Can AI predict someone's breakup? - Thomas Hofweber

169,364 views ใƒป 2024-05-21

TED-Ed


ืื ื ืœื—ืฅ ืคืขืžื™ื™ื ืขืœ ื”ื›ืชื•ื‘ื™ื•ืช ื‘ืื ื’ืœื™ืช ืœืžื˜ื” ื›ื“ื™ ืœื”ืคืขื™ืœ ืืช ื”ืกืจื˜ื•ืŸ.

ืชืจื’ื•ื: zeeva livshitz ืขืจื™ื›ื”: Ido Dekkers
00:07
You and your partner Alex have been in a strong, loving relationship for years,
0
7378
4713
ืืช ื•ื‘ืŸ ื–ื•ื’ืš ืืœื›ืก ื ืžืฆืื™ื ื‘ืžืขืจื›ืช ื™ื—ืกื™ื ื—ื–ืงื” ื•ืื•ื”ื‘ืช ื‘ืžืฉืš ืฉื ื™ื,
00:12
and lately you're considering getting engaged.
1
12091
2461
ื•ืœืื—ืจื•ื ื” ืืช ืฉื•ืงืœืช ืœื”ืชืืจืก.
00:14
Alex is enthusiastic about the idea, but you canโ€™t get over the statistics.
2
14677
4838
ืืœื›ืก ืžืชืœื”ื‘ ืžื”ืจืขื™ื•ืŸ, ืื‘ืœ ืืช ืœื ื™ื›ื•ืœื” ืœื”ืชื’ื‘ืจ ืขืœ ื”ืกื˜ื˜ื™ืกื˜ื™ืงื”.
00:19
You know a lot of marriages end in divorce, often not amicably.
3
19557
3670
ืืช ื™ื•ื“ืขืช ืฉื”ืจื‘ื” ื ื™ืฉื•ืื™ื ืžืกืชื™ื™ืžื™ื ื‘ื’ื™ืจื•ืฉื™ืŸ, ืœืขืชื™ื ืงืจื•ื‘ื•ืช ืœื ื‘ืฆื•ืจื” ื™ื“ื™ื“ื•ืชื™ืช.
00:23
And over 10% of couples in their first marriage get divorced
4
23269
3754
ืœืžืขืœื” ืž -10% ืžื”ื–ื•ื’ื•ืช ื‘ื ื™ืฉื•ืื™ื”ื ื”ืจืืฉื•ื ื™ื ืžืชื’ืจืฉื™ื
00:27
within the first five years.
5
27023
1877
ื‘ื—ืžืฉ ื”ืฉื ื™ื ื”ืจืืฉื•ื ื•ืช.
00:29
If your marriage wouldnโ€™t even last five years,
6
29692
2711
ืื ื”ื ื™ืฉื•ืื™ื ืฉืœืš ืœื ื”ื™ื• ื ืžืฉื›ื™ื ืืคื™ืœื• ื—ืžืฉ ืฉื ื™ื,
00:32
you feel like tying the knot would be a mistake.
7
32403
2378
ืืช ืžืจื’ื™ืฉื” ืฉืœืงืฉื•ืจ ืืช ื”ืงืฉืจ ืชื”ื™ื” ื˜ืขื•ืช.
00:34
But you live in the near future,
8
34989
2086
ืื‘ืœ ืืช ื—ื™ื” ื‘ืขืชื™ื“ ื”ืงืจื•ื‘,
00:37
where a brand-new company just released an AI-based model
9
37075
3545
ืฉื ื—ื‘ืจื” ื—ื“ืฉื” ืœื’ืžืจื™ ืคืจืกืžื” ืžื•ื“ืœ ืžื‘ื•ืกืก AI
00:40
that can predict your likelihood of divorce.
10
40620
3211
ืฉื™ื›ื•ืœ ืœื—ื–ื•ืช ืืช ื”ืกื‘ื™ืจื•ืช ืฉืœืš ืœื’ื™ืจื•ืฉื™ืŸ.
