The Last 7 Years of Human Work - Understanding the AUTOMATION CLIFF!

00:29:07
https://www.youtube.com/watch?v=_wL9DI9pNuc

摘要

TLDRThe video delves into the 'Automation Cliff' concept, emphasizing the stark difference between incremental automation and full-scale automation, predicting a future where many traditional jobs will disappear due to advancements in AI and robotics. The speaker discusses drop-in technologies that facilitate quick adoption, historical examples of successful automation, and industry-specific implications. A timeline for widespread adoption of automation technologies is presented, and the potential challenges of hybrid workflows are highlighted, ultimately painting a picture of a transformative future in various fields, including pharmaceuticals, agriculture, and emergency response.

心得

  • 🚀 Full automation leads to significant improvements compared to incremental changes.
  • ⚙️ Drop-in technologies allow for quick adjustments and adoptions in existing systems.
  • 👨‍⚕️ Industries like pharmaceuticals and agriculture are already benefiting from automation.
  • 📅 A predicted timeline suggests mass automation adoption could happen by 2025-2033.
  • 🤖 Humanoid robots and AI could take over jobs traditionally held by humans.
  • 🌐 Automation could lead to a paradigm shift in knowledge work across many sectors.

时间轴

  • 00:00:00 - 00:05:00

    The speaker introduces the concept of the 'automation cliff,' which emphasizes the importance of fully automating processes rather than making incremental improvements. This idea is illustrated through the comparison of different levels of automation, similar to Tesla's Full Self-Driving (FSD) levels. The speaker discusses the challenges and implications of transitioning from high human involvement to full automation.

  • 00:05:00 - 00:10:00

    The automation cliff occurs alongside 'drop-in technologies' that can completely replace traditional methods of doing tasks without human oversight. Examples of such technologies include USB, cloud integration, and GPS, each representing successful shifts from purely human-operated tasks to automated solutions. These technologies enable rapid adoption due to existing infrastructures, suggesting a potential path for future innovations in various fields.

  • 00:10:00 - 00:15:00

    Further examples illustrate that complete automation can lead to significant benefits, such as reduced mistakes and enhanced efficiency in industries like aviation and pharmaceuticals. There are indications that fully automated processes can outperform systems that involve human supervision, highlighting the potential advantages of pursuing full automation as opposed to gradual implementation.

  • 00:15:00 - 00:20:00

    The speaker then explores why the automation cliff is preferable, noting that mixed systems with human oversight can lead to performance degradation and cognitive overload. The complexities involved in partial automation often necessitate a complete overhaul to achieve full automation effectively, avoiding the inefficiencies of constantly adapting to both human and automated systems.

  • 00:20:00 - 00:29:07

    Finally, the discussion shifts towards the potential for computer using agents and humanoid robots to transform various industries by fully automating tasks. The predicted timeline for the widespread integration of these technologies spans from initial deployment by 2025 to full commercial adoption by 2033. The speaker discusses resistance in traditional industries, emphasizing that as capabilities grow, the barriers to widespread automation will diminish.

显示更多

思维导图

视频问答

  • What is the Automation Cliff?

    The Automation Cliff is the concept of achieving full automation all at once rather than incremental improvements, which can lead to radical changes in efficiency.

  • What are drop-in technologies?

    Drop-in technologies are innovations that can replace existing processes or tools without requiring extensive changes to the existing infrastructure.

  • What examples are provided for successful full automation?

    Examples include autopilots in aircraft, pharmaceutical manufacturing, and automated harvesters.

  • Why is full automation preferable?

    Full automation avoids performance degradation due to human-machine handoffs and reduces cognitive load.

  • What industries are likely to see significant changes due to automation?

    Industries such as pharmaceuticals, agriculture, customer service, and many others are set to experience major transformations.

  • What is the predicted timeline for widespread automation adoption?

    The timeline predicts that widespread adoption could occur between 2025 and 2033, depending on the technology and industry.

  • Can jobs be eliminated by automation?

    Yes, many traditional jobs are at risk of being replaced or significantly altered by advancements in AI and robotics.

  • What are some examples of jobs that may not be safe from automation?

    Jobs in construction, emergency response, and even medical professions may face risks from automation.

  • What role do humanoid robots play in automation?

    Humanoid robots can operate in human environments and use human tools, making them suitable for many jobs currently held by humans.

  • What is the future of knowledge work according to the video?

    The video suggests that knowledge work as we know it may change dramatically, with robots and AI taking over many tasks that are currently performed by humans.

