TED Salon Zebra Technologies
Matt Beane: How do we learn to work with intelligent machines?
Filmed:
Readability: 4.2
1,770,815 views
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.
Matt Beane - Organizational ethnographer
Matt Beane is an Assistant Professor in the Technology Management Program at the University of California, Santa Barbara and a Research Affiliate with MIT's Institute for the Digital Economy. Full bio
Matt Beane is an Assistant Professor in the Technology Management Program at the University of California, Santa Barbara and a Research Affiliate with MIT's Institute for the Digital Economy. Full bio
Double-click the English transcript below to play the video.
00:13
It’s 6:30 in the morning,
0
1292
1875
00:15
and Kristen is wheeling
her prostate patient into the OR.
her prostate patient into the OR.
1
3583
4875
00:21
She's a resident, a surgeon in training.
2
9500
2250
00:24
It’s her job to learn.
3
12333
2167
00:27
Today, she’s really hoping to do
some of the nerve-sparing,
some of the nerve-sparing,
4
15292
3351
00:30
extremely delicate dissection
that can preserve erectile function.
that can preserve erectile function.
5
18667
3875
00:35
That'll be up to the attending surgeon,
though, but he's not there yet.
though, but he's not there yet.
6
23500
3338
00:39
She and the team put the patient under,
7
27625
2393
00:42
and she leads the initial eight-inch
incision in the lower abdomen.
incision in the lower abdomen.
8
30042
3708
00:47
Once she’s got that clamped back,
she tells the nurse to call the attending.
she tells the nurse to call the attending.
9
35042
3586
00:51
He arrives, gowns up,
10
39583
2292
00:54
And from there on in, their four hands
are mostly in that patient --
are mostly in that patient --
11
42458
5792
01:00
with him guiding
but Kristin leading the way.
but Kristin leading the way.
12
48708
2917
01:04
When the prostates out (and, yes,
he let Kristen do a little nerve sparing),
he let Kristen do a little nerve sparing),
13
52875
4643
01:09
he rips off his scrubs.
14
57542
1226
01:10
He starts to do paperwork.
15
58792
1375
01:12
Kristen closes the patient by 8:15,
16
60833
5375
01:18
with a junior resident
looking over her shoulder.
looking over her shoulder.
17
66583
2435
01:21
And she lets him do
the final line of sutures.
the final line of sutures.
18
69042
3083
01:24
Kristen feels great.
19
72833
3042
01:28
Patient’s going to be fine,
20
76250
1559
01:29
and no doubt she’s a better surgeon
than she was at 6:30.
than she was at 6:30.
21
77833
3167
01:34
Now this is extreme work.
22
82208
2834
01:37
But Kristin’s learning to do her job
the way that most of us do:
the way that most of us do:
23
85417
3833
01:41
watching an expert for a bit,
24
89625
1893
01:43
getting involved in easy,
safe parts of the work
safe parts of the work
25
91542
3142
01:46
and progressing to riskier
and harder tasks
and harder tasks
26
94708
2185
01:48
as they guide and decide she’s ready.
27
96917
2333
01:52
My whole life I’ve been fascinated
by this kind of learning.
by this kind of learning.
28
100042
2892
01:54
It feels elemental,
part of what makes us human.
part of what makes us human.
29
102958
3667
01:59
It has different names: apprenticeship,
coaching, mentorship, on the job training.
coaching, mentorship, on the job training.
30
107750
5417
02:05
In surgery, it’s called
“see one, do one, teach one.”
“see one, do one, teach one.”
31
113542
3291
02:09
But the process is the same,
32
117625
1344
02:10
and it’s been the main path to skill
around the globe for thousands of years.
around the globe for thousands of years.
33
118993
4174
02:16
Right now, we’re handling AI
in a way that blocks that path.
in a way that blocks that path.
34
124333
4500
02:21
We’re sacrificing learning
in our quest for productivity.
in our quest for productivity.
