and
  814. 29:18exactly when, that's a decision that we
  815. 29:20don't need to make today because we have
  816. 29:22many more pieces and days of information
  817. 29:24coming in over many many quarters to
  818. 29:27make the right decision that helps do
  819. 29:29those two mandated goals.
  820. 29:32>> Fantastic. No, being data dependent and,
  821. 29:35you know, working I guess you said the
  822. 29:38the markets are complementing.
  823. 29:41>> They're complementing. And they seem to
  824. 29:42understand what we have said we would
  825. 29:45deliver, price stability, and they think
  826. 29:48they look out and see a labor market is
  827. 29:49stable. If you want to deliver price
  828. 29:51stability and a stable labor market with
  829. 29:53good growth, that means
  830. 29:55tightening policy.
  831. 29:57>> So let's maybe I'll ask you a few
  832. 29:58questions about AI and then we'll open
  833. 30:00things up. You know, you had the the
  834. 30:02chart you showed before of of GDP growth
  835. 30:05with and without sort of the I think it
  836. 30:07was the you stripped out CapEx, AI CapEx
  837. 30:10from it. Um so, you know, really high
  838. 30:12level, uh
  839. 30:14how many of our economic eggs are in the
  840. 30:16AI basket? And
  841. 30:18if there is a correction,
  842. 30:21uh you know, pullback in CapEx,
  843. 30:24a correction in the stock market,
  844. 30:25how exposed are are we? What are the
  845. 30:29What are the channels? How big are the
  846. 30:31risks? How are you thinking about, you
  847. 30:32know, this scenario? And obviously
  848. 30:33nobody has a crystal ball, but but we do
  849. 30:37scenario planning, right? It's a plan is
  850. 30:39better than no plan. Uh so how do you
  851. 30:41think about?
  852. 30:41>> So I I mean I I think about it that's
  853. 30:43why I I you know I tried to get at this
  854. 30:44question. It's a great question by the
  855. 30:46way. It's excellent question because you
  856. 30:48look at that chart on GDP growth and you
  857. 30:50think all of our eggs are in one basket.
  858. 30:52But it's not really the experience I
  859. 30:54have when I talk to CEOs in industries
  860. 30:56that are not technology driven. They're
  861. 30:58feeling somewhat constrained by the It
  862. 31:00just there's a one possibility is that
  863. 31:02everybody else is silent and sitting on
  864. 31:05the sidelines and AI is the only thing
  865. 31:06saving us. The other explanation which
  866. 31:08you'll hear from other businesses is
  867. 31:10everything's going up and so I'm kind of
  868. 31:12feeling like I can't do everything I
  869. 31:14want to do because prices are so high.
  870. 31:16So I I do think there's that balance of
  871. 31:18the economy and right now there's a lot
  872. 31:20of investment in technology not just by
  873. 31:22technology companies but by regular
  874. 31:24companies that are non-tech that want to
  875. 31:27get in part of this game. So I think
  876. 31:29that's there. The other thing that I
  877. 31:30think is useful to remember and that's
  878. 31:32why I showed the stock market
  879. 31:33>> I actually pause on that? Like is that
  880. 31:36the sort of flip side of the coin of
  881. 31:38elevated rates from AI uh investment or
  882. 31:43sort of crowding out certain things that
  883. 31:45if there was a pullback then we would
  884. 31:46see an increase in other investment and
  885. 31:48it would be a stabilizing force.
  886. 31:49>> It just it I mean I it's hard to say
  887. 31:51that you could replace all the AI
  888. 31:52investment which is in a surge period
  889. 31:55with other investment. But it isn't the
  890. 31:58it is the case that other businesses
  891. 32:00that are non-tech are investing. And
  892. 32:02when I hear uh businesses across the
  893. 32:05district and even across the country
  894. 32:06talk they're they want to invest more.
  895. 32:09They feel the opportunity of the US
  896. 32:10economy. But that can't be completely
  897. 32:13separated from the surge in growth from
  898. 32:15technology. It's really just a sense
  899. 32:17that maybe we're on the
  900. 32:19I have several people tell me we're on
  901. 32:20the brink of uh a manufacturing
  902. 32:24revolution in the United States. We're
  903. 32:25on the brink of bringing AI to the
  904. 32:27physical infrastructure that we do.
  905. 32:29Isn't this great? What can I do with it?
