How do we learn to work with intelligent machines?
-
0:01 - 0:03It’s 6:30 in the morning,
-
0:04 - 0:08and Kristen is wheeling
her prostate patient into the OR. -
0:10 - 0:12She's a resident, a surgeon in training.
-
0:12 - 0:14It’s her job to learn.
-
0:15 - 0:19Today, she’s really hoping to do
some of the nerve-sparing, -
0:19 - 0:23extremely delicate dissection
that can preserve erectile function. -
0:24 - 0:27That'll be up to the attending surgeon,
though, but he's not there yet. -
0:28 - 0:30She and the team put the patient under,
-
0:30 - 0:34and she leads the initial eight-inch
incision in the lower abdomen. -
0:35 - 0:39Once she’s got that clamped back,
she tells the nurse to call the attending. -
0:40 - 0:42He arrives, gowns up,
-
0:42 - 0:48And from there on in, their four hands
are mostly in that patient -- -
0:49 - 0:52with him guiding
but Kristin leading the way. -
0:53 - 0:58When the prostates out (and, yes,
he let Kristen do a little nerve sparing), -
0:58 - 0:59he rips off his scrubs.
-
0:59 - 1:00He starts to do paperwork.
-
1:01 - 1:06Kristen closes the patient by 8:15,
-
1:07 - 1:09with a junior resident
looking over her shoulder. -
1:09 - 1:12And she lets him do
the final line of sutures. -
1:13 - 1:16Kristen feels great.
-
1:16 - 1:18Patient’s going to be fine,
-
1:18 - 1:21and no doubt she’s a better surgeon
than she was at 6:30. -
1:22 - 1:25Now this is extreme work.
-
1:25 - 1:29But Kristin’s learning to do her job
the way that most of us do: -
1:30 - 1:32watching an expert for a bit,
-
1:32 - 1:35getting involved in easy,
safe parts of the work -
1:35 - 1:37and progressing to riskier
and harder tasks -
1:37 - 1:39as they guide and decide she’s ready.
-
1:40 - 1:43My whole life I’ve been fascinated
by this kind of learning. -
1:43 - 1:47It feels elemental,
part of what makes us human. -
1:48 - 1:53It has different names: apprenticeship,
coaching, mentorship, on the job training. -
1:54 - 1:57In surgery, it’s called
“see one, do one, teach one.” -
1:58 - 1:59But the process is the same,
-
1:59 - 2:03and it’s been the main path to skill
around the globe for thousands of years. -
2:04 - 2:09Right now, we’re handling AI
in a way that blocks that path. -
2:10 - 2:12We’re sacrificing learning
in our quest for productivity. -
2:13 - 2:16I found this first in surgery
while I was at MIT, -
2:16 - 2:19but now I’ve got evidence
it’s happening all over, -
2:19 - 2:22in very different industries
and with very different kinds of AI. -
2:23 - 2:29If we do nothing, millions of us
are going to hit a brick wall -
2:29 - 2:31as we try to learn to deal with AI.
-
2:33 - 2:35Let’s go back to surgery to see how.
-
2:36 - 2:38Fast forward six months.
-
2:38 - 2:43It’s 6:30am again, and Kristen
is wheeling another prostate patient in, -
2:43 - 2:46but this time to the robotic OR.
-
2:48 - 2:49The attending leads attaching
-
2:49 - 2:52a four-armed, thousand-pound
robot to the patient. -
2:53 - 2:55They both rip off their scrubs,
-
2:55 - 2:58head to control consoles
10 or 15 feet away, -
2:59 - 3:03and Kristen just watches.
-
3:04 - 3:07The robot allows the attending
to do the whole procedure himself, -
3:07 - 3:09so he basically does.
-
3:10 - 3:12He knows she needs practice.
-
3:12 - 3:14He wants to give her control.
-
3:14 - 3:18But he also knows she’d be slower
and make more mistakes, -
3:18 - 3:19and his patient comes first.