35
129625
2690
02:25
I found this first in surgery
while I was at MIT,
while I was at MIT,
36
133292
2809
02:28
but now I’ve got evidence
it’s happening all over,
it’s happening all over,
37
136125
2476
02:30
in very different industries
and with very different kinds of AI.
and with very different kinds of AI.
38
138625
3875
02:35
If we do nothing, millions of us
are going to hit a brick wall
are going to hit a brick wall
39
143083
5851
02:40
as we try to learn to deal with AI.
40
148958
2417
02:45
Let’s go back to surgery to see how.
41
153125
1772
02:47
Fast forward six months.
42
155708
1935
02:49
It’s 6:30am again, and Kristen
is wheeling another prostate patient in,
is wheeling another prostate patient in,
43
157667
5476
02:55
but this time to the robotic OR.
44
163167
3166
02:59
The attending leads attaching
45
167667
1684
03:01
a four-armed, thousand-pound
robot to the patient.
robot to the patient.
46
169375
2833
03:04
They both rip off their scrubs,
47
172750
2434
03:07
head to control consoles
10 or 15 feet away,
10 or 15 feet away,
48
175208
3125
03:11
and Kristen just watches.
49
179167
3750
03:16
The robot allows the attending
to do the whole procedure himself,
to do the whole procedure himself,
50
184375
3053
03:19
so he basically does.
51
187452
1583
03:21
He knows she needs practice.
52
189917
2101
03:24
He wants to give her control.
53
192042
1583
03:26
But he also knows she’d be slower
and make more mistakes,
and make more mistakes,
54
194250
3393
03:29
and his patient comes first.
55
197667
1500
03:32
So Kristin has no hope of getting anywhere
near those nerves during this rotation.
near those nerves during this rotation.
56
200250
4625
03:37
She’ll be lucky if she operates more than
15 minutes during a four-hour procedure.
15 minutes during a four-hour procedure.
57
205417
4375
03:42
And she knows that when she slips up,
58
210250
2625
03:45
he’ll tap a touch screen,
and she’ll be watching again,
and she’ll be watching again,
59
213458
3042
03:48
feeling like a kid in the corner
with a dunce cap.
with a dunce cap.
60
216917
2625
03:53
Like all the studies of robots and work
I’ve done in the last eight years,
I’ve done in the last eight years,
61
221583
3501
03:57
I started this one
with a big, open question:
with a big, open question:
62
225108
2118
03:59
How do we learn to work
with intelligent machines?
with intelligent machines?
63
227250
2792
04:02
To find out, I spent two and a half years
observing dozens of residents and surgeons
observing dozens of residents and surgeons
64
230792
5809
04:08
doing traditional and robotic surgery,
interviewing them
interviewing them
65
236625
3476
04:12
and in general hanging out
with the residents as they tried to learn.
with the residents as they tried to learn.
66
240125
3338
04:16
I covered 18 of the top
US teaching hospitals,
US teaching hospitals,
67
244250
3351
04:19
and the story was the same.
68
247625
1458
04:21
Most residents were in Kristen's shoes.
69
249875
2542
04:24
They got to “see one” plenty,
70
252958
1792
04:27
but the “do one” was barely available.
71
255583
2292
04:30
So they couldn’t struggle,
and they weren’t learning.
and they weren’t learning.
72
258333
2528
04:33
This was important news for surgeons, but
I needed to know how widespread it was:
I needed to know how widespread it was:
73
261291
3810
04:37
Where else was using AI
blocking learning on the job?
blocking learning on the job?
74
265125
3833
04:42
To find out, I’ve connected with a small
but growing group of young researchers
but growing group of young researchers
75
270208
4310
04:46
who’ve done boots-on-the-ground studies
of work involving AI
of work involving AI
76
274542
3434
04:50
in very diverse settings
like start-ups, policing,
like start-ups, policing,
77
278000
2976
04:53
investment banking and online education.
78
281000
2601
04:55
Like me, they spent at least a year
and many hundreds of hours observing,
and many hundreds of hours observing,
79
283625
5851
05:01
interviewing and often working
side-by-side with the people they studied.
side-by-side with the people they studied.