  906. 32:31But we're not there yet. I think the
  907. 32:33other thing that is important is and
  908. 32:35this is where that stock market
  909. 32:36valuation chart comes in. Yes, we've
  910. 32:39taken some of the the enthusiasm off,
  911. 32:42right? But it hasn't taken the
  912. 32:44technology away. And ultimately that's
  913. 32:46because there is there there. It may be
  914. 32:50it may take a while to see the full
  915. 32:52impact, but even
  916. 32:5410% of the impact, probably less of what
  917. 32:57people imagine could be possible down
  918. 32:59the road, still contributes to the
  919. 33:01economy and doesn't leave it um
  920. 33:03completely zero. So the worst case
  921. 33:05scenarios are if you go from all of this
  922. 33:07to zero. I don't have a lot of weight on
  923. 33:10the zero hypothesis. I think you could
  924. 33:13slow the growth rate and still have a
  925. 33:14robust economy growing and a stable
  926. 33:18labor market and lower inflation.
  927. 33:21>> And that's it's actually similar. I just
  928. 33:22sort of did some of these numbers. I
  929. 33:24know my colleagues have done too like
  930. 33:26run a reduction in CAPEX, run the wealth
  931. 33:30effect of a correction, you know,
  932. 33:32through basic models. And you you get a
  933. 33:35slowdown, but I don't think you get a
  934. 33:37reversal in in GDP growth. But, you
  935. 33:39know, then there's every, you know, the
  936. 33:42current sort of theory of recessions,
  937. 33:44right? Everyone is different and, you
  938. 33:46know, maybe circular financing or
  939. 33:48something else
  940. 33:49uh it could create some sort of
  941. 33:52more uh deeper slowdown.
  942. 33:54>> You're absolutely right. I mean, that's
  943. 33:56always there's always the risks out
  944. 33:57there. One of the things that that I I
  945. 33:59think was the test case for whether that
  946. 34:02scenario is we're right on the brink of
  947. 34:04it is we've had a sentiment shock,
  948. 34:06right? On AI. And it hasn't unraveled.
  949. 34:09It's been orderly. So, you know, it's
  950. 34:11just not that doesn't I never am a you
  951. 34:13can't be a central banker and never be
  952. 34:14complacent. You can't say, oh, because I
  953. 34:16saw this piece of data, now I feel
  954. 34:17great. But I do think, you know, looking
  955. 34:19for signs of what the likely scenarios
  956. 34:22are are are what's playing out. You
  957. 34:25don't see really burgeoning signs of of
  958. 34:28fragility there.
  959. 34:29>> Right. So let me ask a couple labor
  960. 34:31market questions and we'll open things
  961. 34:32up. Um
  962. 34:34first one I want to go back to, you
  963. 34:36know, you had uh a slide on increasing
  964. 34:40unemployment rates for young college
  965. 34:42graduates. Uh, and I think to to
  966. 34:46summarize your slides and and I think
  967. 34:48broadly held views, you know, we're in a
  968. 34:50a low hire, low fire labor market where
  969. 34:54the most stress is, you know, on young
  970. 34:57college graduates, you know, including
  971. 34:59lots of students that, you know, I talk
  972. 35:02to, I I mentor. Uh,
  973. 35:04and
  974. 35:05there's a host of of theories out there
  975. 35:08for what's going on from work from home.
  976. 35:11I think Nick Bloom has talked about how
  977. 35:13it's harder to monitor a and uh
  978. 35:17and coach a young employee if you're
  979. 35:20more work from home uh to higher
  980. 35:23interest rates in a firms are feeling
  981. 35:25more pressure. This is where pulling
  982. 35:26back to AI,
  983. 35:29uh right that number of the tasks that
  984. 35:31uh young associates would do right now
  985. 35:34AI is doing.
  986. 35:35Uh,
  987. 35:37what is your review as to the different
  988. 35:40factors here? What are you giving the
  989. 35:41most weight? Uh,
  990. 35:43how concerned should we be?