00:43
The model is trained on data sets
11
43915
2127
ื”ืžื•ื“ืœ ืžืื•ืžืŸ ืขืœ ืžืขืจื›ื™ ื ืชื•ื ื™ื
00:46
containing individualsโ€™ social media activity,
12
46042
2627
ื”ืžื›ื™ืœื™ื ืืช ืคืขื™ืœื•ืช ื”ืžื“ื™ื” ื”ื—ื‘ืจืชื™ืช ืฉืœ ืื ืฉื™ื,
00:48
online search histories, spending habits, and history of marriage and divorce.
13
48669
4880
ื”ื™ืกื˜ื•ืจื™ื™ืช ื—ื™ืคื•ืฉ ืžืงื•ื•ืŸ, ื”ืจื’ืœื™ ื”ื•ืฆืื•ืช ื•ื”ื™ืกื˜ื•ืจื™ื” ืฉืœ ื ื™ืฉื•ืื™ืŸ ื•ื’ื™ืจื•ืฉื™ืŸ.
00:53
And using this information,
14
53591
1585
ื•ื‘ืืžืฆืขื•ืช ืžื™ื“ืข ื–ื”,
00:55
the AI can predict if a couple will divorce
15
55176
2669
ื”- AI ื™ื›ื•ืœ ืœื—ื–ื•ืช ืื ื‘ื ื™ ื–ื•ื’
00:57
within the first five years of marriage with 95% accuracy.
16
57845
4380
ื™ืชื’ืจืฉื• ื‘ื—ืžืฉ ื”ืฉื ื™ื ื”ืจืืฉื•ื ื•ืช ืœื ื™ืฉื•ืื™ืŸ ื‘ื“ื™ื•ืง ืฉืœ 95%.
01:02
The only catch is the model doesnโ€™t offer any reasons for its resultsโ€”
17
62767
4212
ื”ืžืœื›ื•ื“ ื”ื™ื—ื™ื“ ื”ื•ื ืฉื”ืžื•ื“ืœ ืื™ื ื• ืžืฆื™ืข ืฉื•ื ืกื™ื‘ื” ืœืชื•ืฆืื•ืชื™ื• -
01:06
it simply predicts that you will or wonโ€™t divorce without saying why.
18
66979
4713
ื”ื•ื ืคืฉื•ื˜ ืžื ื‘ื ืฉืชืชื’ืจืฉื• ืื• ืœื ืชืชื’ืจืฉื• ืžื‘ืœื™ ืœื•ืžืจ ืžื“ื•ืข.
01:12
So, should you decide whether or not to get married
19
72443
3087
ืื–, ื”ืื ืืชื ืฆืจื™ื›ื™ื ืœื”ื—ืœื™ื˜ ืื ืœื”ืชื—ืชืŸ ืื• ืœื
01:15
based on this AIโ€™s prediction?
20
75530
2043
ืขืœ ืกืžืš ื”ืชื—ื–ื™ืช ืฉืœ AI ื–ื”?
01:18
Suppose the model predicts you and Alex
21
78825
2627
ื ื ื™ื— ืฉื”ืžื•ื“ืœ ืžื ื‘ื ืฉืืช ื•ืืœื›ืก
01:21
would divorce within five years of getting married.
22
81452
2920
ืชืชื’ืจืฉื• ืชื•ืš ื—ืžืฉ ืฉื ื™ื ืžืจื’ืข ื”ื ื™ืฉื•ืื™ืŸ.
01:24
At this point, you'd have three options.
23
84997
2253
ื‘ืฉืœื‘ ื–ื”, ื™ื”ื™ื• ืœื›ื ืฉืœื•ืฉ ืืคืฉืจื•ื™ื•ืช.
01:27
You could get married anyway and hope the prediction is wrong.
24
87250
3003
ืืชื ื™ื›ื•ืœื™ื ืœื”ืชื—ืชืŸ ื‘ื›ืœ ืžืงืจื” ื•ืœืงื•ื•ืช ืฉื”ืชื—ื–ื™ืช ืฉื’ื•ื™ื”.
01:30
You could break up now,
25
90294
1252
ืืชื ื™ื›ื•ืœื™ื ืœื”ื™ืคืจื“ ืขื›ืฉื™ื•,
01:31
though thereโ€™s no way to know if ending your currently happy relationship
26
91546
3753
ืื ื›ื™ ืื™ืŸ ื“ืจืš ืœื“ืขืช ืื ืกื™ื•ื ืžืขืจื›ืช ื”ื™ื—ืกื™ื ื”ืžืื•ืฉืจืช ืฉืœื›ื ื›ืจื’ืข
01:35
would cause more harm than letting the prediction run its course.
