คำบรรยาย (284)
0:00If you were in Las Vegas, in January 2023,
sitting in the audience of the opening keynote
0:05of CES—perhaps the most influential tech
conference in the world—you likely had one
0:10question on your mind: “Why is John Deere here?”
After all, they’re the tractor company,
0:15right? They make machines that push and
pull and move and dig, that’s their thing,
0:21right? Well, not according to their CEO, John May.
“We’ve quickly become one of the world’s leading
0:28robotics and AI companies. Our solutions
leverage technology like computer vision,
0:35advanced sensing and compute, machine
learning, and data analytics.”
0:40There’s one key word there: solutions. That’s a
word that gets thrown around a lot by companies
0:46like Apple “…solution…” “…solution…” “…
solution…”; Microsoft “…we provide an
0:54end-to-end tooling solution…” “…deceptively simple
solution…” “…the best end to end solution…”;
1:01Google “…bespoke AI solutions…” “open
sourcing solutions…” “…a great solution…”.
1:07This linguistic mimicry, their mere attendance
at an event like CES, it’s all… peculiar for a
1:14tractor company. Unless, does John Deere
think it’s a tech company? Well… yes,
1:20they do. At least according to themselves, in
this LinkedIn post, sharing an article entitled
1:26“John Deere: ‘We’re a Technology Company.’”
And that assertion appears increasingly less
1:31absurd. While the company lays off hundreds in
its manufacturing plants, it’s simultaneously
1:36staffing up its tech divisions. Of fifteen current
US job listings, twelve are in software, data, or
1:43robotics—just three in manufacturing. The company
has been going through a metamorphosis from one
1:48that makes machines to one that makes solutions.
Apple, for instance, does not merely make
1:55computers or phones or tablets. They
make integrated technology solutions,
2:00blending software, hardware, and services into an
ecosystem that envelops ones digital experience.
2:06Correspondingly, John Deere no longer merely
makes tractors or combines or loaders, but
2:11rather integrated production solutions—blending
software, hardware, and services into an ecosystem
2:17that envelops a farmer’s day-to-day experience.
At the extremes, the company appears entirely
2:23disconnected from its original form as a
small Illinois storefront selling shovels and
2:28pitchforks. Never could John Deere himself have
imagined that his company would eventually go on
2:32to operate, for instance, a satellite network.
But waxing nostalgic about humble beginnings
2:38would mask what the company now truly is: it’s not
an endearing family business, it’s not a scrappy
2:44underdog, it’s a market-domineering behemoth. It
has the business of agriculture, especially in
2:51America, in a stranglehold. And so might Apple,
with consumer electronics, but farming is not
2:57something you pick up or put on. It’s not one’s
digital experience, it’s one's entire experience:
3:03it’s your job, it’s where you live, it’s what
your family does, it’s what your neighbors do,
3:08it’s what your descendants do, it is an all
encompassing way of life whose future is now being
3:12dictated by one tractor company that’s decided
it’s big tech. But John Deere’s power, their
3:18influence, their ability to change the course
of history has been centuries in the making.
3:24It started with this—the self-scouring
steel plow. Plows had existed in some form
3:30for millennia—so many millennia, in fact, that we
can’t even say how old they are—but by the 1800s,
3:35they’d been refined and refined into this: a
single-piece cast-iron plow. These could be
3:40pulled by an animal, and would efficiently loosen
the soil to bring nutrients to the surface before
3:45planting a fresh crop. And these worked great, for
the time, except for here—the American midwest.
3:52The soil of Illinois and its neighbors was thick,
moist, and rooty, in a fashion that would lead
3:56it to clump on the plow, forcing farmers to
stop every once in a while to clean it off.
4:01But John Deere had an idea: he
would manufacture the same plow,
4:05but of polished steel. This cut straight through
the midwestern ground with far greater ease, and
4:11the soil would shed right off rather than clump.
While he was not the first to invent this concept,
4:16he was the first to start manufacturing
a steel plow en masse, and his production
4:19steadily grew into the hundreds per month,
and a later thousands. This innovation played
4:24an instrumental part in spreading agriculture
across the region, and transformed John Deere
4:29from a mere shop into a growing manufacturer.