查看更多视频摘要

即时访问由人工智能支持的免费 YouTube 视频摘要！
字幕
en
自动滚动:
  • 00:00:00
    hello everyone I am really excited about
  • 00:00:02
    this video I've been planning it for a
  • 00:00:04
    while and I kept just adding to it as I
  • 00:00:06
    was doing more and more research uh
  • 00:00:08
    particularly because I was using deep
  • 00:00:10
    research to make it better and better so
  • 00:00:13
    let's Dive Right In uh I have talked
  • 00:00:15
    about the concept of the automation
  • 00:00:17
    Cliff for a while and uh I didn't come
  • 00:00:20
    up with this idea so what I wanted to do
  • 00:00:22
    was actually share some of my personal
  • 00:00:24
    experience but also some research as
  • 00:00:26
    well as some projections based on the
  • 00:00:28
    automation Cliff now uh basically this
  • 00:00:31
    is the idea of the automation cliff and
  • 00:00:34
    we're going to unpack this but just keep
  • 00:00:35
    this graph in mind where you've got the
  • 00:00:36
    stair step versus this more kind of
  • 00:00:40
    catastrophic plunge uh the tldr is that
  • 00:00:44
    if you focus on incremental improvements
  • 00:00:46
    in technology you'll end up with this
  • 00:00:49
    kind of gradual stairstep Improvement of
  • 00:00:51
    level of automation versus level of
  • 00:00:53
    human involvement so before automation
  • 00:00:56
    human involvement is high and then level
  • 00:00:58
    of of automation so you can think of
  • 00:01:00
    this as similar to like uh Tesla's
  • 00:01:02
    levels of FSD you know because right now
  • 00:01:04
    they're at like level three um but
  • 00:01:06
    ideally full self-driving like true
  • 00:01:09
    self-driving is going to be level five
  • 00:01:11
    which is zero human involvement needed
  • 00:01:13
    ever whereas level three is 99% of the
  • 00:01:17
    time it can work but it can't handle
  • 00:01:19
    edge cases and does still require human
  • 00:01:21
    intervention um I don't know the exact
  • 00:01:23
    definition of FSD level three but you
  • 00:01:24
    get the idea so right now most
  • 00:01:27
    Industries are using the stairstep
  • 00:01:29
    approach and we'll talk about why this
  • 00:01:31
    is um but in an Ideal World and in also
  • 00:01:34
    some cases you end up with more of this
  • 00:01:37
    kind of edifice approach where you just
  • 00:01:39
    have this escarpment that just you just
  • 00:01:42
    go careening off of hence the thumbnail
  • 00:01:44
    that you saw for this all right moving
  • 00:01:45
    on so what is the automation Cliff uh
  • 00:01:48
    basically the the principle of the
  • 00:01:50
    automation Cliff says that what you
  • 00:01:52
    should do is wait until you have the
  • 00:01:55
    full process automated end to endend and
  • 00:01:58
    then just full sent just send it off and
  • 00:02:01
    there you go um one of the key
  • 00:02:03
    principles here is tasks should be
  • 00:02:04
    controlled completely by either humans
  • 00:02:06
    or completely by automation systems with
  • 00:02:08
    no middle ground personally as an
  • 00:02:10
    automation engineer in my past life this
  • 00:02:14
    is what I would have advocated for and
  • 00:02:16
    the reason is because you know it does
  • 00:02:18
    nothing until you turn it on and you do
  • 00:02:21
    all the full testing and then you turn
  • 00:02:22
    it on and it automates everything all at
  • 00:02:24
    once that's what I mean by the
  • 00:02:26
    automation Cliff now there's lots and
  • 00:02:28
    lots of other principles in but there's
  • 00:02:31
    also like some problems with getting
  • 00:02:33
    there so let's start to unpack that uh
  • 00:02:35
    in just a moment but what I want to talk
  • 00:02:37
    about is drop in Technologies so let me
  • 00:02:40
    go back a couple slides and show you
  • 00:02:42
    what happens so when you have an
  • 00:02:44
    automation Cliff like this this usually
  • 00:02:46
    happens when you have what's called a
  • 00:02:47
    dropin technology and a dropin
  • 00:02:50
    technology basically means you know
  • 00:02:51
    here's the oldfashioned way of doing
  • 00:02:53
    things where it's 100% humans uh in the
  • 00:02:55
    loop and then you have this new brand
  • 00:02:57
    new technology that means you don't need
  • 00:02:59
    humans in the the loop at all whatsoever
  • 00:03:01
    so let's give you a couple of examples
  • 00:03:03
    about drop in Technologies now in some
  • 00:03:06
    cases these are not automations but
  • 00:03:08
    these are technologies that could just
  • 00:03:10
    come in and completely change everything
  • 00:03:12
    so first was USB um everyone is familiar
  • 00:03:15
    with USB when I was little and people my
  • 00:03:17
    age and people uh older were younger you
  • 00:03:20
    had serials and you had parallels and
  • 00:03:22
    you had all kinds of different
  • 00:03:23
    connections now everything is USB
  • 00:03:25
    Universal serial bus um Cloud
  • 00:03:28
    integration so SAS software is a service
  • 00:03:30
    is another example of where you can just
  • 00:03:32
    switch between softwares um and if you
  • 00:03:35
    get you know uh actually chat Bots are a
  • 00:03:38
    prime example you can go sign up for
  • 00:03:40
    Claude you can sign up for chat GPT you
  • 00:03:41
    can sign up for Gemini and most of them
  • 00:03:43
    are pretty interchangeable so those are
  • 00:03:45
    examples of drop-in Technologies or
  • 00:03:47
    fungible Technologies GPS was another
  • 00:03:50
    thing that once it was there it was
  • 00:03:51
    ubiquitous and now you can use it for
  • 00:03:52
    all kinds of stuff it enabled Google
  • 00:03:54
    Maps uh and you know your your Fitbit
  • 00:03:58
    and everything there's all kinds of
  • 00:04:00
    different cascading effects uh smart
  • 00:04:02
    retrofitting of buildings and other
  • 00:04:04
    infrastructure so an example of a drop
  • 00:04:06
    in technology was dialup modems which
  • 00:04:08
    used that was that's probably honestly
  • 00:04:11