27
95299
3128
01:38
Or, you could stay together and remain unmarried,
28
98511
3378
ืœื—ืœื•ืคื™ืŸ, ืืชื ื™ื›ื•ืœื™ื ืœื”ื™ืฉืืจ ื™ื—ื“ ื•ืœื”ื™ืฉืืจ ืœื ื ืฉื•ืื™ื,
01:41
on the off-chance marriage itself would be the problem.
29
101889
3212
ื‘ืžืงืจื” ืฉื”ื ื™ืฉื•ืื™ืŸ ืขืฆืžื ื™ื”ื™ื• ื”ื‘ืขื™ื”.
01:45
Though without understanding the reasons for your predicted divorce,
30
105560
3753
ืื ื›ื™ ื‘ืœื™ ืœื”ื‘ื™ืŸ ืืช ื”ืกื™ื‘ื•ืช ืœื’ื™ืจื•ืฉื™ืŸ ื”ื—ื–ื•ื™ื™ื ืฉืœืš,
01:49
youโ€™d never know if those mystery issues would still emerge
31
109313
3671
ืœืขื•ืœื ืœื ืชื“ืขื• ืื ืกื•ื’ื™ื•ืช ื”ืžืกืชื•ืจื™ืŸ ื”ืœืœื• ืขื“ื™ื™ืŸ ื™ื•ืคื™ืขื•
01:52
to ruin your relationship.
32
112984
1585
ื›ื“ื™ ืœื”ืจื•ืก ืืช ืžืขืจื›ืช ื”ื™ื—ืกื™ื ืฉืœื›ื.
01:55
The uncertainty undermining all these options
33
115194
3212
ืื™ ื”ื•ื•ื“ืื•ืช ื”ืžืขืจืขืจืช ืืช ื›ืœ ื”ืืคืฉืจื•ื™ื•ืช ื”ืœืœื•
01:58
stems from a well known issue with AI around explainability and transparency.
34
118406
5589
ื ื•ื‘ืขืช ืžื‘ืขื™ื” ื™ื“ื•ืขื” ืขื AI ืกื‘ื™ื‘ ื”ืกื‘ืจ ื•ืฉืงื™ืคื•ืช.
02:04
This problem plagues tons of potentially useful predictive models,
35
124287
4087
ื‘ืขื™ื” ื–ื• ืคื•ื’ืขืช ื‘ื˜ื•ื ื•ืช ืฉืœ ืžื•ื“ืœื™ื ืฉืœ ื ื™ื‘ื•ื™ ืฉืขืฉื•ื™ื™ื ืœื”ื™ื•ืช ืฉื™ืžื•ืฉื™ื™ื,
02:08
such as those that could be used to predict which bank customers
36
128374
3420
ื›ื’ื•ืŸ ืืœื” ืฉื ื™ืชืŸ ืœื”ืฉืชืžืฉ ื‘ื”ื ื›ื“ื™ ืœื—ื–ื•ืช ืื™ืœื• ืœืงื•ื—ื•ืช ื‘ื ืง
02:11
are most likely to repay a loan,
37
131794
1960
ื ื•ื˜ื™ื ื‘ื™ื•ืชืจ ืœื”ื—ื–ื™ืจ ื”ืœื•ื•ืื”,
02:13
or which prisoners are most likely to reoffend if granted parole.
38
133754
4171
ืื• ืื™ืœื• ืืกื™ืจื™ื ื ื•ื˜ื™ื ื‘ื™ื•ืชืจ ืœืคื’ื•ืข ืžื—ื“ืฉ ืื ื™ื™ื ืชืŸ ืœื”ื ืฉื—ืจื•ืจ ืขืœ ืชื ืื™.
02:18
Without knowing why AI systems reach their decisions,
39
138176
3503
ืžื‘ืœื™ ืœื“ืขืช ืžื“ื•ืข ืžืขืจื›ื•ืช AI ืžื’ื™ืขื•ืช ืœื”ื—ืœื˜ื•ืชื™ื”ืŸ,
02:21
many worry we canโ€™t think critically about how to follow their advice.