By the turn of the century, after the company
4:34had passed through the generations of the family,
Deere had become a leading agricultural implements
4:39manufacturer, but the industry landscape was
changing beneath them. Like the plow, tractors
4:45as a concept, had long existed. Through the 19th
century it was typically animals like horse or
4:51oxen that pulled plows and other implements,
yet around the world, across industries, animal
4:56power was being replaced by steam power. Whereas
horse-drawn stagecoaches dominated the past, steam
5:02trains were now the dominant form of long-distance
transport, so the logic carried that steam-powered
5:06tractors could replace animal-power on farms.
And they certainly could, the technology existed,
5:12but steam-powered tractors never became ubiquitous
due to their high up-front and operating cost.
5:18But with the turn of the century came
gas-powered tractors: cheaper to buy,
5:22cheaper to operate. Popularity exploded, and while
Deere was originally reluctant to stray beyond
5:28their agricultural-implement core, they eventually
realized they had no choice if they wanted to
5:33stay relevant. They tried to design their own,
and it was plenty capable, but it was just not
5:38competitive. At about twice the price of that of
equivalent machines, their tractor never had any
5:43shot of commercial success, and then its designer
died from pneumonia following a week of testing in
5:47the wet and cold, so John Deere rather elected to
just simply buy their top competitor. With that,
5:53the Waterloo Gasoline Engine Company was folded
into John Deere, and over the following years,
5:58their tractor, the Waterloo Boy, enjoyed wild
success. Once again, like with the self-scouring
6:04steel plow, Deere didn’t invent the technology,
but they popularized it—they identified the
6:09opportunity, scaled up manufacturing, marketed
successfully, and helped transform the tractor
6:14from a niche, novel technology into the
solution for moving power on the farm.
6:20Over the decades that followed, the company
transformed again from primarily an implement
6:24company into a machine company—offering
combines and balers and planters and sprayers:
6:29essentially anything you needed to turn a field
into a farm. Through much of the 1900s they were
6:34always a significant, but underdog player in the
industry until the 60s and 70s when their primary
6:39competitor, International Harvester, began
to falter. And with its collapse in the 80s,
6:44John Deere took a firm lead in
the industry—becoming the go-to,
6:49ubiquitous source of agricultural equipment in the
United States. But then, another monumental shift
6:55in the field. Like the tractor introduced
the mechanical era of farming, information
7:00technology introduced farming to the digital age.
And again, John Deere had to adjust on the fly.
7:06This started on June 23, 1995 when Rockwell
International Corporation, traditionally a US
7:12defense manufacturer, unveiled its proprietary
Vision System—effectively firing the first shot
7:18of the digital farming revolution. It seems
so simple now, but Rockwell’s Vision System
7:23was poised to usher in a new age of efficiency by
using defense satellites to pinpoint and track a
7:28tractor from above, which, in turn, would allow
a farmer to better monitor their field’s yield,
7:32or when it came time to plant, better disperse
seeds and spray chemicals. Precision farming
7:37had arrived, and while Rockwell was first,
competitors such as Case Corporation, and Agco
7:42Corporation were close on the company’s heels.
So too was Deere, which took it one step further.
7:50To the late-90s American farmer, GPS and precision
agriculture was a handy tool, but still a finicky,
7:56expensive, and difficult-to-use luxury.
It helped, but it wasn’t required.
8:01At least, not until Deere made it a practical
necessity. As they did when developing their
8:06tractor, Deere looked further afield
for help. This started at Stanford,
8:10where the company collaborated with engineers
to develop an autonomous GPS-controlled tractor.
8:15While it worked, it didn’t work well enough
to take to market. The problem was 1990s GPS
8:21just wasn’t accurate or dependable enough.
So, more partnerships. Now Deere, along with
8:27NavCom Technology and NASA’s Jet Propulsion Lab,
sought to figure out how to create a more reliable
8:32positioning system to support not just yield maps
but autonomous guiding—the former a helpful tool,
8:38the latter a potentially revolutionary product.