  991. 35:44>> so I've done I've been a labor economist
  992. 35:45my whole career and one of the things
  993. 35:47I've learned in that period is there's
  994. 35:49no one factor that if you wait it you
  995. 35:51won't regret it the next day uh because
  996. 35:53ultimately all of these factors are
  997. 35:54playing out and probably the combination
  998. 35:56of them, right? Right now we have firms
  999. 35:58that feel cautiously optimistic,
  1000. 36:00underlying cautiously, right? So they
  1001. 36:02see this, but if you want to be if
  1002. 36:04you're cautious, you're uncertain, you
  1003. 36:06certainly don't want to get over your
  1004. 36:08skis so to speak by hiring a lot of
  1005. 36:09individuals that you then have to
  1006. 36:11painfully, you know,
  1007. 36:13let go. So that's one piece. The second
  1008. 36:15piece is AI, you're spending so much
  1009. 36:18time and energy thinking about AI and
  1010. 36:21you really need people who can evaluate
  1011. 36:23whether you're getting the right answers
  1012. 36:25and newly minted people who just started
  1013. 36:27your firm are unlikely to be those great
  1014. 36:29evaluators, probably a little bit why
  1015. 36:31the senior groups are feeling pretty
  1016. 36:33okay. Like they're saying I I know that
  1017. 36:35model's wrong, I know that output's
  1018. 36:36wrong, but ultimately that's a
  1019. 36:38transition dynamic. One we spend a lot
  1020. 36:40of time because this is such a critical
  1021. 36:42issue for the US and for policy making,
  1022. 36:45uh, for the Federal Reserve. We spend a
  1023. 36:47lot of time asking our, uh,
  1024. 36:50vast network of employers,
  1025. 36:52why are you hiring so little? And they
  1026. 36:54give us a array of those answers, but
  1027. 36:56one that you didn't mention that I think
  1028. 36:58is potentially beneficial for younger
  1029. 37:01workers in the transition is that they
  1030. 37:03said we don't know who to hire.
  1031. 37:05So here's an example. We're putting
  1032. 37:07we're trying to rethink everything we do
  1033. 37:09in accounting. If we make sure our
  1034. 37:11customers are are getting the what they
  1035. 37:13need, make sure we're getting what we
  1036. 37:14need. We're revamping the entire
  1037. 37:16accounting production chain for a a
  1038. 37:20national company.
  1039. 37:21If I hire a newly minted CPA and then I
  1040. 37:24don't need as many newly minted CPAs, I
  1041. 37:27need a CPA plus a person who does cyber
  1042. 37:30or something, then I've hired the wrong
  1043. 37:32person and I don't want to hire the
  1044. 37:34wrong person because it's painful to let
  1045. 37:36people go. So I'm going to go skinny.
  1046. 37:38This is often how they say it. I'm going
  1047. 37:39to go skinny until I figured that out.
  1048. 37:41So then the next question we ask is when
  1049. 37:43do you think you'll figure that out? And
  1050. 37:45they say next year, so 27. So this was
  1051. 37:48never going to be a year of rapid hiring
  1052. 37:51because of the the different shocks that
  1053. 37:53we've faced, the higher input costs,
  1054. 37:55they're trying to get their budgets
  1055. 37:56right. But I think that from the AI
  1056. 37:58perspective, you know,
  1057. 38:00it's very hard and this is maybe a theme
  1058. 38:03of the whole conversation. It's
  1059. 38:05extremely hard to take what we see today
  1060. 38:07and extrapolate because if we
  1061. 38:09extrapolate, we're likely to be wrong.
  1062. 38:11And so instead it's like what could
  1063. 38:13happen and what would what would change
  1064. 38:15that equation? And I think right now
  1065. 38:18we're in more or less a perfect storm
  1066. 38:20for hiring. It's cautious,
  1067. 38:23uh,
  1068. 38:24probably trying to get some of the
  1069. 38:25overhang from the post pandemic, you
  1070. 38:27know, labor market hiring down, AI and
  1071. 38:32younger workers are going to be the
  1072. 38:33marginal
  1073. 38:35tariffs,
  1074. 38:36so much input, so many input costs
  1075. 38:38rising and so it's just not going to be
  1076. 38:40a great impetus for hiring. What has
  1077. 38:42been
  1078. 38:44noticeably, um, a sign of resilience in
  1079. 38:47the labor market is there hasn't been a
  1080. 38:48big increase in firing or separation. So
  1081. 38:51I think that's the good news piece of
  1082. 38:52this labor market.
  1083. 38:54>> All right. So let me
  1084. 38:56last question let's actually think
  1085. 38:58through
  1086. 38:59uh you know what may happen over the
  1087. 39:01medium run in in the labor market. And
  1088. 39:05there's two scenarios that I keep in my
  1089. 39:07mind which I find
  1090. 39:09uh clarifying, right? And like you said
  1091. 39:11you know the truth is always a complex
  1092. 39:12combination of different scenarios. So
  1093. 39:14the sort of benign
  1094. 39:17scenario I have for uh
  1095. 39:20AI labor market disruption is
  1096. 39:23you know we have
  1097. 39:25uh
  1098. 39:26job losses or you know
  1099. 39:29changes in hiring mix that happen
  1100. 39:32broadly across the country and happen
  1101. 39:35slowly over time.