    the best example because you started
  • 00:04:13
    with existing phone lines and then you
  • 00:04:15
    said well let's make them digital so
  • 00:04:16
    then you can just have modems call each
  • 00:04:18
    other and exchange uh information that
  • 00:04:21
    way and that was really kind of most
  • 00:04:23
    people's first uh experience of the
  • 00:04:25
    internet and then we did the same thing
  • 00:04:27
    with cable uh cable modems were just
  • 00:04:29
    using the fatter uh bandwidth that uh
  • 00:04:32
    the digital cable had uh available uh
  • 00:04:35
    and then streaming media is another
  • 00:04:36
    example you know you can swap between
  • 00:04:37
    Netflix and Disney and all those other
  • 00:04:39
    things so these are examples of drop in
  • 00:04:41
    technologies that once you have enough
  • 00:04:42
    infrastructure you can just put in a new
  • 00:04:44
    technology and you can adopt it very
  • 00:04:46
    quickly personally uh on an individual
  • 00:04:49
    consumer level you can adopt it almost
  • 00:04:51
    instantly overnight um some of these
  • 00:04:53
    Technologies do take longer for larger
  • 00:04:55
    organizations to adopt just because
  • 00:04:57
    there's a lot of inertia in those
  • 00:04:58
    organizations so these are a few
  • 00:05:00
    examples of the fact that we have had
  • 00:05:01
    dropin Technologies um that allowed for
  • 00:05:04
    that kind of that almost saltatory leap
  • 00:05:07
    that very Square wave um of adoption not
  • 00:05:10
    all of it goes that way so um before we
  • 00:05:14
    uh get a little bit further I want to
  • 00:05:15
    talk uh also about examples of where
  • 00:05:17
    full automation does tend to be superior
  • 00:05:21
    so if you can achieve full automation if
  • 00:05:23
    you can achieve that full automation
  • 00:05:25
    Cliff that is going to be more desirable
  • 00:05:27
    so one example is um autopilots so
  • 00:05:30
    originally autopilots would basically
  • 00:05:31
    just maintain your altitude and speed
  • 00:05:33
    and a little bit more but now uh as
  • 00:05:35
    airplanes are more and more
  • 00:05:36
    sophisticated autopilots basically there
  • 00:05:39
    are stories of Pilots just going to
  • 00:05:41
    sleep with the autopilot on uh meaning
  • 00:05:43
    that they run 100% of the aircraft um
  • 00:05:46
    another example is pharmaceutical
  • 00:05:47
    production so these by the way these are
  • 00:05:49
    all numbers that were surfaced using
  • 00:05:51
    deep research um so yeah maybe I should
  • 00:05:54
    start including the links to those deep
  • 00:05:56
    research articles as uh as as evidence
  • 00:06:00
    anyways let me know what you think in
  • 00:06:01
    the comments so uh having supported and
  • 00:06:05
    worked with and consulted for and talked
  • 00:06:07
    to people in the Pharmaceuticals
  • 00:06:08
    Industries um the pharmaceutical
  • 00:06:10
    industry is one of the most heavily
  • 00:06:12
    automated Industries out there
  • 00:06:14
    particularly with the actual production
  • 00:06:17
    of drugs um and when you look at what
  • 00:06:20
    when we talk about lights out
  • 00:06:21
    manufacturing lights out manufacturing
  • 00:06:23
    basically means no humans need to be
  • 00:06:25
    present or even observing um and so that
  • 00:06:28
    that took the effect rate from 0.1% to
  • 00:06:34
    0.001% um so in in other words keeping
  • 00:06:37
    human supervisors was actually a net
  • 00:06:40
    negative it was actually better to get
  • 00:06:42
    humans completely out of the loop um
  • 00:06:44
    here's another example is automated
  • 00:06:45
    Harvesters so John Deere so these are
  • 00:06:47
    like the big combines that you see like
  • 00:06:49
    that you know go over fields and harvest
  • 00:06:51
    everything um they're fully autonomous
  • 00:06:54
    combines uh reduced yield loss from 15%
  • 00:06:57
    to 2.3% by eliminating operator fatigue
  • 00:07:00
    and operator errors basically you know
  • 00:07:02
    humans make mistakes and if you're
  • 00:07:03
    driving a tractor for 10 hours a day you
  • 00:07:06
    get kind of bored so on and so forth you
  • 00:07:08
    get the idea we don't need to go through
  • 00:07:09
    every single example but these are some
  • 00:07:12
    these you know between uh autopilots
  • 00:07:14
    pharmaceutical manufacturing and uh
  • 00:07:17
    harvesting you can see that there are
  • 00:07:19
    several examples across several
  • 00:07:21
    different domains where full automation
  • 00:07:23
    is actually preferable if you can
  • 00:07:25
    achieve it so moving on um now one
  • 00:07:29
    question you might might be wondering is
  • 00:07:30
    like okay well why why why is the
  • 00:07:31
    automation Cliff preferable um if if you
  • 00:07:35
    know you could just gradually uh
  • 00:07:37
    Implement things number one is the uh
  • 00:07:40
    performance degradation of handoffs so
  • 00:07:44
    you know you see videos of people you
  • 00:07:46
    know driving their Tesla and it's like
  • 00:07:47
    oh you know you're distracted and then
  • 00:07:49
    you have to intervene um that's one
  • 00:07:51
    example now what I will say is as a
  • 00:07:53
    counter example to that is that the is
  • 00:07:56
    that splitting your cognitive attention
  • 00:07:57
    with using tools like deep research it's
  • 00:07:59
    like oh here go do go do a research uh
  • 00:08:02
    topic for me briefly and then I'll come
  • 00:08:04
    back in 5 minutes and that actually
  • 00:08:06
    gives your brain a CH a chance to rest
  • 00:08:08
    there's actually a brief story that I
  • 00:08:09
    have where um uh at a software company I
  • 00:08:12
    worked at gosh 2012 2011 2012 so that
  • 00:08:15
    was a long time ago um I built out their
  • 00:08:18
    their uh virtual infrastructure and we
  • 00:08:20
    built more build servers for them and
  • 00:08:22
    their build process went from 24 hours
  • 00:08:24
    to 2 hours and they were like Dave can