While autonomous guiding may have seemed a
8:44lofty ambition, its use case was well grounded.
Before the rise of precision farming, farming was,
8:51well, remarkably imprecise. Take, for instance,
actually planting a field. Now, laying down seed
8:57is actually a rather complicated process with
a whole host of decisions to make and factors
9:02to consider from when to seed to how to establish
then plant the field’s headlands and borders. But
9:07regardless of such considerations—or what crop one
is even planting in the first place—each and every
9:12farmer, since the dawn of the tractor, has dealt
with one major inefficiency: overlap. In farming,
9:20overlapping is practically unavoidable—as
a tractor operator threads rows back and
9:24forth across their field, it’s nearly impossible
for there not to be slivers of field—whether it
9:28be where rows meet headlands, or just between
rows themselves—where the farmer doesn’t pass
9:33over a small section twice. And considering the
alternative—what farmers call sparing—this makes
9:38sense: if a farmer is to miss a small sliver
entirely while drilling, there will, of course
9:43be no crops, if they miss it with pesticides
or fertilizer, the section’s yield will drop,
9:48if they miss it during harvesting, well that’d
be an expensive and embarrassing mistake, too.
9:53So farmers overlap. But they try to do so as
little as possible. Experience helps with this, as
10:00hours in the chair, along with a long-established
sense of pride in maintaining straight rows,
10:04keeps overlap down. So too do generally normal,
rectangular fields, should a farmer have such a
10:10luxury. And then there are tricks: spray foams
to mark areas already hit, guideposts along
10:15fences for visual reference, thoughtfully laid out
tractor paths calibrated to align with the width
10:19of the farmer’s equipment. But tricks only go so
far. One study has put numbers on the overlaps.
10:25Across the study’s 17 locations and four years of
planting, the combine driller overlapped at 7.7%,
10:31spin disk fertilizer at 9.5%, while the sprayer
overlapped at 15.7%, and the cultivator reached
10:3719%. At every step of the process of growing
something, then, the farmer’s overlap is costing
10:43them—8% of their seed is being wasted, 10% of
their expensive fertilizer is being overapplied,
10:48nearly 16% in pesticide and herbicide is doing
more harm than good, and almost a fifth of their
10:53field is being turned over by the cultivator
for no reason. This means more materials. It
10:59also means more fuel, it means more time in the
field and in the chair, it means more hours put
11:03on the machines, and therefore more hours in
the shop and fewer functional seasons. Such
11:08costs really add up, too, as a bad year will see
costs outpace income, while 10 year averages, in
11:13the case of Kansas farms from 2010 to 2019, will
only net meager 11.8% profit margins. With such
11:20touchy and tight finances, unnecessarily wasting
8 to 19% of one’s time and money on overlapping
11:26is a massive inefficiency. One that Deere was
seeking to address at the dawn of a new century.
11:33The answer was called Starfire which, by
correcting notoriously inaccurate GPS data
11:38with ground location data, offered farmers field
mapping accurate to within 3 feet or 1 meter,
11:43rather than the 10-to-30 foot or three-to-ten
meter accuracy of traditional GPS. With further
11:48work on the product, by 2004, Starfire 2 provided
accuracy within 1.5 inches or 4.5 centimeters.
11:55Through the collaboration with Stanford, NASA,
and Navcom who they eventually acquired, Starfire
12:00positioned the company again on the cutting edge
of the precision farming revolution—not only
12:05did their product provide superior accuracy
for yield and seed mapping, it was accurate
12:09enough to address the fundamental inefficiency
of overlapping. Simply equip a machine with a
12:14Starfire receiver and a monitor then purchase
Deere’s Autotrac program and farmers could now
12:19guide by precise lines laid out on a screen and
even let the autonomous feature take the wheel.