  1102. 39:37And that sort of transition gets almost
  1103. 39:41lost in the huge amount of churn we see
  1104. 39:43in the labor market as it is, right? You
  1105. 39:45know
  1106. 39:46communities hire fewer people in this
  1107. 39:49role, more people in this role uh to get
  1108. 39:52lost in the churn and we can manage it.
  1109. 39:53We have a very dynamic economy.
  1110. 39:56You know the other scenario when I think
  1111. 39:57about
  1112. 39:59the hollowing out of the labor market
  1113. 40:01and in particular the the loss of these
  1114. 40:03sort of routine jobs are you think about
  1115. 40:06Charlie Chapman in in modern times doing
  1116. 40:08the same thing over and over and again.
  1117. 40:10Like if you look in the data it's
  1118. 40:12something like almost 90% of routine
  1119. 40:14jobs were lost during recessions of the
  1120. 40:17six months afterwards.
  1121. 40:19When firms are cutting back and then
  1122. 40:21they figure out how to do more with less
  1123. 40:23uh and it's cheap to install capital
  1124. 40:26because interest rates are low and wages
  1125. 40:28are sticky so wages are high. Uh so you
  1126. 40:30have
  1127. 40:31job losses compounded
  1128. 40:34structural job losses compounded with
  1129. 40:37cyclical shocks.
  1130. 40:38And I think that's the scenario I'm most
  1131. 40:40worried about with AI that there is a
  1132. 40:42recession and employers figure out how
  1133. 40:45to do more with less and we don't hire
  1134. 40:47back then the white collar service
  1135. 40:50workers, and we have, you know, the
  1136. 40:53longest jobless recovery that we've ever
  1137. 40:56seen. So that's my worst case scenario.
  1138. 40:58What what are the scenarios that you're
  1139. 41:00thinking about?
  1140. 41:02What do you think about the scenarios
  1141. 41:03that I just laid out here? Uh, and then
  1142. 41:05hopefully people will have uh more
  1143. 41:07optimistic questions we can finish on.
  1144. 41:09>> I I think those are two reasonable
  1145. 41:12scenarios, but I think there's many more
  1146. 41:13scenarios in the in between than I also
  1147. 41:15think about. And and that's because, you
  1148. 41:17know, ultimately we don't know, as you
  1149. 41:19rightly said, those are two you know,
  1150. 41:21kind of put them at the benchmark part.
  1151. 41:23One thing that I will say since you
  1152. 41:25brought up you a recession and this is
  1153. 41:26when it happens, another time that you
  1154. 41:28can see job transformation and employers
  1155. 41:31really thinking about how do I automate
  1156. 41:33is in response to a hot job market. So
  1157. 41:36we had a very very tight labor market
  1158. 41:40coming out of the pandemic, very tight.
  1159. 41:42I mean, people were losing employees two
  1160. 41:45weeks after they hired them or maybe
  1161. 41:47they hired them and they didn't even
  1162. 41:48show up. That's how remember ghosting
  1163. 41:50and all this. So that's how tight it
  1164. 41:52was.
  1165. 41:53The a response to that type of tight
  1166. 41:55labor market is I'm automating. I can't
  1167. 41:58be left
  1168. 42:00to
  1169. 42:01wait for workers to come. I'm going to
  1170. 42:02automate things I can. That happened to
  1171. 42:05collide or correspond, I guess is a
  1172. 42:07better word, with the AI boom. And so I
  1173. 42:10think one of the reasons we're seeing a
  1174. 42:12little bit of this right now without a
  1175. 42:14recession, as you say, this is a
  1176. 42:16classic, you know, historical example,
  1177. 42:18but without that is because the firms
  1178. 42:19already got the discipline of, you know,
  1179. 42:21I better not rely on a labor force
  1180. 42:24that's always available. And then you
  1181. 42:26put onto that the fact that our labor
  1182. 42:29force growth is really slow going
  1183. 42:31forward in the absence of immigration it
  1184. 42:33could get to zero by 2030 in terms So
  1185. 42:35the employers like, okay, if that's what
  1186. 42:38I've got, then how can this technology
  1187. 42:40really help me continue output growth
  1188. 42:44while I do more with fewer employees?
