1
00:00:00,000 --> 00:00:06,000
Imagine a machine with a simple job: send circles to one tray and squares to another.

2
00:00:06,128 --> 00:00:10,360
Here is a teal circle. Here is a teal square.

3
00:00:10,380 --> 00:00:15,240
You can see the difference. But the machine's sensor reports only colour.

4
00:00:15,280 --> 00:00:19,600
For both cards, it receives exactly the same message: teal.

5
00:00:19,740 --> 00:00:22,040
How can it choose the right tray?

6
00:00:22,400 --> 00:00:26,120
With this information alone, it cannot always get the answer right.

7
00:00:26,187 --> 00:00:29,880
It could guess. It could always choose the circle tray.

8
00:00:29,900 --> 00:00:36,280
But the two cards need different responses, and nothing in its input tells it which response to choose.

9
00:00:36,368 --> 00:00:39,720
That is why we ask what an observer can tell apart.

10
00:00:39,780 --> 00:00:46,760
The answer sets a limit on what it can reliably do, including when a decision needs another observation.

11
00:00:47,264 --> 00:00:52,496
This brings us to a question about mind: how could a system represent the limits of its own

12
00:00:52,496 --> 00:00:54,840
view, and use that knowledge?

13
00:00:57,720 --> 00:01:01,480
The website makes this problem small enough to inspect completely.

14
00:01:01,507 --> 00:01:07,944
There are four possibilities: a teal circle, a teal square, a gold circle, and a

15
00:01:07,944 --> 00:01:12,320
gold square. Their two digits are just labels for colour and shape.

16
00:01:12,728 --> 00:01:16,960
Here, observer means a system with access to specified tests.

17
00:01:17,000 --> 00:01:22,760
We are not assuming that it has an inner experience, or imagining a little person inside the machine.

18
00:01:22,808 --> 00:01:28,720
With colour alone, both teal cards give one answer, and both gold cards give another.

19
00:01:28,744 --> 00:01:33,200
Four different possibilities produce only two distinguishable records.

20
00:01:33,340 --> 00:01:38,480
Now enable shape as well. Each card produces a different pair of answers.

21
00:01:38,600 --> 00:01:42,000
A rule can use the shape answer to select the correct tray.

22
00:01:42,080 --> 00:01:47,229
The cards have stayed the same. What changed is the information available for choosing an

23
00:01:47,229 --> 00:01:52,800
action. The demonstration isolates that change so we can understand it precisely.

24
00:01:55,680 --> 00:01:57,840
There is a second lesson hiding here.

25
00:01:57,863 --> 00:02:03,220
Colour alone makes two groups. Shape alone also makes two groups.

26
00:02:03,220 --> 00:02:07,000
But only the shape groups support perfect sorting by shape.

27
00:02:07,086 --> 00:02:12,440
Change the task to sorting by colour, and the useful distinction changes too.

28
00:02:12,613 --> 00:02:15,480
So the number of distinctions is not enough.

29
00:02:15,520 --> 00:02:21,720
We need to ask what those distinctions allow the system to predict, decide, or control.

30
00:02:21,820 --> 00:02:27,200
For the shape task, treating the two circles as equivalent is perfectly adequate.

31
00:02:27,200 --> 00:02:32,040
Their colours differ, but that difference does not affect the required answer.

32
00:02:32,064 --> 00:02:37,964
This gives us a small model of how a category can serve a purpose: it preserves a relevant

33
00:02:37,964 --> 00:02:40,120
difference while setting another aside.

34
00:02:40,128 --> 00:02:43,320
That does not explain human concepts in full.

35
00:02:43,387 --> 00:02:49,024
It gives a proposed explanation something concrete to specify: which differences are

36
00:02:49,024 --> 00:02:53,720
preserved, which are ignored, and why those choices matter for the task.

37
00:02:53,840 --> 00:02:58,880
We will return to this problem when a robot must tell whether a dark image comes from its

38
00:02:58,880 --> 00:03:01,320
surroundings or from its own camera.

39
00:03:01,373 --> 00:03:06,200
First, we need to be precise about what its evidence establishes.

40
00:03:10,080 --> 00:03:15,640
The first linked page asks a different question: what can the available evidence establish?

41
00:03:15,787 --> 00:03:21,720
We now change the rules. The website's switches gave us complete yes-or-no answers.

42
00:03:21,728 --> 00:03:26,120
Here, a test supplies a positive confirmation: a certificate.

43
00:03:26,187 --> 00:03:29,540
It tells the observer which claim the evidence supports.

44
00:03:29,540 --> 00:03:33,880
The observer receives these signals, rather than our view of the card.

45
00:03:33,920 --> 00:03:37,120
We can certify gold, or certify square.

46
00:03:37,160 --> 00:03:43,000
Watch this gold card. We can see its colour, but the observer has no confirmation yet.