  • 00:08:26
    you take those servers out they were
  • 00:08:28
    kind of joking but they're like can you
  • 00:08:29
    remove those servers because you know uh
  • 00:08:32
    we we we now have less time to actually
  • 00:08:37
    you know work on our after action
  • 00:08:39
    reports we're used to we're used to the
  • 00:08:40
    build process taking 24 hours so then we
  • 00:08:43
    have a full day to keep working um I was
  • 00:08:46
    like I'm not going to do that like you
  • 00:08:48
    you wanted me to make things faster I
  • 00:08:49
    made things faster by a factor of 12
  • 00:08:51
    deal with it so anyways um you have
  • 00:08:54
    trust issues workload problems
  • 00:08:56
    monitoring partial automation can uh
  • 00:08:58
    increase the C itive load which that's
  • 00:09:01
    particularly true if you're monitoring
  • 00:09:02
    different automation stations um so that
  • 00:09:06
    can and and if information is coming at
  • 00:09:08
    you faster that will tire tire you out
  • 00:09:10
    much much more quickly uh and those
  • 00:09:12
    sorts of things so in many cases if it
  • 00:09:14
    is possible then you want to use the
  • 00:09:16
    full automation Cliff you want to go
  • 00:09:18
    straight from the current way of doing
  • 00:09:20
    things to the new way of doing things
  • 00:09:22
    without much um without much
  • 00:09:24
    interstitial time another reason is
  • 00:09:26
    because you don't want to keep
  • 00:09:27
    Reinventing the wheel um that was
  • 00:09:28
    something that I include in this slide
  • 00:09:30
    but basically every time you have to
  • 00:09:31
    reinvent the infrastructure or Implement
  • 00:09:33
    new infrastructure that handles you know
  • 00:09:36
    human affordances and partial Automation
  • 00:09:38
    and then you have to do it again to get
  • 00:09:40
    to full automation often it's better to
  • 00:09:42
    just wait and then implement the full
  • 00:09:44
    automation all at
  • 00:09:46
    once um and so this is this is talking a
  • 00:09:49
    little bit more this slide is we talk a
  • 00:09:51
    little bit more about how um in reality
  • 00:09:55
    usually full automation is just not an
  • 00:09:57
    option so this is one of the things that
  • 00:09:58
    we're going to be talking about with the
  • 00:10:00
    rise of agents and robots so um also my
  • 00:10:04
    dog is under the desk so if I seem
  • 00:10:06
    distracted I'm petting my dog um she'll
  • 00:10:09
    make a guest appearance one day um so
  • 00:10:12
    first and foremost uh is the economic
  • 00:10:14
    barriers so full end in automation can
  • 00:10:16
    often be very expensive and as many of
  • 00:10:18
    you have pointed out in the comments and
  • 00:10:19
    on Twitter and other places the first
  • 00:10:21
    90% is usually actually really easy it's
  • 00:10:24
    the last mile of automation that is
  • 00:10:26
    really hard and that's where 90 to 99%
  • 00:10:29
    of of your automation effort will go
  • 00:10:31
    into is uh what was it one of you said
  • 00:10:33
    something like you know when in in the
  • 00:10:35
    space of automation you realize that
  • 00:10:36
    everything is edge cases and that's
  • 00:10:39
    that's not a bad way of thinking about
  • 00:10:40
    it is because yes 90 to 99% of what
  • 00:10:43
    you're doing is routine robust uh or not
  • 00:10:47
    not robust um uh routine but or brain
  • 00:10:50
    dead simple I don't know what word I was
  • 00:10:51
    thinking of um but it's it's uh it's
  • 00:10:54
    it's very repetitive maybe that's the
  • 00:10:55
    word I was thinking of it's routine and
  • 00:10:57
    repetitive but then you do need that
  • 00:10:59
    level of high level adaptation for every
  • 00:11:02
    single exception every single edge case
  • 00:11:03
    and those sorts of things and that is
  • 00:11:06
    the technical complexity where it's like
  • 00:11:07
    some things it's just too complex to
  • 00:11:09
    automate unless or until you get to a
  • 00:11:12
    general purpose general intelligence
  • 00:11:14
    whether it's a computer using agent or a
  • 00:11:17
    robot um which is going to be more
  • 00:11:19
    cognitively flexible than a human then
  • 00:11:21
    the technical complexity is no longer a
  • 00:11:23
    barrier that has honestly been the
  • 00:11:25
    biggest barrier to automation up to this
  • 00:11:27
    point but with the rise of generative AI
  • 00:11:29
    language models and cognitive
  • 00:11:31
    architectures that's no longer going to
  • 00:11:33
    be a barrier um risk management resource
  • 00:11:36
    constraints you can imagine how all of
  • 00:11:38
    these things play out but really it's
  • 00:11:40
    the economics and the technical
  • 00:11:41
    complexity are the two biggest barriers
  • 00:11:43
    or constraints to full automation um but
  • 00:11:46
    as robots you know become more
  • 00:11:48
    ubiquitous and become more intelligent
  • 00:11:50
    and as computer using agents also become
  • 00:11:52
    more ubiquitous more robust and more
  • 00:11:54
    intelligent those barriers are going to
  • 00:11:56
    disappear very quickly um so speak
  • 00:11:59
    speaking of barriers and adoption rates
  • 00:12:01
    one of the things that I have been
  • 00:12:02
    pointing out to people is that
  • 00:12:04
    technology adoption rates have been
  • 00:12:05
    accelerating so the automobile took a
  • 00:12:08
    long time to reach a point of saturation
  • 00:12:09
    oh and by the way this graph is a little
  • 00:12:11
    bit dated because you know the internet
  • 00:12:13
    has been around for more than 10 years
  • 00:12:14
    now um so the data in this graph is a
  • 00:12:17
    little bit dated but you get the point
  • 00:12:19
    where the television once it got cheap
  • 00:12:21
    enough it took off really fast
  • 00:12:23
    electricity took off really fast but
  • 00:12:25
    these are things that took you know
  • 00:12:27
    decades to a century to get fully
  • 00:12:29
    adopted but then mobile phones PCS