  1189. 42:47And that's a response to those two
  1190. 42:49things and I think that's a scenario
  1191. 42:51that's probably top my mind right now is
  1192. 42:54what is a transition dynamic look like
  1193. 42:56when you have two accelerants, right?
  1194. 42:58The slower labor force growth coming off
  1195. 43:00a frothy labor market and you have the
  1196. 43:03AI boom. How do employers combine those
  1197. 43:06two things and how does that change the
  1198. 43:08steady state demand for workers? But I
  1199. 43:10always come back to something I was on
  1200. 43:11stage with Patrick Collison of Stripe
  1201. 43:13one about a two years ago at this point,
  1202. 43:16but it's still it's still something he
  1203. 43:17says. And he said, look, I'm hiring as
  1204. 43:20many workers as I'm
  1205. 43:22not as I'm letting go and the thing is
  1206. 43:25they're just totally different people.
  1207. 43:26So when you think about a labor market
  1208. 43:28that is able to continue and thrive,
  1209. 43:31it's about how we intentionally help all
  1210. 43:34of our workforce learn the skills that
  1211. 43:36are going to be necessary
  1212. 43:38to embark on the new technology and the
  1213. 43:40new jobs and the new tasks that come out
  1214. 43:41of that and not you know, kind of stay
  1215. 43:45behind on the old
  1216. 43:47on the old
  1217. 43:48like spreadsheet and say, well, that's a
  1218. 43:50job, that's a job, that's a job. I think
  1219. 43:52we're we're likely to see many many new
  1220. 43:54jobs as we lose older jobs and I just
  1221. 43:57want to make sure our workforce is
  1222. 43:58ready.
  1223. 43:59>> Yeah, no, that's that's a great scenario
  1224. 44:02also to keep top of mind. Let me flip to
  1225. 44:04questions and just seeing one from from
  1226. 44:06Karen. I don't know oh, wait, there she
  1227. 44:08is.
  1228. 44:09She asks, what indicators will you be
  1229. 44:12watching to assess whether AI indeed
  1230. 44:14will support productivity and lower
  1231. 44:16interest rates? And I think we didn't
  1232. 44:17touch on that. So yeah, what what are
  1233. 44:20the
  1234. 44:21you know, we all know the productivity
  1235. 44:22series is noisy and it's it's hard to
  1236. 44:26just focus on that. What are you looking
  1237. 44:27at?
  1238. 44:28>> So you you know, Karen, thanks for the
  1239. 44:29question.
  1240. 44:31We talked we've talked about this a lot,
  1241. 44:33but you know, one of the things that and
  1242. 44:34I I was benefited by starting at the Fed
  1243. 44:37in 1996. I was a a microeconomist, labor
  1244. 44:39economist and then Chairman Greenspan
  1245. 44:42would say, I think productivity growth
  1246. 44:44is there and he'd set teams across the
  1247. 44:46country to look for evidence or no
  1248. 44:48evidence of this. And one of the things
  1249. 44:51that I experienced was companies telling
  1250. 44:54us that they were making these
  1251. 44:55productivity gains and it was helping
  1252. 44:57them protect margins while they
  1253. 44:58continued to create output growth. And I
  1254. 45:02the lesson I learned from that isn't the
  1255. 45:03technology shocks are always great for
  1256. 45:06productivity growth right away, but that
  1257. 45:08you're never going to find it in the
  1258. 45:09data until it's already happened. And so
  1259. 45:12to be able to find it
  1260. 45:15you got to look away from the published
  1261. 45:16data on productivity growth, not to
  1262. 45:18discount it. I think it's useful, but
  1263. 45:20that is not going to give you a leading
  1264. 45:22indicator of what's happening. To do to
  1265. 45:24get that indicator, you have to talk to
  1266. 45:26firms. You have to ask them what they're
  1267. 45:28doing, how they're doing it, what's
  1268. 45:29changed. The second thing that is really
  1269. 45:32important that I look for is
  1270. 45:34you know, I I go back I'm a I love
  1271. 45:36history, so I go back I read about
  1272. 45:37deeply about electrification. If you
  1273. 45:39read any historical piece on
  1274. 45:41electrification, they tell you that
  1275. 45:43electricity is invented. It takes you
  1276. 45:45know, 100 years to to see that transform
  1277. 45:48the economy. The transformative part
  1278. 45:51wasn't even when they put an electric
  1279. 45:52motor at the end of the line to replace
  1280. 45:54the steam engine. That saved money, but
  1281. 45:56it did not change everything. What
  1282. 45:58transformed everything is the invention
  1283. 46:00of the unit motor and then you get a
  1284. 46:02unit motor which can be moved anywhere
  1285. 46:04and the unit motor itself isn't the
  1286. 46:06transformative part. What's
  1287. 46:07transformative is if businesses say,
  1288. 46:09well, my gosh, if I don't have to have
  1289. 46:10one continuous line, then I can move
  1290. 46:13around, I can move machines, I can
  1291. 46:15respond to output needs, I can do a
  1292. 46:17variety of things and that transforms
  1293. 46:20manufacturing. So that's the piece I
  1294. 46:22look for. What are businesses doing to
  1295. 46:24change not their technologies, but their
  1296. 46:27processes? And when they change their
  1297. 46:29processes, we will have the kind of boom
  1298. 46:32that that people talk about.