12:25Today, through a combination of six uplink sites
on three continents, 46 reference sites around
12:29the globe, and leased bandwidth from Inmarsat
satellites, the shovel and pitchfork company
12:34is able to provide greater accuracy than the
public global positioning alternative, optimizing
12:39every single thread and turn across a farmer’s
field. The influence of such guidance can’t be
12:45understated. Only 10% of farmers used any sort of
auto steer and guidance system in 2004, but as of
12:512019, those numbers stood in the mid 50 to 60%
range, and on bigger, thousand-acre farms where
12:57the economies of scale blunt the upfront cost and
the waste of overlap is only magnified, adoption
13:02rates of such systems have reached over 80%.
By applying the same playbook they did with the
13:07plow and the tractor—embracing then perfecting
new technology through upfront R&D investment
13:11while also acquiring sector leaders like
Navcom—Deere helped push farming into a
13:16new epoch. But that epoch isn’t over, and Deere’s
only dug themselves further into the digital turn.
13:23Across the dozen American companies Deere has
acquired since 2007, only four are traditional
13:29hardware manufacturers, the rest, broadly, are in
tech, and increasingly in artificial intelligence,
13:35machine learning, and automation. In 2017, for
just north of $300 million, Deere purchased Blue
13:41River Technology, who had recently been testing
their new product called See & Spray—what they
13:46called the world’s first smart sprayer, which,
by feeding hundreds of thousands of plant images
13:51through deep learning algorithms was capable of
identifying crops and weeds before then spraying
13:55herbicide within a quarter-inch accuracy. Not
long after, Deere’s See & Spray Select entered
14:01the market. Then, in 2022, See & Spray Ultimate
became available for factory installation on 2023
14:08model 410R, 412R, and 612R Sprayers. With a camera
positioned along every meter of the carbon-fiber
14:15spray boom, the product would reduce spray volume
by two thirds, saving money on herbicides and, by
14:20extending trips between refills, saving time and
fuel. The benefits of the next step in precision
14:26agriculture also provided an environmental benefit
beyond the farmer too, as this product, the
14:31company projected, would reduce the airborne drift
of chemicals by up to 87% and chemical run-off by
14:36up to 93%. In this new era of AI and machine
learning precision farming, John Deere was not
14:43the first, as a Dyson subsidiary entered the smart
spray space earlier. Nor is it alone, as AgZen,
14:49a commercial outgrowth of MIT research, is pushing
into the space, too. Whether John Deere wins out
14:54here, as they have so many times in the past with
new technological innovations, remains to be seen,
15:00but given their history, it feels like a safe bet.
Regardless of competition, though, this service,
15:06capable of plugging right into the broader
John Deere ecosystem, should be a boon for
15:11the American farmer conscious of cost, yield,
and overall environmental impact of their work.
15:16Or at least, that’s what it would seem.
By standard metrics, farming in the US
15:21has gotten better across the sector’s
digital revolution—we’re wasting less,
15:25making better informed decisions,
and growing more than ever: just
15:28look at average yields for corn, soy, and cotton.
But consider the position of the American farmer.
15:34In the past, being an all green farm—that is
running strictly John Deere equipment—was a point
15:39of pride. Today, though, it’s increasingly feeling
like an expensive necessity without alternatives.
15:45New-found hyper efficiency comes with a cost, or
really, a whole host of costs. Say an Illinois soy
15:51farmer is sizing up purchasing a See & Spray
attachment—well, first they’ll need to have a
15:56fairly new sprayer to begin with, which if they
don’t have, will be in the ballpark of $50,000.
16:02Then add on another $25,000 for equipment and
install, which can only be done at an authorized
16:07dealership. Still, given that soybean pesticides
have reached an all-time high this decade coming
16:11in at $77 per acre in 2022, and given that
this farmer owns the median sized farm for the
16:16state at 4,500 acres, considering scale, such an
upfront investment may well be worth it. Without
16:22See and Spray, pesticides would cost $350,000,
with the product, assuming it cuts spraying
16:29down to a third, the farmer would only need about
$117,000 in pesticides—so with upfront costs,
16:35savings total about $62,000. But then another
cost: See and Spray subscriptions cost $4 per
16:42acre so cut out another $18,000 and the
economics become slightly less appealing.