  • 00:12:32
    internet everything is getting adopted
  • 00:12:34
    much much faster here you're talking
  • 00:12:36
    about adoption curves that are in
  • 00:12:38
    measured in the 10 to 20 years um now
  • 00:12:41
    that uh that the internet has reached a
  • 00:12:43
    certain level of saturation anything
  • 00:12:45
    that can be delivered on the internet
  • 00:12:46
    gets adopted much much faster and that
  • 00:12:49
    includes artificial intelligence such as
  • 00:12:52
    uh chat Bots and those sorts of things
  • 00:12:54
    robots require a lot of infrastructure
  • 00:12:56
    to be built out um then you have to ship
  • 00:12:58
    the robots and those sorts of things so
  • 00:13:00
    because there's a physical layer to the
  • 00:13:02
    robots there's going to be a little bit
  • 00:13:03
    more friction but on the other hand
  • 00:13:05
    robots that are in humanoid shape are a
  • 00:13:07
    perfect dropin technology so uh before
  • 00:13:10
    we move on I want to point out my uh sub
  • 00:13:13
    uh not substack my link tree real quick
  • 00:13:14
    which has my substack um it's got my
  • 00:13:17
    patreon my school Community this is my
  • 00:13:19
    learning community I update uh two to
  • 00:13:21
    three lessons per week over there on
  • 00:13:22
    patreon we have an exclusive Discord I'm
  • 00:13:24
    also on substack Twitter um I also just
  • 00:13:27
    added my SoundCloud to uh to my link
  • 00:13:29
    tree which is where I put all my AI
  • 00:13:31
    generated music um it's not for everyone
  • 00:13:34
    but I listen to my own music a lot um so
  • 00:13:36
    if you're into psychedelic Space Rock
  • 00:13:38
    I've got a lot of it up there um also
  • 00:13:40
    I'm on GitHub and Spotify and a few
  • 00:13:42
    other things so go check it out all
  • 00:13:43
    right back to the
  • 00:13:44
    show now um what this slide is talking
  • 00:13:48
    about like where are we actually trying
  • 00:13:50
    to automate things so I I what I did was
  • 00:13:53
    I had deep research say go find the
  • 00:13:56
    problems that people are trying to
  • 00:13:57
    automate today right right now with
  • 00:13:59
    generative Ai and Robotics so here are
  • 00:14:02
    the examples that it came up with number
  • 00:14:03
    one is contact centers so we've all by
  • 00:14:06
    now probably heard some of the stories
  • 00:14:07
    where call centers have had some of
  • 00:14:09
    their their Staffing reduced by 90%
  • 00:14:12
    there's also been some stories of
  • 00:14:13
    they've had to rehire Some Humans for
  • 00:14:15
    all those edge cases that we were
  • 00:14:16
    talking about but at the same time a lot
  • 00:14:19
    of those call centers that have switched
  • 00:14:21
    to fully or mostly AI um the cat scores
  • 00:14:24
    also go up and cat is customer
  • 00:14:26
    satisfaction so that's that's MBA jargon
  • 00:14:29
    for how happy are your customers in many
  • 00:14:31
    cases if you go to full automation the
  • 00:14:34
    customers get happier because then uh
  • 00:14:36
    there the quality of their service goes
  • 00:14:38
    up and they have more faith in your
  • 00:14:39
    service or your company or your product
  • 00:14:42
    um so however with that being said you
  • 00:14:44
    know if a call center can only get rid
  • 00:14:46
    of 90% of its people but it still needs
  • 00:14:48
    10 for those edge cases that's not full
  • 00:14:50
    automation um and furthermore there's
  • 00:14:52
    plenty of other kinds of call centers
  • 00:14:54
    that you just cannot fully automate away
  • 00:14:56
    yet that is still a very high Target
  • 00:14:59
    taret because that's what we would call
  • 00:15:00
    low hanging fruit uh another example is
  • 00:15:02
    retail checkout um so in uh for instance
  • 00:15:06
    if you've ever gone to those self
  • 00:15:07
    checkouts um those self checkouts
  • 00:15:10
    sometimes they break or so on and so
  • 00:15:11
    forth sometimes theft also goes up
  • 00:15:13
    because it's like you have a self
  • 00:15:14
    checkout but then you have like a human
  • 00:15:16
    supervising and then but the human gets
  • 00:15:18
    bored and stuff still gets stolen and
  • 00:15:20
    yada yada yada so then you need more
  • 00:15:22
    computer vision for the security and
  • 00:15:25
    yeah and so you end up with all these
  • 00:15:27
    other what about what about what about
  • 00:15:29
    uh kinds of things that make full
  • 00:15:30
    automation of the checkout uh a little
  • 00:15:33
    bit harder uh Warehouse robotics uh this
  • 00:15:36
    is another example so you've probably
  • 00:15:38
    seen some of the videos of like the
  • 00:15:39
    Amazon robots where it's like it there
  • 00:15:43
    Amazon has warehouses that are not human
  • 00:15:45
    navigable anymore um at the same time
  • 00:15:48
    sometimes those systems still get gummed
  • 00:15:49
    up because of a complex emergent
  • 00:15:51
    behavior that happens when you have
  • 00:15:53
    hundreds and hundreds of uh item
  • 00:15:55
    fetching robots and they get you know
  • 00:15:56
    all uh jammed up I don't mean physically
  • 00:15:59
    jammed up I mean you know like the
  • 00:16:00
    traffic gets congested and so on and so
  • 00:16:03
    forth so these are these are current
  • 00:16:05
    challenges that we have not yet solved
  • 00:16:07
    and it's like okay well if we can't
  • 00:16:08
    fully automate call centers and Retail
  • 00:16:10
    checkout and warehouses then clearly
  • 00:16:14
    like a lot of jobs are still safe
  • 00:16:15
    however keep in mind that as robots get
  • 00:16:17
    more intelligent every every step of
  • 00:16:20
    intelligence they that they gain and
  • 00:16:22
    this also includes computer using agents
  • 00:16:24
    that dramatically expands what they can
  • 00:16:27
    do without human intervention so so
  • 00:16:29
    you're going to see some of these leaps
  • 00:16:30
    some of these um some of these sigmoid
  • 00:16:32
    curves or these step functions where