  1299. 46:35>> Uh so speaking of uh electrification, I
  1300. 46:39have a question on
  1301. 46:41uh energy in the here and now. So let me
  1302. 46:43let me read it out maybe in a bridge
  1303. 46:45form. Power hungry data centers are
  1304. 46:47demanding uh immense amounts of energy.
  1305. 46:50How's the Fed thinking about this demand
  1306. 46:52reverberating through markets and
  1307. 46:54contributing to broader long-term
  1308. 46:55inflation in electricity markets, in
  1309. 46:57other energy sectors, and in markets for
  1310. 47:00energy sensitive goods.
  1311. 47:03>> Now, we think about it a lot and and one
  1312. 47:05of the things that we have to think
  1313. 47:06about is what are communities doing in
  1314. 47:07response. And, you know, we have we talk
  1315. 47:10to we have round tables with the power
  1316. 47:12companies. They tell us the same thing.
  1317. 47:14It's hard to keep up with the demand for
  1318. 47:16energy
  1319. 47:17that communities have as we have growing
  1320. 47:19communities, but data centers have. It
  1321. 47:21was already hard to keep up and now we
  1322. 47:22have a demand shock from from
  1323. 47:25data centers. So, these are real
  1324. 47:27problems. The infrastructure needs are
  1325. 47:29present and the Fed can't solve those,
  1326. 47:32but you're right. We have to think about
  1327. 47:33them. So, one of the things I
  1328. 47:35just to put a practical component on how
  1329. 47:38I think of it as a policy maker is, you
  1330. 47:40know, energy prices go up, but it really
  1331. 47:43is all the people who pay for the energy
  1332. 47:45that matters for the overall price
  1333. 47:47inflation. If businesses are already
  1334. 47:49paying higher prices for oil, now they
  1335. 47:52pay higher prices for electricity.
  1336. 47:54They're paying higher prices for other
  1337. 47:55inputs due to tariffs or competing
  1338. 47:58interest from AI, then this their input
  1339. 48:00costs are going to rise and so that
  1340. 48:02means their margins will fall unless
  1341. 48:04they raise their prices. And so that's
  1342. 48:06the dynamic you have to watch. So, we
  1343. 48:08can't simply say, well, this is just
  1344. 48:10electricity, so not a big deal. These
  1345. 48:13are changing the price dynamics and if
  1346. 48:15other prices don't adjust downward, then
  1347. 48:18we're going to have a a higher pressure
  1348. 48:21on the on the inflation rate than we
  1349. 48:23would absent these shocks. And so this
  1350. 48:26has got to be something we really think
  1351. 48:27about. And, you know, if you really
  1352. 48:29really geeky, people talk about this in
  1353. 48:31the context of what's the long run
  1354. 48:33interest rate that's going to be normal
  1355. 48:35in an environment like this? The answer
  1356. 48:37is nobody knows, but you can definitely
  1357. 48:39put a table together and say things that
  1358. 48:41move it up, things that move it down,
  1359. 48:43and the the list of things that move it
  1360. 48:45up are growing relative to the things
  1361. 48:47that move it down.
  1362. 48:48>> Yep.
  1363. 48:49Um,
  1364. 48:50we'll finish with with two questions
  1365. 48:52sort of in the similar vein. First from
  1366. 48:56Peter Blair here in the front. Oh, there
  1367. 48:59are lots researchers and private sector
  1368. 49:01firms in the room. What research and
  1369. 49:03data would be most useful to the Fed and
  1370. 49:05policy making in the age of AI uh to
  1371. 49:08support the dual mandate? What's missing
  1372. 49:10from the current research efforts of
  1373. 49:12scholars and firms?