40
141679
4296
ืจื‘ื™ื ื—ื•ืฉืฉื™ื ืฉืื ื—ื ื• ืœื ื™ื›ื•ืœื™ื ืœื—ืฉื•ื‘ ื‘ืื•ืคืŸ ื‘ื™ืงื•ืจืชื™ ื›ื™ืฆื“ ืœืขืงื•ื‘ ืื—ืจ ืขืฆืชื.
02:26
But the transparency problem doesnโ€™t just prevent us
41
146100
2711
ืื‘ืœ ื‘ืขื™ื™ืช ื”ืฉืงื™ืคื•ืช ืœื ืจืง ืžื•ื ืขืช ืžืื™ืชื ื•
02:28
from understanding these models,
42
148811
1794
ืœื”ื‘ื™ืŸ ืžื•ื“ืœื™ื ืืœื”,
02:30
it also impacts the userโ€™s accountability.
43
150605
3086
ื”ื™ื ื’ื ืžืฉืคื™ืขื” ืขืœ ืื—ืจื™ื•ืช ื”ืžืฉืชืžืฉ.
02:34
For example, if the AI's prediction led you to break up with Alex,
44
154150
4046
ืœื“ื•ื’ืžื”, ืื ื”ืชื—ื–ื™ืช ืฉืœ ื”ื‘ื™ื ื” ื”ืžืœืื›ื•ืชื™ืช ื”ื•ื‘ื™ืœื” ืื•ืชืš ืœื”ื™ืคืจื“ ืžืืœื›ืก,
02:38
what explanation could you reasonably offer them?
45
158196
2752
ืื™ื–ื” ื”ืกื‘ืจ ืชื•ื›ืœื™ ืœื”ืฆื™ืข ืœื”ื ื‘ืื•ืคืŸ ืกื‘ื™ืจ?
02:41
That you want to end your happy relationship
46
161115
2127
ืฉืืช ืจื•ืฆื” ืœืกื™ื™ื ืืช ืžืขืจื›ืช ื”ื™ื—ืกื™ื ื”ืžืื•ืฉืจืช ืฉืœื›ื
02:43
because some mysterious machine predicted its demise?
47
163242
3253
ื›ื™ ืื™ื–ื• ืžื›ื•ื ื” ืžืกืชื•ืจื™ืช ื—ื–ืชื” ืืช ืžื•ืชื”?
02:46
That hardly seems fair to Alex.
48
166579
2127
ื–ื” ืœื ื ืจืื” ื”ื•ื’ืŸ ืœืืœื›ืก.
02:48
We donโ€™t always owe people an explanation for our actions,
49
168831
3337
ืื ื—ื ื• ืœื ืชืžื™ื“ ื—ื™ื™ื‘ื™ื ืœืื ืฉื™ื ื”ืกื‘ืจ ืœืžืขืฉื™ื ื•,
02:52
but when we do,
50
172168
1168
ืื‘ืœ ื›ืฉืื ื—ื ื• ืขื•ืฉื™ื ื–ืืช,
02:53
AIโ€™s lack of transparency can create ethically challenging situations.
51
173336
4754
ื—ื•ืกืจ ื”ืฉืงื™ืคื•ืช ืฉืœ AI ื™ื›ื•ืœ ืœื™ืฆื•ืจ ืžืฆื‘ื™ื ืžืืชื’ืจื™ื ืžื‘ื—ื™ื ื” ืืชื™ืช.
02:58
And accountability is just one of the tradeoffs we make
52
178299
2961
ื•ืื—ืจื™ื•ืช ื”ื™ื ืจืง ืื—ืช ื”ืคืฉืจื•ืช ืฉืื ื• ืžืงื‘ืœื™ื
03:01
by outsourcing important decisions to AI.
53
181260
2962
ืขืœ ื™ื“ื™ ืžื™ืงื•ืจ ื—ื•ืฅ ืฉืœ ื”ื—ืœื˜ื•ืช ื—ืฉื•ื‘ื•ืช ืœ- AI.
03:04
If youโ€™re comfortable deferring your agency to an AI model
54
184889
3545
ืื ืืชื ืžืจื’ื™ืฉื™ื ื‘ื ื•ื— ืœื“ื—ื•ืช ืืช ื”ื‘ื—ื™ืจื” ืฉืœื›ื ืœืžื•ื“ืœ AI,
03:08
itโ€™s likely because youโ€™re focused on the accuracy of the prediction.