16:47And then there’s the less tangible costs.
Fundamentally, precision agriculture is
16:52changing what it means to be a farmer. What was
once an occupation defined by individual autonomy,
16:57problem-solving, and improvisation is now
increasingly beholden to monitors, screens,
17:02software subscriptions, and the availability of
manufacturer-authorized technicians. Undoubtedly,
17:07this is an issue of nostalgia, but it
permeates in costly and frustrating ways too.
17:13Consider the solar storms that pushed northern
lights as far south as the American midwest. While
17:18it might’ve been a once in a lifetime experience
for the farmers who stayed up late to see it,
17:21it caused far more costly problems when the storm
knocked out their navigational systems. Just at
17:27the moment farmers needed to be out planting corn
their precision navigation systems failed them.
17:32While solar storms are few and far between,
issues with software programs and machinery
17:37that’s now more complicated than ever are far too
common and far too difficult to get figured out
17:42for your life-long farmer. Rather than hauling
a tractor back to the barn to fix a hydraulic
17:46leak and get back out on the field that same
day, when new-era hardware fails, there’s a
17:50good chance a farmer will be out of their depth
if it’s on the technology-side of the machine,
17:54which as the far more finicky side, it likely
is. So rather than fixing it and getting back
17:59out on the field, the farmer’s left waiting for
an authorized technician who will be expensive
18:03to pay and costly on time, as it’s unlikely
they’ll be available at the drop of a hat.
18:08And all that’s without considering whether the
farmer has the tools and information to make the
18:11fix in the first place—which is also a matter of
contention. Across the past decade, John Deere has
18:17found itself in the middle of a battle over the
right to repair. For those savvy enough or bold
18:22enough to fix their own issues, they often need
access to the diagnostic software to begin with,
18:26which is something Deere’s been slow to hand over.
Their stated reason to keep software restricted is
18:31a matter of liability and responsibility.
If they hand over the keys, they figure,
18:35their machines might get used and altered in
ways they shouldn’t. But for an increasingly
18:39boisterous farming community, this withholding
of key information is simply another way to make
18:44sure that farmer is also on the hook for costly
repair bills that make their way back to Deere,
18:49thus providing the company yet another revenue
stream. Regardless as to who is really telling
18:54the truth, what’s undeniable is that the
farmer is as financially squeezed as ever,
18:58and with the rise of big tech in farming, they are
increasingly being moved out of the driver’s seat.
19:04But Deere is facing their own financial
pressure—competition is rising, so they
19:08have to adapt to maintain their relevance. Over
the past decade, venture capital money has poured
19:13into startups that insist they can disrupt the
world of food production. Whereas in 2013 there
19:18were 42 funds focused on the AgriFood space,
today there are almost 300 reaching a peak
19:24of $53 billion of investment in 2021. A simple
thesis is presented to potential investors—the
19:31global share of land dedicated to agriculture is
peaking as more and more of the world urbanizes,
19:36yet simultaneously, the global population is
expected to continue increasing for at least
19:41half a century more. Therefore, it is objectively
true that we will have to produce more from less,
19:47and these startups believe the way to do
that is through technological innovation.
19:52Different companies have different solutions
to this problem—some are focused on “controlled
19:56environment agriculture,” growing indoors to
eliminate the threats and resource-losses from the
20:00outdoors; others are leveraging big data analytics
and machine learning to remove the inefficiencies
20:04of guesswork; while still others are working to
increase outdoor production yields and lower labor
20:09cost through various forms of autonomy. Building
on early successes in precision agriculture,
20:13John Deere has committed to developing a fully
autonomous production system for corn and soybean
20:18by 2030—that means every step from plowing
through planting through harvest without direct
20:23human involvement. And that’s remarkably
believable. Corn and soybean are planted
20:29in straight rows with relatively high distance
between each plant, meaning there’s already the
20:33predictability and margin for error that makes
it easiest for autonomous systems to succeed.