  1374. 49:13>> So I'm going to I'm going to talk about
  1375. 49:15something Peter, thanks for the
  1376. 49:17question. I appreciate it. You know, one
  1377. 49:19of the things I want to talk about on
  1378. 49:20that which you mentioned is you know,
  1379. 49:22the
  1380. 49:23can and then I'll get to the AI part
  1381. 49:24specifically, but the conventional
  1382. 49:26wisdom in economics is that you can look
  1383. 49:28through shocks, but in a world where the
  1384. 49:31pressure on inflation we're going to be
  1385. 49:33most central banks will be fighting
  1386. 49:34inflation from above our target, pulling
  1387. 49:36it down. And we've got these sequence of
  1388. 49:38shocks, global shocks, national shocks
  1389. 49:40that I think we just have to have a
  1390. 49:42broader research hunger for how do I
  1391. 49:45think of relative price shocks? What are
  1392. 49:46the avenues that that it make them uh
  1393. 49:50more durable, more sticky? How do they
  1394. 49:52change the underlying level of inflation
  1395. 49:55underlying rate of inflation? I think
  1396. 49:56that's a ripe for this. On the AI
  1397. 49:58specifically, you know, I've really
  1398. 50:00appreciated all the research that's been
  1399. 50:03done where you can't find it in the
  1400. 50:05aggregate data. This is going to hook
  1401. 50:07Karen's question to your question,
  1402. 50:08Peter. Can't find it in the aggregate
  1403. 50:10data, but we know a lot about where to
  1404. 50:12look and all the studies that are
  1405. 50:14looking at case studies and saying how
  1406. 50:16does it happen? What's the mechanism
  1407. 50:18that makes it just cost saving because
  1408. 50:21AI is doing some work to transformative
  1409. 50:24to your business? That is really great
  1410. 50:26research to think about. And then the
  1411. 50:28other part that I I really want would
  1412. 50:30love the research community to embrace
  1413. 50:32and think about is
  1414. 50:35not what bad things can happen to the
  1415. 50:37labor market eventually, but what can we
  1416. 50:40do to prevent those things happening? I
  1417. 50:44I do feel like it's it's easier
  1418. 50:46oftentimes to itemize what could go
  1419. 50:48wrong, but it's not that
  1420. 50:51it's not an instinct of most humans
  1421. 50:53really to say, well, what do we have to
  1422. 50:55change in the models or in the data to
  1423. 50:57have that go better? Like what would
  1424. 50:59have to happen? That's not just a
  1425. 51:01regulatory response, but it's actually
  1426. 51:03like a training response. What would
  1427. 51:05What how would it be different? And
  1428. 51:06looking across the uh the globe would be
  1429. 51:09helpful. Uh we had a a talk in our bank
  1430. 51:11at the We have a Technology and Growth
  1431. 51:13Center. We kicked it off a couple of
  1432. 51:15weeks ago and Philip Lane from the ECB
  1433. 51:18was giving a talk and he he talked about
  1434. 51:20how different perceptions of AI are for
  1435. 51:22workers here in the United States than
  1436. 51:24they are in the EU. I think that's
  1437. 51:26something to learn from. They're much
  1438. 51:27less pessimistic about AI than our
  1439. 51:29younger workers are. Well, why? And what
  1440. 51:32what aspects of their countries and
  1441. 51:34their policies and their programs do
  1442. 51:35they have that might mitigate some of
  1443. 51:38that fear?
  1444. 51:38>> Yeah, you know, I I love that point. I
  1445. 51:40you know, I've said to other people, we
  1446. 51:41do too much predicting and not enough
  1447. 51:44planning around AI.
  1448. 51:45>> a great way to say it.
  1449. 51:46>> Uh and you know, I think part of the
  1450. 51:49uh
  1451. 51:49purpose and you know, substance of
  1452. 51:52today's event is to also do that sort of
  1453. 51:55scenario planning and and start building
  1454. 51:57that muscle ahead of when we need it. So
  1455. 52:00let me finish with, you know, I think a
  1456. 52:01very hard but important question. We
  1457. 52:03have lots of young people, undergrads
  1458. 52:06choosing majors, choosing where to get
  1459. 52:08an internship, uh pre-docs, thinking
  1460. 52:10about whether they're going to go to
  1461. 52:12grad school in economics, go to the
  1462. 52:14private sector.
  1463. 52:15We're in a moment of huge uncertainty.
  1464. 52:17Now what's your advice to to young
  1465. 52:20people today?
