WEBVTT

1
00:00:00.200 --> 00:00:05.947
A digital label can change suddenly while the quantity underneath changes smoothly. Here

2
00:00:05.947 --> 00:00:11.213
Being is the position in the cycle now; Becoming is how that position evolves.

3
00:00:11.213 --> 00:00:16.333
To predict what comes next, we may need more than the current label or signal height.

4
00:00:16.333 --> 00:00:21.933
Watch the relationship between the evolving state, its coarse observation, and the role

5
00:00:21.933 --> 00:00:22.680
of history.

6
00:00:23.920 --> 00:00:26.140
Leave the finite machine behind.

7
00:00:26.140 --> 00:00:32.213
The next existing explainer plots a smoothly repeating input curve, called a sine wave.

8
00:00:32.213 --> 00:00:36.120
Phase means where we are around that repeating cycle.

9
00:00:36.120 --> 00:00:41.120
For our added prediction question, allow the starting phase to be unknown.

10
00:00:41.120 --> 00:00:49.500
A complete model can record that position around the cycle; the observer sees only the current height or its zero-or-one summary.

11
00:00:49.500 --> 00:00:57.840
The film first shows how an observation rule turns the input into a bit, then deliberately changes that rule to include memory.

12
00:00:58.630 --> 00:01:03.174
A system can change smoothly while its logical description changes suddenly.

13
00:01:03.174 --> 00:01:10.854
Here is a continuous input signal moving up and down. For this example, let its value be the sine of time.

14
00:01:10.854 --> 00:01:16.998
Name the observation rule before drawing the bit: report one when the value is at least zero,

15
00:01:16.998 --> 00:01:22.110
including exactly zero, and report zero when it is below zero.

16
00:01:22.110 --> 00:01:26.802
The signal moves continuously. The recorded bit flips at a threshold.

17
00:01:26.802 --> 00:01:31.610
This is a small starting point for the proposed Evolutionary Logic direction:

18
00:01:31.610 --> 00:01:36.463
how do continuously changing processes acquire discrete logical descriptions?

19
00:01:36.463 --> 00:01:39.317
There are three ingredients to keep separate.

20
00:01:39.317 --> 00:01:44.370
First, the signal trajectory: how the underlying quantity changes.

21
00:01:44.370 --> 00:01:50.836
Second, the observation map: the rule that turns that quantity into a reported value.

22
00:01:50.836 --> 00:01:56.716
Third, the resulting logical history: the sequence of zeros and ones.

23
00:01:56.716 --> 00:01:59.736
The threshold did not cause the sine wave to jump.

24
00:01:59.736 --> 00:02:02.756
It changed how a continuous value was described.

25
00:02:02.756 --> 00:02:07.576
And many different continuous trajectories could produce the same Boolean history.

26
00:02:07.576 --> 00:02:13.736
A one-bit description discards information about amplitudes and motion within each side of the threshold.

27
00:02:14.526 --> 00:02:17.886
We can already ask a precise question about stability.

28
00:02:17.886 --> 00:02:29.346
If the signal is well above zero, a small perturbation leaves the recorded bit unchanged. If the signal is very close to zero, a perturbation of the same size may flip it.

29
00:02:29.346 --> 00:02:43.406
For this exact rule, any perturbation smaller than the absolute value of a nonzero signal value cannot cross the threshold. At the threshold itself, arbitrarily small changes can alter the answer.

30
00:02:43.406 --> 00:02:52.606
The Boolean description is therefore robust in some regions and sensitive in others. That depends on both the signal value and the chosen observation rule.

31
00:02:53.926 --> 00:02:57.306
Suppose we want to avoid repeated flipping near zero.

32
00:02:57.306 --> 00:03:03.706
One possible alternative is hysteresis: a rule that remembers its previous output.

33
00:03:03.706 --> 00:03:07.016
Specify an initial output.

34
00:03:07.016 --> 00:03:13.893
Switch to one strictly above the positive threshold and to zero strictly below the negative threshold.

35
00:03:13.893 --> 00:03:18.064
At either threshold and between them, keep the previous output.

36
00:03:18.064 --> 00:03:25.240
That rule can ignore small oscillations around zero after an initial output has been specified.

37
00:03:25.240 --> 00:03:27.380
But we have changed the model.