  • 00:16:34
    you're going to have new abilities that
  • 00:16:36
    are going to just say oh all that stuff
  • 00:16:38
    that we couldn't automate a year ago we
  • 00:16:40
    can automate all of it now and I have
  • 00:16:42
    seen that personally back in my back in
  • 00:16:44
    my corporate days I've also seen it in
  • 00:16:46
    some of the clients that I've consulted
  • 00:16:48
    for where there are things that you can
  • 00:16:49
    automate today that a lot of people
  • 00:16:50
    don't even believe that you can automate
  • 00:16:52
    and that's one of the reasons that I
  • 00:16:53
    make these videos is to say hey the
  • 00:16:55
    thing that you think that you can't
  • 00:16:56
    automate maybe you actually can
  • 00:16:59
    so moving on um now I've talked about
  • 00:17:02
    humanoid robots on this channel quite a
  • 00:17:04
    bit but I want to talk about how this is
  • 00:17:05
    really the ultimate drop in solution so
  • 00:17:08
    one of the key things is that humanoid
  • 00:17:09
    robots can operate in human spaces using
  • 00:17:12
    human tools human vehicles and uh pretty
  • 00:17:15
    much everything else so if you have a
  • 00:17:17
    human robot that is as smart as or
  • 00:17:20
    honestly if you put you know gp4 or gp5
  • 00:17:23
    in it or you know clae 4 whatever
  • 00:17:25
    whatever model comes out then it's going
  • 00:17:28
    to be smarter than the vast majority of
  • 00:17:29
    humans already then if you have watched
  • 00:17:32
    the Boston Dynamics videos where those
  • 00:17:34
    robots are far more agile than humans
  • 00:17:36
    they can do standing back flips I cannot
  • 00:17:38
    do a standing backflip so they're
  • 00:17:40
    stronger they're smarter they're faster
  • 00:17:42
    they're going to have more dexterity
  • 00:17:43
    than humans that means that it is a
  • 00:17:45
    perfect drop in solution which means
  • 00:17:47
    that basically any job that a human does
  • 00:17:49
    with their hands and eyes and body sorry
  • 00:17:51
    hit the microphone um these robots will
  • 00:17:54
    be able to do very soon and those that
  • 00:17:58
    general purpose form function means that
  • 00:17:59
    it can even sit in front of a computer
  • 00:18:01
    and use a keyboard and mouse if it needs
  • 00:18:03
    to um but we can use computer using
  • 00:18:06
    agents for that so you can just remove
  • 00:18:07
    the whole robot entirely um so this this
  • 00:18:11
    represents a full automation solution
  • 00:18:14
    and this is what I mean by the
  • 00:18:15
    automation Cliff once you start shipping
  • 00:18:18
    you know super intelligent super strong
  • 00:18:19
    super dextrous super agile robots it's
  • 00:18:22
    like game over for 90% of human jobs
  • 00:18:25
    next is the computer using agents so the
  • 00:18:27
    computer using agents are what you've
  • 00:18:28
    seen like um uh operator and repet and
  • 00:18:32
    all those other different tools out
  • 00:18:34
    there um what you need to think of and I
  • 00:18:37
    still have people saying Dave why don't
  • 00:18:38
    we just focus on apis and so for those
  • 00:18:41
    that aren't familiar an API is an
  • 00:18:42
    application programming interface which
  • 00:18:44
    is basically allows one computer program
  • 00:18:46
    to call and talk directly to another
  • 00:18:48
    computer program with without any other
  • 00:18:50
    user interface but keyboard video Mouse
  • 00:18:53
    KVM is the universal API furthermore
  • 00:18:57
    think about how the vast majority of
  • 00:18:59
    what humans do also my dogs are
  • 00:19:00
    wrestling in the background so if you do
  • 00:19:01
    hear that I apologize um so the vast
  • 00:19:05
    majority of human knowledge work is done
  • 00:19:07
    with KVM keyboard video Mouse if you can
  • 00:19:10
    do it with KVM and an operator can do it
  • 00:19:12
    with KVM that's a universal UI that's a
  • 00:19:15
    universal interface that you don't need
  • 00:19:17
    any other infrastructure for you don't
  • 00:19:19
    need custom apis you don't need custom
  • 00:19:21
    API discoveries that is the API the KVM
  • 00:19:25
    is the universal API and so what that
  • 00:19:27
    means is is that instead of even having
  • 00:19:30
    a robot using the computer you just drop
  • 00:19:32
    that agent onto any computer or servers
  • 00:19:34
    and they can be virtual servers by the
  • 00:19:35
    way and you have literally the
  • 00:19:38
    equivalent of hundreds thousands
  • 00:19:40
    millions of of employees all using you
  • 00:19:43
    know their own own laptop screen
  • 00:19:45
    basically but on a virtual server in the
  • 00:19:47
    cloud somewhere um that's really what
  • 00:19:50
    we're heading towards and the roll out
  • 00:19:51
    of this so this is this ties back to
  • 00:19:53
    that um what I said about you know the
  • 00:19:55
    adoption of cloud services um it's going
  • 00:19:58
    over the Internet so that means it's
  • 00:20:00
    really really fast to roll out um now
  • 00:20:04
    here's my personal timeline so this is
  • 00:20:07
    the automation wave optimistic timeline
  • 00:20:10
    um and this is based on the kind of
  • 00:20:13
    seven-year time Horizon and the seven
  • 00:20:16
    years is basically about how long it
  • 00:20:18
    took for companies to adopt
  • 00:20:20
    virtualization uh which was my area of
  • 00:20:22
    specialty as well as Cloud software uh
  • 00:20:24
    or software as a service which was uh
  • 00:20:26
    adjacent to what I was doing and when
  • 00:20:28
    you think that computer using agents are
  • 00:20:30
    basically virtualization and Cloud
  • 00:20:33
    software and it took seven years to
  • 00:20:35
    adopt those then we're looking at about
  • 00:20:37
    seven years for full commercial adoption
  • 00:20:40
    from this year because this year is when
  • 00:20:41
    we're first uh deploying agents so
  • 00:20:44
    initial launch is 2025 computer using
  • 00:20:47
