Latest / Elon Musk Podcast / Anthropic says White Collar jobs are going away

Transcript

Anthropic says White Collar jobs are going away · Elon Musk Podcast

• 0:00
  Generative artificial intelligence is technically
• 0:02
  capable of completing nearly all the tasks required of computer
• 0:06
  programmers and mathematicians, yet actual workplace usage shows
• 0:10
  it is currently performing only a third of those tasks.
• 0:13
  That capability gap is just massive.
• 0:16
  I mean, looking at the sources we have on labor market
• 0:18
  research, you really start to realize the immense delta
• 0:21
  between what artificial intelligence can do and what it
• 0:24
  is actually doing across the white collar workforce.
• 0:27
  The reality inside Cororate offices just looks very
• 0:30
  different from the theoretical benchmarks.
• 0:32
  Right. So if the technology has this
• 0:34
  massive untapped potential to do our work, why aren't we seeing
• 0:38
  massive unemployment? And who is actually taking the
• 0:41
  economic hit right now? Well, we're seeing a complete
• 0:44
  shift in automation targets for a long time.
• 0:46
  You know, the assumption was always that machines would come
• 0:49
  for physical labor first, like the factory worker or the
• 0:52
  warehouse Packer. The delivery driver.
• 0:55
  Exactly. But the automation we are seeing
• 0:57
  right now is focusing directly on cognitive tasks.
• 1:01
  To understand this, we have to look at a concept called
• 1:03
  observed exposure. This measures what these systems
• 1:07
  are actually erforming in real rofessional settings, rather
• 1:10
  than just what they're programmed to achieve in theory.
• 1:13
  Which basically means researchers are looking at the
• 1:15
  actual screens of workers on a random afternoon, right?
• 1:18
  They're measuring how many keystrokes or how many e-mail
• 1:22
  drafts are being handled by human fingers versus automated
• 1:25
  Co pilots. Yeah, and the data shows the
• 1:27
  professions most exposed to automation are no longer manual
• 1:31
  laborers. They are highly educated, higher
• 1:33
  paid professionals. We're looking at computer
• 1:35
  programmers, customer service representatives, data entry
• 1:39
  keyers. And financial analysts?
• 1:40
  Right. Oh, absolutely.
• 1:41
  They sit right at the top of that exposure list.
• 1:43
  And when you look at the demographics of this exposure,
• 1:46
  it completely flips our traditional understanding of job
• 1:49
  security. The workers facing the highest
• 1:51
  risk are disproportionately older, female and hold graduate
• 1:55
  degrees. Oh.
• 1:56
  Wow, that's really counterintuitive.
• 1:58
  It is A worker in the most exposed group earns
• 2:01
  significantly more on average, and is like nearly four times as
• 2:05
  likely to hold a graduate degree compared to someone in the least
• 2:08
  exposed group. Think about the contrast there.
• 2:11
  You have a 50 year old financial analyst with a master's degree
• 2:14
  who is highly exposed while physical roles like cooks,
• 2:17
  mechanics and bartenders show 0 exposure.
• 2:20
  Right, because the bartender relies on physical dexterity,
• 2:23
  reading the room and managing chaotic real world physics.
• 2:27
  Software cannot pour a drink or fix a shattered glass.
• 2:30
  But the financial analyst, her whole job is concentrated
• 2:33
  entirely on information processing, linguistic synthesis
• 2:37
  and complex decision making. That is exactly the architecture
• 2:40
  of the software is built to handle.
• 2:41
  Wait back up. If these high paying jobs are so
• 2:44
  exposed, are these workers actually getting fired?
• 2:47
  Well, no. Because of organizational
• 2:49
  friction, legal constraints, and just basic trust issues, the
• 2:53
  technology is largely being used to assist rather than replace.
• 2:57
  This limits immediate job losses, meaning we aren't
• 3:00
  currently experiencing a massive spike in unemployment for white
• 3:02
  collar workers. Ah, I see.
• 3:04
  Yeah, the reason for this comes down to how corporate
• 3:07
  environments actually function in the real world.
• 3:10
  Deploying automated systems requires a level of trust and
• 3:13
  legal compliance that the software simply cannot guarantee
• 3:17
  on its own. Right.
• 3:18
  Think about the legal constraints alone.
• 3:20
  If an automated system hallucinates a bad financial
• 3:23
  projection or drafts a contract with a fatal flaw, and a client
• 3:27
  loses millions of dollars, who gets sued?
• 3:30
  You cannot sue an algorithm. No, you can't put a piece of
• 3:33
  software in jail. Companies still need human
• 3:35
  judgement to verify accuracy and act as a human shield for
• 3:38
  liability. They need a senior employee to
• 3:41
  sign their name on the dotted line and take legal
• 3:43
  responsibility for the output. Exactly.
• 3:46
  And beyond the legal liability, there's the necessity of
• 3:49
  managing complex client relationships and handling
• 3:51
  sensitive exceptions. Business is fundamentally about
• 3:54
  human relationships and trust. So if a major client has a
• 3:58
  highly specific, nuanced problem that does not fit neatly into a
• 4:02
  standard template, the software struggles.