38
00:03:27.380 --> 00:03:31.840
The output now depends on memory as well as the current continuous value.

39
00:03:31.840 --> 00:03:38.120
Two identical current values can produce different outputs if their histories differ.

40
00:03:38.120 --> 00:03:43.800
That is a feature we must state, not an invisible adjustment to the original threshold rule.

41
00:03:44.670 --> 00:03:51.270
A first formal study could choose a small dynamical system, define its observation resolution,

42
00:03:51.270 --> 00:03:53.970
and ask which properties survive the mapping.

43
00:03:53.970 --> 00:04:00.030
For example: when are observed transitions stable under a specified perturbation bound?

44
00:04:00.030 --> 00:04:05.198
Can a coarse observer predict the next recorded bit from its current recorded state?

45
00:04:05.198 --> 00:04:08.010
When does that require extra memory?

46
00:04:08.010 --> 00:04:14.850
A good worked model would include a case where the desired property holds and a boundary case where it fails.

47
00:04:14.850 --> 00:04:21.541
The current site presents this as a research direction, not as an already completed bridge from continuous

48
00:04:21.541 --> 00:04:22.990
dynamics to thought.

49
00:04:22.990 --> 00:04:27.190
The example establishes a modest but useful distinction.

50
00:04:27.190 --> 00:04:34.165
A continuous process, a discrete description, and a rule for updating that description are separate

51
00:04:34.165 --> 00:04:35.966
mathematical objects.

52
00:04:35.966 --> 00:04:43.130
Once we specify them, we can investigate their relationship instead of treating a threshold crossing as an

53
00:04:43.130 --> 00:04:44.830
explanation of cognition.

54
00:04:44.830 --> 00:04:51.330
That is a concrete way to approach Becoming: say what changes, say how it is observed,

55
00:04:51.330 --> 00:04:54.270
and test which features of the change remain visible.

56
00:04:55.060 --> 00:04:59.180
Choose thresholds minus one quarter and plus one quarter.

57
00:04:59.180 --> 00:05:04.360
Set memory to zero before the first input in each history.

58
00:05:04.360 --> 00:05:10.750
The first history processes minus one half, plus one half, then zero.

59
00:05:10.750 --> 00:05:16.370
The second processes plus one half, minus one half, then zero.

60
00:05:16.370 --> 00:05:25.018
What is the output after the final zero in each history?

61
00:05:25.018 --> 00:05:28.198
In the first it is one; in the second it is zero.

62
00:05:28.198 --> 00:05:35.698
The full output rows, after processing each input, are zero-one-one and one-zero-zero.

63
00:05:35.698 --> 00:05:40.906
The rule retained memory, so the same final input gave different outputs.

64
00:05:40.906 --> 00:05:44.698
This is a different observation rule from the original threshold bit.

65
00:05:46.018 --> 00:05:50.658
Now return to the original threshold rule without hysteresis.

66
00:05:50.658 --> 00:05:55.264
A repeating signal can be at the same value while rising or falling.

67
00:05:55.264 --> 00:06:02.130
Choose one point on the rising part of the cycle and one on the falling part, both at height one half.

68
00:06:02.130 --> 00:06:05.068
Both current bits are one.

69
00:06:05.068 --> 00:06:08.848
Both follow the same constant cycle rate.

70
00:06:08.848 --> 00:06:14.767
Advance both by one sixth of a full cycle, using the same elapsed time.

71
00:06:14.767 --> 00:06:20.687
The rising case reaches height one; the falling case reaches minus one half.

72
00:06:20.687 --> 00:06:24.860
Their next bits are therefore one and zero.

73
00:06:24.860 --> 00:06:28.989
The current value or bit alone did not specify the future.

74
00:06:28.989 --> 00:06:35.940
Remembering something can help only if the model says what memory is available and why it is sufficient.

75
00:06:36.490 --> 00:06:42.290
The hysteresis example changed the definition of the output by including its previous value.

76
00:06:42.290 --> 00:06:47.983
The phase example asked which description predicts a separately defined output.

77
00:06:47.983 --> 00:06:50.058
Those are different tasks.

78
00:06:50.058 --> 00:07:02.090
Next we move to R4, a finite actuator-tracking case with its own dynamics and observations, to see how a system maintains an estimate of its own changing machinery.