  1466. 52:21>> I'm happy to give it. Um and then you
  1467. 52:23can tell me you can call me up 10 years
  1468. 52:25from now and say that was terrible
  1469. 52:27advice or that was excellent advice.
  1470. 52:29>> you have the disclaimer, right? That you
  1471. 52:31>> do have the disclaimer, but I'll I'll
  1472. 52:33I'll take this on my on myself here and
  1473. 52:36and say that, you know, I think the
  1474. 52:38the instinct that many have when you see
  1475. 52:41uncertainty is to stall out
  1476. 52:43and do less. I think the instinct that I
  1477. 52:46want you all to have is to lean in and
  1478. 52:48do more.
  1479. 52:49The economy's always going to be
  1480. 52:51dynamic. You're you are coming out in a
  1481. 52:54period of great uncertainty about what
  1482. 52:57the world will look like, but we're not
  1483. 52:58uncertain about where it is today. And
  1484. 53:00today you know you have to have AI
  1485. 53:02skills. You have to be able to
  1486. 53:04interrogate things with great curiosity.
  1487. 53:07I think there's a lot of human skills
  1488. 53:10that AI hasn't been able to replicate
  1489. 53:12yet that are going to be in high demand.
  1490. 53:14In many ways, the scarcest skills are
  1491. 53:16going to be leadership, uh, quality
  1492. 53:18control, knowing when to use a model,
  1493. 53:20when not to use it, knowing how to
  1494. 53:22interrogate or investigate what is You
  1495. 53:25know, you you started by talking about
  1496. 53:26evidence-based. What's the standard for
  1497. 53:28evidence-based? It's not that, you know,
  1498. 53:30an LLM told you the answer. The standard
  1499. 53:32for evidence-based is hypothesis
  1500. 53:34testing, you know, having that
  1501. 53:36curiosity, coming back with the data. So
  1502. 53:39I think being positioning yourself so
  1503. 53:41you're not mono-lined in a specific
  1504. 53:43thing, but you're really open. You could
  1505. 53:45be sitting
  1506. 53:47right here at the beginnings of
  1507. 53:49something that, you know, Neil and I
  1508. 53:51will be jealous of because we weren't in
  1509. 53:53the labor market. I don't think that's
  1510. 53:55where you sit right now, but you want to
  1511. 53:57be ready for that, right? You want to be
  1512. 53:59ready for that. I also deeply believe
  1513. 54:01that younger people are where all of our
  1514. 54:04the idealism meets the the practicality.
  1515. 54:07And if we want to look like a country
  1516. 54:09and a a world we want 10 years from now,
  1517. 54:11we're going to have to have each and
  1518. 54:12every one of you participate. So I say
  1519. 54:14lean in.
  1520. 54:15>> Yeah. No,
  1521. 54:17we all know as investors we should have
  1522. 54:18a diversified portfolio. One thing I
  1523. 54:20took from that comment is in terms of
  1524. 54:22our human capital, right? We should lean
  1525. 54:24in. We should be diversified, you know,
  1526. 54:26not just the cognitive skills, the human
  1527. 54:28skills,
  1528. 54:29uh, and, you know, that leaves us best
  1529. 54:31prepared for what opportunities are
  1530. 54:34going to wait us in our uncertain
  1531. 54:35future. Thank you so much.
  1532. 54:37>> Can I say what one of my interns told me
  1533. 54:38this year? He said, Mary, the thing that
  1534. 54:39I learned by working at the Fed this
  1535. 54:41year is that I'm not just what I do, I
  1536. 54:44am how I do it. And I think if you take
  1537. 54:46that to that out, you're going to find
  1538. 54:49that how you do things matters as much
  1539. 54:51as what you do. What you do can change
  1540. 54:53every three years, certainly every 10
  1541. 54:55years, but how you do it, how you come
  1542. 54:57across in the labor market, how you work
  1543. 54:59with others, how you collaborate, that's
  1544. 55:00a durable skill.
  1545. 55:01>> you you are a beacon in terms of what
  1546. 55:04you do and how you you it. Uh, and it's
  1547. 55:06been a delight to have you here. Thank
  1548. 55:09you so much.
  1549. 55:11>> Thank you.

About this transcript

This page contains the full transcript of Keynote Remarks by Mary C. Daly at SIEPR’s Policy Forum by Federal Reserve Bank of San Francisco, generated from the public captions YouTube serves with the video. The transcript has 10,393 words across 1,549 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.