80
289500
3917
05:06
We shared data, and I looked for patterns.
81
294458
2417
05:09
No matter the industry, the work,
the AI, the story was the same.
the AI, the story was the same.
82
297917
5208
05:16
Organizations were trying harder
and harder to get results from AI,
and harder to get results from AI,
83
304042
3642
05:19
and they were peeling learners away from
expert work as they did it.
expert work as they did it.
84
307708
3542
05:24
Start-up managers were outsourcing
their customer contact.
their customer contact.
85
312333
2875
05:27
Cops had to learn to deal with crime
forecasts without experts support.
forecasts without experts support.
86
315833
4042
05:32
Junior bankers were getting
cut out of complex analysis,
cut out of complex analysis,
87
320875
3250
05:36
and professors had to build
online courses without help.
online courses without help.
88
324500
3083
05:41
And the effect of all of this
was the same as in surgery.
was the same as in surgery.
89
329125
3226
05:44
Learning on the job
was getting much harder.
was getting much harder.
90
332375
2917
05:48
This can’t last.
91
336958
1417
05:51
McKinsey estimates that between half
a billion and a billion of us
a billion and a billion of us
92
339542
4267
05:55
are going to have to adapt to AI
in our daily work by 2030.
in our daily work by 2030.
93
343833
4125
06:01
And we’re assuming
that on-the-job learning
that on-the-job learning
94
349000
2011
06:03
will be there for us as we try.
95
351035
1917
06:05
Accenture’s latest workers survey showed
that most workers learned key skills
that most workers learned key skills
96
353500
4268
06:09
on the job, not in formal training.
97
357792
2291
06:13
So while we talk a lot about its
potential future impact,
potential future impact,
98
361292
3517
06:16
the aspect of AI
that may matter most right now
that may matter most right now
99
364833
3685
06:20
is that we’re handling it in a way
that blocks learning on the job
that blocks learning on the job
100
368542
3375
06:24
just when we need it most.
101
372375
1625
06:27
Now across all our sites,
a small minority found a way to learn.
a small minority found a way to learn.
102
375458
6042
06:35
They did it by breaking and bending rules.
103
383625
3042
06:39
Approved methods weren’t working,
so they bent and broke rules
so they bent and broke rules
104
387083
4643
06:43
to get hands-on practice with experts.
105
391750
1976
06:45
In my setting, residents got involved
in robotic surgery in medical school
in robotic surgery in medical school
106
393750
5601
06:51
at the expense
of their generalist education.
of their generalist education.
107
399375
3583
06:56
And they spent hundreds of extra hours
with simulators and recordings of surgery,
with simulators and recordings of surgery,
108
404417
5851
07:02
when you were supposed to learn in the OR.
109
410292
2541
07:05
And maybe most importantly,
they found ways to struggle
they found ways to struggle
110
413375
3476
07:08
in live procedures
with limited expert supervision.
with limited expert supervision.
111
416875
3750
07:13
I call all this “shadow learning,”
because it bends the rules
because it bends the rules
112
421792
4309
07:18
and learner’s do it out of the limelight.
113
426125
2000
07:21
And everyone turns a blind eye
because it gets results.
because it gets results.
114
429542
4101
07:25
Remember, these are
the star pupils of the bunch.
the star pupils of the bunch.
115
433667
3166
07:29
Now, obviously, this is not OK,
and it’s not sustainable.
and it’s not sustainable.
116
437792
3208
07:33
No one should have to risk getting fired
117
441708
2185
07:35
to learn the skills
they need to do their job.
they need to do their job.
118
443917
2150
07:38
But we do need to learn from these people.
119
446792
2056
07:41
They took serious risks to learn.
120
449917
2250
07:44
They understood they needed to protect
struggle and challenge in their work
struggle and challenge in their work
121
452792
4351
07:49
so that they could push themselves
to tackle hard problems
to tackle hard problems
122
457167
2892
07:52
right near the edge of their capacity.