55
188434
3837
ืกื‘ื™ืจ ืœื”ื ื™ื— ืฉืืชื ืžืชืžืงื“ื™ื ื‘ืจืžืช ื”ื“ื™ื•ืง ืฉืœ ื”ืชื—ื–ื™ืช.
03:12
In this mindset, it doesnโ€™t really matter why you and Alex might break upโ€”
56
192271
4380
ื‘ื—ืฉื™ื‘ื” ื”ื–ื•, ื–ื” ืœื ืžืžืฉ ืžืฉื ื” ืžื“ื•ืข ืืช ื•ืืœื›ืก ืขืฉื•ื™ื™ื ืœื”ื™ืคืจื“ -
03:16
simply that you likely will.
57
196651
1918
ืคืฉื•ื˜ ืกื‘ื™ืจ ืœื”ื ื™ื— ืฉืชืคืจื“ื•.
03:18
But if you prioritize authenticity over accuracy,
58
198736
3754
ืื‘ืœ ืื ืืชื ืžืชืขื“ืคื™ื ืื•ืชื ื˜ื™ื•ืช ืขืœ ืคื ื™ ื“ื™ื•ืง,
03:22
then you'll need to understand and appreciate the reasons
59
202490
3086
ืชืฆื˜ืจื›ื• ืœื”ื‘ื™ืŸ ื•ืœื”ืขืจื™ืš ืืช ื”ืกื™ื‘ื•ืช
03:25
for your future divorce before ending things today.
60
205576
3170
ืœื’ื™ืจื•ืฉื™ืŸ ื”ืขืชื™ื“ื™ื™ื ืฉืœื›ื ืœืคื ื™ ืฉืชืกื™ื™ืžื• ืืช ื”ื“ื‘ืจื™ื ื”ื™ื•ื.
03:28
Authentic decision making like this is essential for maintaining accountability,
61
208955
4879
ืงื‘ืœืช ื”ื—ืœื˜ื•ืช ืื•ืชื ื˜ื™ืช ื›ื–ื• ื—ื™ื•ื ื™ืช ืœืฉืžื™ืจื” ืขืœ ืื—ืจื™ื•ืช,
03:33
and it might be your best chance to prove the prediction wrong.
62
213876
3921
ื•ื–ื” ืขืฉื•ื™ ืœื”ื™ื•ืช ื”ืกื™ื›ื•ื™ ื”ื˜ื•ื‘ ื‘ื™ื•ืชืจ ืฉืœื›ื ืœื”ื•ื›ื™ื— ืฉื”ืชื—ื–ื™ืช ืฉื’ื•ื™ื”.
03:37
On the other hand, itโ€™s also possible the model
63
217880
2920
ืžืฆื“ ืฉื ื™, ื™ื™ืชื›ืŸ ื’ื ืฉื”ืžื•ื“ืœ
03:40
already accounted for your attempts to defy it,
64
220800
2836
ื›ื‘ืจ ืœืงื— ื‘ื—ืฉื‘ื•ืŸ ืืช ื”ื ื™ืกื™ื•ื ื•ืช ืฉืœื›ื ืœื”ืชื ื’ื“ ืœื•,
03:43
and youโ€™re just setting yourself up for failure.
65
223636
2502
ื•ืืชื ืคืฉื•ื˜ ืžื•ืœื™ื›ื™ื ืืช ืขืฆืžื›ื ืœื›ื™ืฉืœื•ืŸ.
03:46
95% accuracy is high, but itโ€™s not perfectโ€”
66
226931
3587
ื“ื™ื•ืง ืฉืœ 95% ื”ื•ื ื’ื‘ื•ื”, ืื‘ืœ ื–ื” ืœื ืžื•ืฉืœื -
03:50
that figure means 1 in 20 couples will receive a false prediction.
67
230601
4672
ื ืชื•ืŸ ื–ื” ืื•ืžืจ ืฉืื—ื“ ืžื›ืœ 20 ื–ื•ื’ื•ืช ื™ืงื‘ืœ ืชื—ื–ื™ืช ื›ื•ื–ื‘ืช.