20:38The combination of innovations like
precision agriculture, indoor growing,
20:42autonomous production, and more will yield amazing
benefits for us all: in sum, they create a food
20:47production system that is less expensive and less
resource intensive. But they come at a cost—a very
20:53literal, incredibly significant, upfront cost.
The economics of paying an exorbitant amount for
20:59a fully autonomous wheat production system work
out first for the absolute largest farms. Just
21:05as with See & Spray, every innovation promises to
improve efficiency by a certain, small percent,
21:10so the larger the overall operation, the
more valuable that small percent can be,
21:15and therefore the more likely the upfront cost
is worth it. So innovations can be worth it,
21:20but only if you grow a ton of food. This
has been true for a while—there have been
21:25greater and greater economies of scale
in agriculture—which has contributed to
21:29a long-term trend of consolidation. Over the past
25 years, the average size of an American farm has
21:34grown by 7% even as total farmland has declined
8%—as small farms face increasing cost-pressure
21:41by more-efficient big-ag operations, they
either shut down or sell their land to big ag.
21:48And as big tech encroaches into farming,
innovation is accelerating, which is great
21:52by itself, but this leaves the small family farm
behind. A layperson’s perception of farming,
21:59as a mom and pop living in a homestead in Kansas,
working the fields around, answering to no boss
22:03but themselves, is becoming a cinematic fiction.
Increasingly, those living in regions dominated
22:10by agriculture work not for themselves, but
for landowners holding hundreds of thousands
22:15of acres. Often, the owners of this land
live time zones away, meaning profits from
22:20production are not spent at the local diner
or car dealership, but rather distributed to
22:24a multitude of investors and left to sit in
mutual funds. This contributes to a further
22:29gutting of the economy of rural America—one of the
rare ways to build a business outside of cities is
22:35becoming an increasing impossibility, and rural
resources are being extracted for urban gain.
22:42This is, in many ways, inevitable. Tech
innovation, across essentially any industry,
22:48has primarily benefited larger corporations
and incentivized consolidation. Farming,
22:53being so far from urban areas and so culturally
isolated from Silicon Valley, has long been
22:59shielded from these forces. Yet today, John Deere
and others have recognized the upside of bringing
23:05these two worlds together. But this progress will
hurt. The death of the family farm is upon us,
23:11and the autonomous tractor sits just beyond
the horizon, waiting to unleash its destructive
23:16ability to incrementally optimize yields.
The reason why John Deere’s See & Spray
23:22technology is so powerful is that it’s able to
interpret visual information and make decisions
23:27on what a given plant needs in an instant, at
a huge scale. What makes this possible behind
23:33the scenes is a neural network—the software was
fed millions of images and taught to interpret
23:38them as a human would. Neural networks are
fascinating as they’re essentially recreating
23:42a brain in a digital environment, and they’re
also some of the most wide scale instances of
23:46machine-learning right now, so it’s worthwhile to
understand how they work. And for that, there’s
23:51our sponsor Brilliant.org. Their class on neural
networks starts by teaching very basic principles
23:57through interactive problem-solving exercises
and straightforward, intuitive explanations,
24:01and then goes on to bring these concepts together
into practical applications. I really think they
24:06understand how to make learning more effective
and more engaging, so they’re perfect if you’re
24:09the person who really likes to learn a lot
efficiently. In addition to their classes
24:13on AI topics like neural networks and large
language models, they have dozens more in math,
24:18computer science, data analysis, and other STEM
subjects, each thoughtfully designed. And to make
24:23it all practical, they break everything up into
small chunks that can be completed in those small
24:27moments of your day, either on your computer or
on your phone, making it a perfect way to replace
24:31your scrolling with something that’ll better your
brain. Whether you’re in school and want to brush
24:35up on core principles or in the workforce and want
to better understand fundamental math and science
24:40topics that affect everything, Brilliant.org is
really the best place to learn. And you don’t
24:44have to trust me, you can see for yourself since
you can try everything Brilliant has to offer,
24:48for free, for a full thirty days when you go to
Brilliant.org/Wendover. Plus, you’ll also get
24:5420% off an annual premium subscription and
help support the channel when you sign up.