123
460083
1959
07:54
They also made sure
there was an expert nearby
there was an expert nearby
124
462458
2216
07:56
to offer pointers and to backstop
against catastrophe.
against catastrophe.
125
464698
3094
08:00
Let’s build this combination
of struggle and expert support
of struggle and expert support
126
468875
3458
08:04
into each AI implementation.
127
472708
2750
08:08
Here’s one clear example
I could get of this on the ground.
I could get of this on the ground.
128
476375
2828
08:12
Before robots,
129
480125
1226
08:13
if you were a bomb disposal technician,
you dealt with an IED by walking up to it.
you dealt with an IED by walking up to it.
130
481375
4792
08:19
A junior officer was
hundreds of feet away,
hundreds of feet away,
131
487333
2143
08:21
so could only watch and help
if you decided it was safe
if you decided it was safe
132
489500
3309
08:24
and invited them downrange.
133
492833
1417
08:27
Now you sit side-by-side
in a bomb-proof truck.
in a bomb-proof truck.
134
495208
3893
08:31
You both watched the video feed.
135
499125
1809
08:32
They control a distant robot,
and you guide the work out loud.
and you guide the work out loud.
136
500958
4310
08:37
Trainees learn better than they
did before robots.
did before robots.
137
505292
3208
08:41
We can scale this to surgery,
start-ups, policing,
start-ups, policing,
138
509125
3933
08:45
investment banking,
online education and beyond.
online education and beyond.
139
513082
2625
08:48
The good news is
we’ve got new tools to do it.
we’ve got new tools to do it.
140
516375
2500
08:51
The internet and the cloud mean we don’t
always need one expert for every trainee,
always need one expert for every trainee,
141
519750
4082
08:56
for them to be physically near each other
or even to be in the same organization.
or even to be in the same organization.
142
524167
4458
09:01
And we can build AI to help:
143
529292
3041
09:05
to coach learners as they struggle,
to coach experts as they coach
to coach experts as they coach
144
533167
5059
09:10
and to connect those two groups
in smart ways.
in smart ways.
145
538250
2542
09:15
There are people at work
on systems like this,
on systems like this,
146
543375
2542
09:18
but they’ve been mostly focused
on formal training.
on formal training.
147
546333
2792
09:21
And the deeper crisis
is in on-the-job learning.
is in on-the-job learning.
148
549458
2584
09:24
We must do better.
149
552417
1851
09:26
Today’s problems demand we do better
150
554292
2583
09:29
to create work that takes full advantage
of AI’s amazing capabilities
of AI’s amazing capabilities
151
557375
4875
09:35
while enhancing our skills as we do it.
152
563042
2750
09:38
That’s the kind of future
I dreamed of as a kid.
I dreamed of as a kid.
153
566333
2750
09:41
And the time to create it is now.
154
569458
2167
09:44
Thank you.
155
572333
1226
09:45
(Applause)
156
573583
3625
ABOUT THE SPEAKER
Matt Beane - Organizational ethnographerMatt Beane is an Assistant Professor in the Technology Management Program at the University of California, Santa Barbara and a Research Affiliate with MIT's Institute for the Digital Economy.
Why you should listen
Matt Beane does field research on work involving robots to help us understand the implications of intelligent machines for the broader world of work. Any of his projects mean many hundreds of hours -- sometimes years -- watching, interviewing and often working side by side with people trying to work with robots to get their jobs done.
Beane has studied robotic surgery, robotic materials transport and robotic telepresence in healthcare, elder care and knowledge work. He has published in top management journals such as Administrative Science Quarterly, he was selected in 2012 as a Human Robot Interaction Pioneer and is a regular contributor to popular outlets such as Wired, MIT Technology Review, TechCrunch, Forbes and Robohub. He also took a two-year hiatus from his doctoral studies to help found and fund Humatics, an MIT-connected, full-stack IoT startup.
Beane is an Assistant Professor in the Technology Management Program at the University of California, Santa Barbara and a Research Affiliate with MIT's Institute for the Digital Economy. He received his PhD from the MIT Sloan School of Management.
Matt Beane | Speaker | TED.com