-
3:20 - 3:25So Kristin has no hope of getting anywhere
near those nerves during this rotation. -
3:25 - 3:30She’ll be lucky if she operates more than
15 minutes during a four-hour procedure. -
3:30 - 3:33And she knows that when she slips up,
-
3:33 - 3:36he’ll tap a touch screen,
and she’ll be watching again, -
3:37 - 3:40feeling like a kid in the corner
with a dunce cap. -
3:42 - 3:45Like all the studies of robots and work
I’ve done in the last eight years, -
3:45 - 3:47I started this one
with a big, open question: -
3:47 - 3:50How do we learn to work
with intelligent machines? -
3:51 - 3:57To find out, I spent two and a half years
observing dozens of residents and surgeons -
3:57 - 4:00doing traditional and robotic surgery,
interviewing them -
4:00 - 4:03and in general hanging out
with the residents as they tried to learn. -
4:04 - 4:08I covered 18 of the top
US teaching hospitals, -
4:08 - 4:09and the story was the same.
-
4:10 - 4:12Most residents were in Kristen's shoes.
-
4:13 - 4:15They got to “see one” plenty,
-
4:16 - 4:18but the “do one” was barely available.
-
4:18 - 4:21So they couldn’t struggle,
and they weren’t learning. -
4:21 - 4:25This was important news for surgeons, but
I needed to know how widespread it was: -
4:25 - 4:29Where else was using AI
blocking learning on the job? -
4:30 - 4:35To find out, I’ve connected with a small
but growing group of young researchers -
4:35 - 4:38who’ve done boots-on-the-ground studies
of work involving AI -
4:38 - 4:41in very diverse settings
like start-ups, policing, -
4:41 - 4:44investment banking and online education.
-
4:44 - 4:49Like me, they spent at least a year
and many hundreds of hours observing, -
4:50 - 4:53interviewing and often working
side-by-side with the people they studied. -
4:54 - 4:57We shared data, and I looked for patterns.
-
4:58 - 5:03No matter the industry, the work,
the AI, the story was the same. -
5:04 - 5:08Organizations were trying harder
and harder to get results from AI, -
5:08 - 5:11and they were peeling learners away from
expert work as they did it. -
5:12 - 5:15Start-up managers were outsourcing
their customer contact. -
5:16 - 5:20Cops had to learn to deal with crime
forecasts without experts support. -
5:21 - 5:24Junior bankers were getting
cut out of complex analysis, -
5:24 - 5:28and professors had to build
online courses without help. -
5:29 - 5:32And the effect of all of this
was the same as in surgery. -
5:32 - 5:35Learning on the job
was getting much harder. -
5:37 - 5:38This can’t last.
-
5:40 - 5:44McKinsey estimates that between half
a billion and a billion of us -
5:44 - 5:48are going to have to adapt to AI
in our daily work by 2030. -
5:49 - 5:51And we’re assuming
that on-the-job learning -
5:51 - 5:53will be there for us as we try.
-
5:54 - 5:58Accenture’s latest workers survey showed
that most workers learned key skills -
5:58 - 6:00on the job, not in formal training.
-
6:01 - 6:05So while we talk a lot about its
potential future impact, -
6:05 - 6:09the aspect of AI
that may matter most right now -
6:09 - 6:12is that we’re handling it in a way
that blocks learning on the job -
6:12 - 6:14just when we need it most.
-
6:15 - 6:22Now across all our sites,
a small minority found a way to learn. -
6:24 - 6:27They did it by breaking and bending rules.
-
6:27 - 6:32Approved methods weren’t working,
so they bent and broke rules -
6:32 - 6:34to get hands-on practice with experts.
-
6:34 - 6:39In my setting, residents got involved
in robotic surgery in medical school -
6:39 - 6:43at the expense
of their generalist education. -
6:44 - 6:50And they spent hundreds of extra hours
with simulators and recordings of surgery, -
6:50 - 6:53when you were supposed to learn in the OR.
-
6:53 - 6:57And maybe most importantly,
they found ways to struggle -
6:57 - 7:01in live procedures
with limited expert supervision. -
7:02 - 7:06I call all this “shadow learning,”
because it bends the rules -
7:06 - 7:08and learner’s do it out of the limelight.