    agents begin deployment um and not and
  • 00:20:49
    digital knowledge work and humanoid
  • 00:20:51
    robot uh humanoid robots are being
  • 00:20:53
    ramped up this year as well Mass
  • 00:20:55
    adoption happens 2026 and 2027 um so
  • 00:20:59
    this is when Fortune 500 companies are
  • 00:21:00
    going to really start using both
  • 00:21:02
    computer using agents and humanoid
  • 00:21:04
    robots um in Mass there are Fortune 500
  • 00:21:07
    companies already using Tesla Optimus
  • 00:21:09
    and other robots just want to point that
  • 00:21:11
    out I think BMW was the first car
  • 00:21:13
    company that started using them other
  • 00:21:15
    than Tesla of course um then so that's
  • 00:21:18
    the that's the uh early early Mass
  • 00:21:21
    adoption and then you're going to have
  • 00:21:23
    full integration happening in 2028 to
  • 00:21:25
    2030 and then you're going to have the
  • 00:21:28
    the fin laggards the the the
  • 00:21:31
    optimization happening in the 2031 to
  • 00:21:33
    2032 range and then by 2033 you're going
  • 00:21:37
    to have offices full of robots and
  • 00:21:40
    computer using agents and all that fun
  • 00:21:42
    stuff that's my personal prediction is
  • 00:21:43
    that we're looking at seven years until
  • 00:21:47
    you know knowledge work as we know it is
  • 00:21:49
    over and done with in every industry um
  • 00:21:52
    now I want to use this graph so this
  • 00:21:54
    this graph is the adoption curve um so
  • 00:21:57
    this is like a very similar
  • 00:21:59
    version to the other adoption curve that
  • 00:22:01
    I showed you so this is this is a linear
  • 00:22:04
    adoption curve which is just at what
  • 00:22:06
    point does the technology become
  • 00:22:07
    saturated but another way to look at the
  • 00:22:09
    adoption curve is this which is at what
  • 00:22:12
    point does each um each type of company
  • 00:22:14
    adopted so right now or or I guess 2024
  • 00:22:18
    and earlier were the innovators so these
  • 00:22:20
    are all the people that you know watched
  • 00:22:22
    my YouTube channel since 2022 2023 these
  • 00:22:25
    are the people that have been
  • 00:22:26
    experimenting with cognitive
  • 00:22:27
    architectures and agents
  • 00:22:29
    since you know before chat GPT came out
  • 00:22:31
    or when chat GPT first came out think
  • 00:22:33
    about back to the era of baby AGI and
  • 00:22:35
    those sorts of things that was the
  • 00:22:37
    innovators so that was the bleeding edge
  • 00:22:38
    innovators that was the first
  • 00:22:40
    2.5% this year and 2026 are going to be
  • 00:22:43
    the early adopter so this is where uh
  • 00:22:47
    this is where all the first movers are
  • 00:22:48
    saying okay there's actual commercial
  • 00:22:50
    value here let's pull the trigger then
  • 00:22:53
    20 2027 to 2028 is going to be the early
  • 00:22:57
    majority this is where you know your
  • 00:23:00
    your mom and pop shops may maybe maybe
  • 00:23:02
    not you know your bakery but what I mean
  • 00:23:04
    is you know your average run of the mill
  • 00:23:06
    companies are going to start adopting
  • 00:23:08
    some of these Technologies you know U I
  • 00:23:11
    know lawyers and law firms that are
  • 00:23:12
    already using some of these AI tools um
  • 00:23:15
    uh but they're they're still kind of the
  • 00:23:16
    early adopters so then the majority of
  • 00:23:18
    law firms and doctor's offices and those
  • 00:23:20
    sorts of things will start adopting then
  • 00:23:22
    and then you'll have the uh the group of
  • 00:23:24
    people in the late majority so these are
  • 00:23:26
    the more Skeptics these are the more uh
  • 00:23:28
    mortar kind of stores so like you know I
  • 00:23:31
    would imagine that like Home Depot
  • 00:23:33
    they'll probably be a little bit later
  • 00:23:34
    to adopt these just because that
  • 00:23:36
    business model hasn't changed in more
  • 00:23:38
    than a century you know it's like you
  • 00:23:40
    have hardware and tools and you sell
  • 00:23:42
    Hardwares and tools to real people in
  • 00:23:45
    front of you um so some businesses some
  • 00:23:48
    Industries are going to be a little bit
  • 00:23:49
    more resistant to it um rather than
  • 00:23:52
    people that are going to be more on the
  • 00:23:53
    front end now heavy Industries like
  • 00:23:55
    Mining and construction they will
  • 00:23:57
    probably be in the early majority if I
  • 00:23:59
    had to guess just because human labor is
  • 00:24:01
    really expensive and loss of life and
  • 00:24:04
    injury is also really expensive but if a
  • 00:24:06
    robot gets crushed under a rockfall
  • 00:24:08
    that's just a tax write off you can't
  • 00:24:10
    write off human lives uh sorry that's
  • 00:24:12
    not how it works and then 2030 plus this
  • 00:24:15
    is going to be as the rest of the world
  • 00:24:17
    catches up so this is my preferred
  • 00:24:20
    timeline now I asked deep research to
  • 00:24:24
    take all of this into account and make
  • 00:24:26
    its own timeline and it gave a much more
  • 00:24:28
    conservative timeline so it's its
  • 00:24:31
    timeline based on historical evidence
  • 00:24:32
    and those longer adoption curves which
  • 00:24:34
    we saw earlier says that the initial
  • 00:24:36
    wave will be 2025 to 2030 um so I I I
  • 00:24:42
    need to emphasize this is not my
  • 00:24:43
    personal timeline I'm just showing you
  • 00:24:45
    what the AI said as a as a more
  • 00:24:47
    conservative or realistic timeline so
  • 00:24:50
    2025 to 2030 this is when we're going to
  • 00:24:52
    see digital knowledge work uh get
  • 00:24:54
    replaced the early majority won't be
  • 00:24:56
    till 2030 to 2035 again
  • 00:24:59
    I don't believe
  • 00:25:00
    that um service integration so this is
  • 00:25:03
    where you start to see kind of the the
  • 00:25:04
    early uh early and late majority so some
  • 00:25:07
    of the more uh resistive s uh uh uh
  • 00:25:10