47
00:03:43,056 --> 00:03:45,560
Gold has not been established for it.

48
00:03:45,568 --> 00:03:48,200
That does not mean gold has been ruled out.

49
00:03:48,200 --> 00:03:51,160
The gold confirmation covers two cards.

50
00:03:51,160 --> 00:03:53,630
The square confirmation covers two.

51
00:03:53,630 --> 00:03:58,590
Combine them with and, and their overlap identifies the gold square.

52
00:03:58,590 --> 00:04:03,600
Combine them with or, and they cover every card except the teal circle.

53
00:04:03,618 --> 00:04:09,320
Together with the empty region and the whole set, these regions form a small topology.

54
00:04:09,400 --> 00:04:15,280
Any OR combination, and any finite AND combination, stays in this collection.

55
00:04:15,380 --> 00:04:21,580
Why collect these regions? They let us check which claims the evidence can support, and compare

56
00:04:21,580 --> 00:04:24,280
how that changes when we add confirmations.

57
00:04:24,320 --> 00:04:28,440
Gold and square can jointly certify one particular card.

58
00:04:28,460 --> 00:04:33,840
But nothing made from these positive confirmations alone singles out the teal circle.

59
00:04:33,904 --> 00:04:36,760
This is why the choice of evidence matters.

60
00:04:36,920 --> 00:04:43,080
A complete negative answer and the absence of a positive confirmation are different resources.

61
00:04:43,100 --> 00:04:46,200
The mathematics keeps that difference explicit.

62
00:04:51,080 --> 00:04:55,160
The second linked page studies changes to the observation structure.

63
00:04:55,229 --> 00:04:58,664
Return to the complete yes-or-no records.

64
00:04:58,664 --> 00:05:03,280
Keep every old test with its answers unchanged, and add a new one.

65
00:05:03,286 --> 00:05:07,200
States that already gave different answers remain distinguishable.

66
00:05:07,360 --> 00:05:11,440
A shared group may split if the new test separates its members.

67
00:05:11,507 --> 00:05:14,400
Repeating a test need not split anything.

68
00:05:14,456 --> 00:05:16,880
That is different from a change in the world.

69
00:05:17,000 --> 00:05:23,720
A teal circle might become a teal square while a colour-only observer continues to receive the same answer.

70
00:05:23,820 --> 00:05:30,720
One change concerns the state. The other concerns the observer's access to distinctions between states.

71
00:05:30,800 --> 00:05:36,509
The page distinguishes several kinds of change to an observational representation.

72
00:05:36,509 --> 00:05:43,360
Adding a test is one case. Explaining how an agent finds a useful test requires a further model.

73
00:05:47,240 --> 00:05:50,080
How does this connect to mind and awareness?

74
00:05:50,187 --> 00:05:56,280
Think about the difference between choosing an answer and recognising that you may have answered

75
00:05:56,280 --> 00:06:02,745
badly. A theory of mind may need to explain both the use of information and a system's

76
00:06:02,745 --> 00:06:04,960
assessment of its own performance.

77
00:06:05,062 --> 00:06:09,360
Research on metacognition studies that second question.

78
00:06:09,400 --> 00:06:15,304
For example, experiments can measure how well a person's confidence distinguishes their correct

79
00:06:15,304 --> 00:06:20,800
decisions from their errors, separately from how well they perform the original task.

80
00:06:20,920 --> 00:06:27,462
The cards address an earlier, narrower issue: which differences are available in the specified

81
00:06:27,462 --> 00:06:35,280
observations at all? They do not yet show memory, attention, confidence, or a model of the observer itself.

82
00:06:35,400 --> 00:06:40,600
And there is the question of experience: whether there is anything it feels like to be that

83
00:06:40,600 --> 00:06:46,656
system. Successful sorting, or successful self-monitoring, would not by itself

84
00:06:46,656 --> 00:06:48,799
establish an answer to that question.

85
00:06:48,820 --> 00:06:54,880
Keeping these questions distinct lets us state what an explanation actually explains.

86
00:06:54,973 --> 00:07:01,240
It gives The MIND a way to formulate specific problems without treating a simple demonstration

87
00:07:01,240 --> 00:07:03,680
as a complete account of awareness.

88
00:07:07,560 --> 00:07:11,200
Imagine a robot trying to identify a shape on a table.

89
00:07:11,267 --> 00:07:17,752
In our toy design, its camera returns the same black frame when it is on in a dark room and

90
00:07:17,752 --> 00:07:19,680
when it is off in a lit room.

91
00:07:20,060 --> 00:07:23,520
The same image can therefore have two different causes.

92
00:07:23,560 --> 00:07:28,693
One concerns the environment. The other concerns the robot's own condition.

93
00:07:28,940 --> 00:07:32,640
Now give it a reliable record of whether its camera is on.