• 4:05
  Yeah, and that reliance on human supervision acts as a massive
• 4:08
  buffer. It's the primary reason why we
• 4:11
  aren't seeing senior analysts and experienced programmers
• 4:13
  getting handed pink slips on mass.
• 4:16
  OK. So the senior folks are
• 4:17
  relatively safe. For now, yes, but the labor
• 4:21
  market is experiencing A shrinking hiring pipeline for
• 4:24
  junior employees. The hiring rate for young
• 4:27
  workers, specifically those ages 22 to 25 in highly exposed
• 4:31
  fields has dropped by approximately 14 to 16%.
• 4:35
  Wow. Yeah.
• 4:36
  And overall entry level job postings have fallen roughly
• 4:39
  35%. That is a massive drop.
• 4:42
  A 35% reduction in entry level postings means that over a third
• 4:46
  of the doors normally open to recent college graduates have
• 4:49
  just been nailed shut. It's what's driving this trend
• 4:51
  of quiet hiring. Instead of actively recruiting
• 4:54
  new talent, companies are simply letting senior staff retire
• 4:57
  without backfilling the roles. Or they're relying on their
• 5:02
  current experienced workers who are now using artificial
• 5:05
  intelligence to work faster to absorb all the junior level
• 5:09
  workload. Think about a traditional
• 5:12
  biology ecosystem. If you clear cut all the
• 5:14
  saplings, the old growth forest looks incredibly healthy today.
• 5:18
  Right, the tall trees are still there, getting plenty of
• 5:20
  sunlight. But the ecosystem faces a crisis
• 5:22
  when the old trees eventually fall.
• 5:24
  When those senior executives retire, there will be no new
• 5:27
  canopy to replace them because the young trees were never
• 5:30
  allowed to grow. That analogy perfectly captures
• 5:33
  the structural problem. If the bottom rung of the career
• 5:36
  ladder vanishes, it severely limits the development of future
• 5:40
  senior talent. Organizations risk a massive
• 5:42
  expertise deficit because no one is gaining the necessary
• 5:45
  foundational experience. Think about your own job when
• 5:49
  you first started, how much of your day was spent just
• 5:51
  summarizing meeting notes, organizing spreadsheets, or
• 5:54
  drafting standard reports? Oh, almost all of it.
• 5:57
  But that grunt work was actually teaching you how your industry
• 6:00
  worked. The tasks that young
• 6:03
  professionals historically use to learn their industry are
• 6:06
  exactly the tasks being automated.
• 6:08
  Without those entry level opportunities, the mechanisms
• 6:11
  for transferring tacit knowledge and building intuition disappear
• 6:15
  completely. Tacit knowledge isn't something
• 6:18
  you can learn from reading a corporate handbook.
• 6:20
  Right. You learn it by sitting in a
• 6:21
  meeting, being told to write the summary, and listening to how
• 6:25
  the senior partners negotiate. You learn the unspoken rules of
• 6:29
  your profession by doing the basic repetitive tasks over and
• 6:32
  over until you understand the underlying structure of the
• 6:35
  business. So if the software writes the
• 6:37
  summary, the junior employee is never in the room.
• 6:40
  Exactly. They never hear the negotiation
• 6:42
  and they never build that critical intuition.
• 6:44
  If you've got a decent mic and laptop and some free time,
• 6:47
  Babble Audio is paying people to record speech data and annotate
• 6:51
  audio for AI training. No minimums, no fixed hours.
• 6:56
  You work when you want and get paid weekly via PayPal, Venmo,
• 6:59
  or bank transfer. They pay per recorded or
• 7:01
  annotated hour, plus bonus challenges for hitting weekly
• 7:05
  goals. And if you sign up through our
• 7:06
  link, you get priority processing and a $15 bonus.
• 7:10
  Links in the show notes. So getting back to the sources,
• 7:13
  we see that artificial intelligence alters daily tasks
• 7:15
  by acting as a great leveller for novices.
• 7:18
  The technology provides the largest productivity boost to
• 7:21
  lower skilled or less experience, right?
• 7:23
  Allowing them to perform tasks entirely outside their core
• 7:26
  domain. For example, imagine a junior
• 7:29
  data scientist who's great with numbers but has no background in
• 7:31
  marketing. Oh yeah, with these new tools,
• 7:34
  they can stretch into marketing analytics, write compelling copy
• 7:37
  for campaigns, and generate visual assets.
• 7:39
  It makes them a highly versatile full stack employee who could
• 7:42
  operate across multiple departments.
• 7:44
  There is a real danger of deskilling here, though.