    Industries so like healthc care is a
  • 00:25:12
    very resistant industry education very
  • 00:25:13
    resistant industry you're it doesn't
  • 00:25:16
    expect that we're going to see full
  • 00:25:17
    automation there for the next 10 to 15
  • 00:25:20
    years again I this will not age well um
  • 00:25:24
    and then by then there will be enough
  • 00:25:26
    regulatory pressure on States and
  • 00:25:28
    federal governments to say okay we need
  • 00:25:30
    to do things differently and that's 15
  • 00:25:32
    to 20 years out now keep in mind that
  • 00:25:34
    2045 is like Singularity so if teachers
  • 00:25:37
    unions are still preventing AI in the
  • 00:25:39
    classroom when Singularity hits oh boy
  • 00:25:41
    are they in for a roote Awakening
  • 00:25:43
    anyways like I said I this this timeline
  • 00:25:47
    is way too conservative for me but I I
  • 00:25:49
    felt like just for the sake of argument
  • 00:25:51
    I had to put this is what the AI thinks
  • 00:25:52
    the timeline is going to be um now what
  • 00:25:56
    I do predict is that as computer using
  • 00:25:58
    agents and robots ramp up in terms of
  • 00:26:00
    intelligence and ubiquity we are going
  • 00:26:02
    to see total Workforce automation as we
  • 00:26:04
    understand it today now we can talk
  • 00:26:05
    about post labor economics there's there
  • 00:26:07
    will be some kinds of jobs like
  • 00:26:08
    influencers I hope will stick around um
  • 00:26:11
    entertainers will probably stick around
  • 00:26:13
    like musicians and stuff there might be
  • 00:26:15
    entirely new classes of jobs there
  • 00:26:16
    probably will but the vast majority of
  • 00:26:19
    economic uh activity will not be done by
  • 00:26:21
    humans in the near future so you look at
  • 00:26:24
    Medical Precision superhuman surgical
  • 00:26:27
    robots with perfect Steady Hand hands or
  • 00:26:28
    multiple hands um combined with computer
  • 00:26:31
    using agents that are constantly
  • 00:26:32
    researching the best medical procedures
  • 00:26:35
    you will not have a human doctor you
  • 00:26:37
    will not want a human doctor in this
  • 00:26:39
    potential world next is construction a
  • 00:26:41
    lot of people say oh well I'm a boiler
  • 00:26:43
    maker or I'm a welder and Y yada yada
  • 00:26:45
    and my job safe no it isn't um consider
  • 00:26:48
    that robot that industrial robots
  • 00:26:50
    already do better Precision welds than
  • 00:26:52
    humans do the only difference between
  • 00:26:54
    like those Factory line welders and a
  • 00:26:57
    human welder is is that the human is in
  • 00:26:58
    a form factor that is more mobile um
  • 00:27:01
    that's not an advantage in the long run
  • 00:27:03
    electricians plumbers construction
  • 00:27:05
    workers uh welders you guys like you're
  • 00:27:09
    on notice I'm I'm telling you I'm I'm
  • 00:27:10
    trying to warn you ahead of time um that
  • 00:27:13
    that job is probably going away next is
  • 00:27:16
    emergency response so this is everything
  • 00:27:18
    from um uh emergency medical technicians
  • 00:27:21
    to Firefighters to even police um or or
  • 00:27:25
    whatever like all kinds of emergency
  • 00:27:27
    respons you take the human out of the
  • 00:27:29
    loop they you know you have machines
  • 00:27:31
    that are immune to smoke heat biological
  • 00:27:33
    radi radiological chemical attacks
  • 00:27:35
    whatever like you know there was um
  • 00:27:38
    there was a movie called surrogates
  • 00:27:40
    which was a really cool movie it didn't
  • 00:27:42
    make that bit much at the box office but
  • 00:27:43
    it's a Bruce Willis movie and one of the
  • 00:27:45
    scenes in that movie was really cool
  • 00:27:47
    where there's like a bunch of soldiers
  • 00:27:48
    and they're like all kids like playing
  • 00:27:50
    VR but they're piloting little humanoid
  • 00:27:52
    robots across a battlefield um and it's
  • 00:27:54
    just like oh you know robot gets you
  • 00:27:56
    know nuked and you know the person's
  • 00:27:58
    like ah darn and they you know spawn up
  • 00:28:01
    into another robot and to them it's just
  • 00:28:02
    a game um science science and
  • 00:28:05
    engineering that would you know I don't
  • 00:28:08
    think I have to really sell this for my
  • 00:28:09
    audience because you guys are like
  • 00:28:10
    paying attention to The Cutting Edge of
  • 00:28:11
    like Alpha fold and all that fun stuff
  • 00:28:13
    but you know we have like somewhere
  • 00:28:16
    between eight and 25 million scientists
  • 00:28:19
    uh you know phds uh or doctorates
  • 00:28:22
    globally right now we're going to have
  • 00:28:23
    the equivalent of billions or trillions
  • 00:28:25
    here real soon um and so therefore the
  • 00:28:27
    vast majority of scientific research
  • 00:28:28
    will be automated you combine Those
  • 00:28:30
    computer using agents those digital
  • 00:28:32
    agents um or those narrow AIS with
  • 00:28:34
    robots and you won't even need humans in
  • 00:28:36
    the loop if you don't want it now
  • 00:28:38
    obviously you still want humans saying
  • 00:28:39
    hey you know hey Mr Robot maybe stop
  • 00:28:41
    making VX gas we don't want you to make
  • 00:28:43
    that because that's really dangerous for
  • 00:28:45
    us but you get the idea and then finally
  • 00:28:48
    um you know uh government um
  • 00:28:51
    particularly if AI is provisioned uh of
  • 00:28:53
    the people for the people and by the
  • 00:28:54
    people um and the AI is is is directly
  • 00:28:57
    accountable to the people then what role
  • 00:29:00
    does elected politicians play
  • 00:29:02
    anymore I don't know so anyways thanks
  • 00:29:05
    for watching I hope you got a lot out of
  • 00:29:06
    this cheers
标签
  • Automation Cliff
  • Drop-in Technologies
  • Full Automation
  • AI
  • Robotics
  • Humanoid Robots
  • Technology Adoption
  • Pharmaceutical Industry
  • Agriculture Automation
  • Future of Work