94
00:07:32,653 --> 00:07:37,840
In these two cases, it can distinguish the causes and choose an appropriate next step.

95
00:07:38,147 --> 00:07:42,896
We have returned to the card problem, but one of the relevant properties now belongs to the

96
00:07:42,896 --> 00:07:44,560
observing system itself.

97
00:07:44,860 --> 00:07:51,144
A proposed self-model would represent how the robot's own sensing process affects what it can

98
00:07:51,144 --> 00:07:54,480
find out. It would make predictions that we can check.

99
00:07:54,573 --> 00:08:00,080
In our lit-room case, the model predicts that switching its camera on will make the shape

100
00:08:00,080 --> 00:08:04,223
distinguishable. We can then compare that prediction with what happens.

101
00:08:04,223 --> 00:08:07,750
Switching the camera on changes the robot's state.

102
00:08:07,750 --> 00:08:13,840
Holding the possible scenes fixed, we can also compare what it could distinguish with its camera

103
00:08:13,840 --> 00:08:18,800
off and with it on. The card mathematics describes that comparison;

104
00:08:18,800 --> 00:08:21,800
the action moves the robot between those conditions.

105
00:08:21,830 --> 00:08:27,695
Knowing one's own limitations now has a concrete target: predicting when observations are

106
00:08:27,695 --> 00:08:30,880
insufficient for a task, and what could improve them.

107
00:08:34,760 --> 00:08:39,023
But an extra input and an effective self-model are different things.

108
00:08:39,023 --> 00:08:44,740
Giving one system camera status while hiding it from another would demonstrate the value of that

109
00:08:44,740 --> 00:08:49,160
information. It would not establish that a particular self-model is better.

110
00:08:49,293 --> 00:08:54,428
To test that further claim, we would compare specified models with the same available

111
00:08:54,428 --> 00:08:57,360
observations and comparable resources.

112
00:08:57,453 --> 00:09:03,100
We would vary room lighting and camera status independently, and test predictions after changing

113
00:09:03,100 --> 00:09:07,200
the camera's state, including cases outside their training examples.

114
00:09:07,400 --> 00:09:12,960
Can the explicit model better predict whether the next image will let it identify the shape?

115
00:09:12,984 --> 00:09:17,280
Does its confidence distinguish correct identifications from errors?

116
00:09:17,288 --> 00:09:23,280
Does it request another observation when that improves identification enough to justify the cost?

117
00:09:23,453 --> 00:09:28,000
It might succeed. It might add no benefit over a simpler model.

118
00:09:28,029 --> 00:09:31,480
Either outcome would teach us something about the proposal.

119
00:09:31,544 --> 00:09:36,240
This is a possible next step for The MIND's reflective-self-model direction.

120
00:09:36,307 --> 00:09:41,120
It is a research target, not a reported result of the current manuscripts.

121
00:09:44,000 --> 00:09:49,091
The point of the example is that a difference in the world does not automatically become a

122
00:09:49,091 --> 00:09:51,160
difference a system can use.

123
00:09:51,264 --> 00:09:55,627
Once we make that gap explicit, we can ask better questions.

124
00:09:55,627 --> 00:09:58,160
Which distinctions support the task?

125
00:09:58,160 --> 00:10:00,280
What evidence makes them available?

126
00:10:00,304 --> 00:10:03,380
What changes when a new test arrives?

127
00:10:03,380 --> 00:10:06,520
And can a system model the limits of its own access?

128
00:10:06,880 --> 00:10:12,700
B-Theory's tools make the available distinctions and supported claims explicit, so we can

129
00:10:12,700 --> 00:10:15,400
compare observation conditions precisely.

130
00:10:15,440 --> 00:10:21,728
The MIND programme proposes further models of how a system uses and changes those conditions,

131
00:10:21,728 --> 00:10:24,022
including its own sensing process.

132
00:10:24,022 --> 00:10:29,800
Connecting those models to actual minds requires further theory and evidence.

133
00:10:29,840 --> 00:10:33,160
The robot leaves us with two questions to explore.

134
00:10:33,200 --> 00:10:36,280
What can its specified evidence establish?

135
00:10:36,320 --> 00:10:42,896
Follow From observations to topology. How does the structure of its distinctions change

136
00:10:42,896 --> 00:10:45,771
when another test becomes available?

137
00:10:45,771 --> 00:10:51,907
Follow What changes when a new test is added. The MIND asks what further models and

138
00:10:51,907 --> 00:10:55,800
experiments could turn these tools into an account of self-monitoring.

139
00:10:55,872 --> 00:11:02,213
Then the four cards become a starting point for a larger investigation: what a system can

140
00:11:02,213 --> 00:11:08,600
distinguish, what it can do with those distinctions, and what it can find out about its own way of observing.