• 7:47
  Internal surveys from software engineers show a growing anxiety
• 7:51
  within the profession. They worry that relying on code
• 7:54
  generation will cause them to lose their deep technical
• 7:57
  competence. Like losing their ability to
• 8:00
  properly supervise the output. Yeah, it's essentially cognitive
• 8:04
  offloading. Just like how you lose your
• 8:05
  sense of direction when you blindly follow a GPS application
• 8:08
  every time you drive, your mental muscles begin to atrophy
• 8:12
  when you stop doing the heavy lifting of complex problem
• 8:15
  solving. Hold on, so the exact tool
• 8:17
  making you faster today could actually make you worse at your
• 8:20
  job tomorrow? Yes, it creates a dangerous
• 8:22
  recycle. When workers constantly delegate
• 8:25
  complex problem solving to an automated system, their own
• 8:28
  critical faculties can begin to decay.
• 8:30
  So if a software engineer stops writing the fundamental logic
• 8:34
  structures and only acts as an editor for automated code, they
• 8:38
  slowly forget the underlying architecture.
• 8:40
  And when a truly novel problem arises, something the software
• 8:43
  has never seen before, they lack the mental muscle to solve it.
• 8:47
  Their capacity to identify subtle errors or design original
• 8:50
  solutions is deeply compromised. Which completely changes how we
• 8:54
  value human capital. It commoditizes narrow
• 8:57
  expertise, driving down the wage premium for certain highly
• 9:00
  specialized skills. Exactly.
• 9:03
  If an automated system can instantly perform the highly
• 9:05
  specialized coding trick you spend years mastering, that
• 9:09
  specific skill is no longer scarce.
• 9:10
  You cannot charge a premium for it anymore.
• 9:12
  Instead, it rewards workers who have broad adaptability and
• 9:16
  strong critical thinking to verify automated outputs.
• 9:19
  True value shifts toward those who can manage multiple domains
• 9:23
  and synthesize disparate information.
• 9:26
  This brings into focus the concept of pro worker
• 9:28
  technology. This involves using artificial
• 9:31
  intelligence to make human skills more valuable by giving
• 9:34
  workers the ability to tackle entirely new, complex problems.
• 9:37
  Right, which stands in direct contrast to pure automation,
• 9:41
  which simply replaces human labor entirely in order to cut
• 9:44
  costs. We see specific examples of this
• 9:46
  augmentation in the field right now.
• 9:48
  Yeah, consider an electrician's assistant tool.
• 9:51
  This software analyzes complex sensor data from industrial
• 9:54
  machinery and automatically drafts detailed maintenance
• 9:57
  reports. And it cuts the report writing
• 9:59
  time in half. But the electrician does not
• 10:02
  lose their job. Instead, because the software
• 10:05
  handles the tedious paperwork, the field technician can spend
• 10:08
  their time completing far more challenging physical repairs.
• 10:12
  They're up on the ladder managing complex wiring systems,
• 10:15
  doing the high value physical work that the software cannot
• 10:18
  touch. We see a very similar dynamic in
• 10:20
  the healthcare sector. The technology handles the
• 10:23
  incredibly heavy documentation burdens that plague medical
• 10:26
  professionals. Medical coding, patient
• 10:29
  charting, and insurance summaries are largely handled by
• 10:32
  the software. And this is fueling massive
• 10:34
  projected growth for roles like nurse practitioners.
• 10:37
  They can spend significantly more time on direct patient
• 10:40
  care, having actual conversations with the people
• 10:42
  they are treating, because the software manages the
• 10:45
  administrative load behind the scenes.
• 10:47
  So if businesses choose to deploy these tools
• 10:49
  collaboratively rather than is pure automation, it opens up the
• 10:53
  potential to raise global labor productivity by roughly 3%
• 10:56
  annually. Changes the fundamental equation
• 10:59
  for the economy. When businesses deploy
• 11:02
  technology to augment human capabilities, they create
• 11:04
  entirely new tasks and new business models, ultimately
• 11:08
  stimulating demand for human labor.
• 11:10
  Think about the economic multiplier effect here.
• 11:13
  When goods and services become far more efficient to produce,
• 11:17
  the overall cost drops. As the cost drops, overall
• 11:21
  economic demand grows. Right.
• 11:23
  Consumers have more capital to spend elsewhere, which generates
• 11:26
  entirely new categories of work. We saw this with the
• 11:29
  introduction of spreadsheet software in the past.
• 11:32
  It did not eliminate the accounting profession.
• 11:34
  It made financial calculations so cheap and easy that the
• 11:37
  demand for complex financial analysis exploded.
• 11:40
  Collaborative deployment requires human judgment,
• 11:43
  empathy, and strategic thinking to guide the software, creating
• 11:46
  a massive new frontier for human labor.
• 11:48
  The disruption of white collar work is arriving as a quiet
• 11:51
  structural shift that squeezes out entry level opportunities
• 11:55
  and forces us to completely rethink how human expertise is
• 11:58
  built and valued. If the foundational, repetitive
• 12:01
  tasks that usually teach us our professions are entirely
• 12:04
  automated, how will you acquire the deep, tacit knowledge
• 12:08
  required to lead the workforce of the future?
• 12:11
  If you're not subscribed yet, take a second and hit follow on
• 12:14
  whatever a you're using. It hels us Kee making this.
• 12:16
  We appreciate you being here.