03:55
And as more people use this service,
68
235356
2210
ื•ื›ื›ืœ ืฉื™ื•ืชืจ ืื ืฉื™ื ืžืฉืชืžืฉื™ื ื‘ืฉื™ืจื•ืช ื–ื”,
03:57
the likelihood increases that someone who was predicted to divorce
69
237566
3879
ื”ืกื‘ื™ืจื•ืช ืขื•ืœื” ืฉืžื™ืฉื”ื• ืฉื ื—ื–ื” ืœื”ืชื’ืจืฉ ื™ืขืฉื” ื–ืืช
04:01
will do so just because the AI predicted they would.
70
241445
3420
ืจืง ื‘ื’ืœืœ ืฉื”ื‘ื™ื ื” ื”ืžืœืื›ื•ืชื™ืช ื—ื–ืชื” ืฉื”ื ื™ืขืฉื• ื–ืืช.
04:05
If that happens to enough newlyweds,
71
245366
2294
ืื ื–ื” ื™ืงืจื” ืœืžืกืคื™ืง ื–ื•ื’ื•ืช ืฆืขื™ืจื™ื,
04:07
the AI's success rate could be artificially maintained
72
247660
3712
ืฉื™ืขื•ืจ ื”ื”ืฆืœื—ื” ืฉืœ ื”- AI ื™ื›ื•ืœ ืœื”ื™ืฉืžืจ ื‘ืื•ืคืŸ ืžืœืื›ื•ืชื™
04:11
or even increased by these self-fulfilling predictions.
73
251372
3670
ืื• ืืคื™ืœื• ืœื’ื“ื•ืœ ื™ื“ื™ ืชื—ื–ื™ื•ืช ืืœื” ืฉืžื’ืฉื™ืžื•ืช ืืช ืขืฆืžืŸ.
04:15
Of course, no matter what the AI might tell you,
74
255167
2878
ื›ืžื•ื‘ืŸ, ืœื ืžืฉื ื” ืžื” ื”- AI ืขืฉื•ื™ ืœื•ืžืจ ืœื›ื,
04:18
whether you even ask for its prediction is still up to you.
75
258045
3796
ืื ืืชื ืืคื™ืœื• ืžื‘ืงืฉื™ื ืืช ื”ืชื—ื–ื™ืช ืฉืœื” ืขื“ื™ื™ืŸ ืชืœื•ื™ ื‘ื›ื.
ืขืœ ืืชืจ ื–ื”

ืืชืจ ื–ื” ื™ืฆื™ื’ ื‘ืคื ื™ื›ื ืกืจื˜ื•ื ื™ YouTube ื”ืžื•ืขื™ืœื™ื ืœืœื™ืžื•ื“ ืื ื’ืœื™ืช. ืชื•ื›ืœื• ืœืจืื•ืช ืฉื™ืขื•ืจื™ ืื ื’ืœื™ืช ื”ืžื•ืขื‘ืจื™ื ืขืœ ื™ื“ื™ ืžื•ืจื™ื ืžื”ืฉื•ืจื” ื”ืจืืฉื•ื ื” ืžืจื—ื‘ื™ ื”ืขื•ืœื. ืœื—ืฅ ืคืขืžื™ื™ื ืขืœ ื”ื›ืชื•ื‘ื™ื•ืช ื‘ืื ื’ืœื™ืช ื”ืžื•ืฆื’ื•ืช ื‘ื›ืœ ื“ืฃ ื•ื™ื“ืื• ื›ื“ื™ ืœื”ืคืขื™ืœ ืืช ื”ืกืจื˜ื•ืŸ ืžืฉื. ื”ื›ืชื•ื‘ื™ื•ืช ื’ื•ืœืœื•ืช ื‘ืกื ื›ืจื•ืŸ ืขื ื”ืคืขืœืช ื”ื•ื•ื™ื“ืื•. ืื ื™ืฉ ืœืš ื”ืขืจื•ืช ืื• ื‘ืงืฉื•ืช, ืื ื ืฆื•ืจ ืื™ืชื ื• ืงืฉืจ ื‘ืืžืฆืขื•ืช ื˜ื•ืคืก ื™ืฆื™ืจืช ืงืฉืจ ื–ื”.

https://forms.gle/WvT1wiN1qDtmnspy7