-
7:10 - 7:14And everyone turns a blind eye
because it gets results. -
7:14 - 7:17Remember, these are
the star pupils of the bunch. -
7:18 - 7:21Now, obviously, this is not OK,
and it’s not sustainable. -
7:22 - 7:24No one should have to risk getting fired
-
7:24 - 7:26to learn the skills
they need to do their job. -
7:27 - 7:29But we do need to learn from these people.
-
7:30 - 7:32They took serious risks to learn.
-
7:33 - 7:37They understood they needed to protect
struggle and challenge in their work -
7:37 - 7:40so that they could push themselves
to tackle hard problems -
7:40 - 7:42right near the edge of their capacity.
-
7:42 - 7:45They also made sure
there was an expert nearby -
7:45 - 7:48to offer pointers and to backstop
against catastrophe. -
7:49 - 7:52Let’s build this combination
of struggle and expert support -
7:53 - 7:55into each AI implementation.
-
7:56 - 7:59Here’s one clear example
I could get of this on the ground. -
8:00 - 8:01Before robots,
-
8:01 - 8:06if you were a bomb disposal technician,
you dealt with an IED by walking up to it. -
8:07 - 8:09A junior officer was
hundreds of feet away, -
8:10 - 8:13so could only watch and help
if you decided it was safe -
8:13 - 8:14and invited them downrange.
-
8:15 - 8:19Now you sit side-by-side
in a bomb-proof truck. -
8:19 - 8:21You both watched the video feed.
-
8:21 - 8:25They control a distant robot,
and you guide the work out loud. -
8:25 - 8:28Trainees learn better than they
did before robots. -
8:29 - 8:33We can scale this to surgery,
start-ups, policing, -
8:33 - 8:36investment banking,
online education and beyond. -
8:36 - 8:39The good news is
we’ve got new tools to do it. -
8:40 - 8:44The internet and the cloud mean we don’t
always need one expert for every trainee, -
8:44 - 8:49for them to be physically near each other
or even to be in the same organization. -
8:49 - 8:52And we can build AI to help:
-
8:53 - 8:58to coach learners as they struggle,
to coach experts as they coach -
8:58 - 9:01and to connect those two groups
in smart ways. -
9:03 - 9:06There are people at work
on systems like this, -
9:06 - 9:09but they’ve been mostly focused
on formal training. -
9:09 - 9:12And the deeper crisis
is in on-the-job learning. -
9:12 - 9:14We must do better.
-
9:14 - 9:17Today’s problems demand we do better
-
9:17 - 9:22to create work that takes full advantage
of AI’s amazing capabilities -
9:23 - 9:26while enhancing our skills as we do it.
-
9:26 - 9:29That’s the kind of future
I dreamed of as a kid. -
9:29 - 9:32And the time to create it is now.
-
9:32 - 9:34Thank you.
-
9:34 - 9:37(Applause)
- Title:
- How do we learn to work with intelligent machines?
- Speaker:
- Matt Beane
- Description:
-
The path to skill around the globe has been the same for thousands of years: train under an expert and take on small, easy tasks before progressing to riskier, harder ones. But right now, we're handling AI in a way that blocks that path -- and sacrificing learning in our quest for productivity, says organizational ethnographer Matt Beane. What can be done? Beane shares a vision that flips the current story into one of distributed, machine-enhanced mentorship that takes full advantage of AI’s amazing capabilities while enhancing our skills at the same time.
- Video Language:
- English
- Team:
- closed TED
- Project:
- TEDTalks
- Duration:
- 09:50
Hiroshi Uchiyama commented on English subtitles for How do we learn to work with intelligent machines? | ||
Brian Greene edited English subtitles for How do we learn to work with intelligent machines? | ||
Brian Greene edited English subtitles for How do we learn to work with intelligent machines? | ||
Brian Greene edited English subtitles for How do we learn to work with intelligent machines? | ||
Brian Greene approved English subtitles for How do we learn to work with intelligent machines? | ||
Brian Greene edited English subtitles for How do we learn to work with intelligent machines? | ||
Brian Greene accepted English subtitles for How do we learn to work with intelligent machines? | ||
Brian Greene edited English subtitles for How do we learn to work with intelligent machines? |
Hiroshi Uchiyama
The name of trainee varies between "Kristen" and "Kristin" time to time. I am working on Japanese translation based on "Kristin" which sounds more familiar as a female name for me.