WEBVTT

1
00:00:00.200 --> 00:00:06.552
In this series, Being is the system's state now. Becoming is how that state changes. We

2
00:00:06.552 --> 00:00:08.714
have used models to describe both.

3
00:00:08.714 --> 00:00:14.328
Before trusting a model's explanation or prediction about a real system, we need to ask

4
00:00:14.328 --> 00:00:19.387
what has actually been established. A correct calculation alone does not show that the

5
00:00:19.387 --> 00:00:20.880
model fits the world.

6
00:00:20.880 --> 00:00:27.096
Use this final episode to separate assumptions, derived results and empirical evidence.

7
00:00:27.096 --> 00:00:30.440
Keep asking: which claims still need a test?

8
00:00:31.680 --> 00:00:34.027
We have used several kinds of statements.

9
00:00:34.027 --> 00:00:43.460
We specified small models, derived consequences, checked calculations and examined a reported finite-model comparison.

10
00:00:43.460 --> 00:00:49.560
A proposed application to a person or another machine adds a correspondence claim.

11
00:00:49.560 --> 00:00:55.436
Evidence that supports one of these statements does not automatically establish all the others.

12
00:00:56.196 --> 00:01:00.756
In M0, u and v have different after-PULSE outputs.

13
00:01:00.756 --> 00:01:07.976
For w and z, either no RESET occurs or the first RESET brings both cases to u.

14
00:01:07.976 --> 00:01:13.149
That argument supports agreement for every finite sequence of the supplied buttons.

15
00:01:13.149 --> 00:01:17.276
Local partition and path checks corroborated the calculation.

16
00:01:17.276 --> 00:01:25.976
Testing a finite selection of paths alone would not justify every future sequence, and none of these checks observed a biological mind.

17
00:01:26.766 --> 00:01:36.466
The noisy-reader model M1’s updated probabilities followed from a supplied prior, sensor likelihoods and independence assumptions.

18
00:01:36.466 --> 00:01:39.986
We did not measure that reader in a laboratory.

19
00:01:39.986 --> 00:01:49.686
R4 used inherited finite-benchmark evidence; its film reports model-average failures and limited policy comparisons.

20
00:01:49.686 --> 00:01:54.299
These examples can explain mechanisms and bounded results.

21
00:01:54.299 --> 00:02:02.219
They do not by themselves validate a general theory of cognition or show that a self-model label improves an architecture.

22
00:02:02.929 --> 00:02:09.422
Suppose we propose that a representation predicts which differences a particular observer can use.

23
00:02:09.422 --> 00:02:15.349
First specify what the states and observations represent in that system.

24
00:02:15.349 --> 00:02:20.889
Declare the task, available information and measured outcome.

25
00:02:20.889 --> 00:02:28.419
Compare a relevant alternative under matched access and any resource constraints the claim requires.

26
00:02:28.419 --> 00:02:34.779
Keep some cases out of model development, and state what result would count against the claim.

27
00:02:34.779 --> 00:02:40.091
This is a proposal for evaluation, not a study conducted by this playlist.

28
00:02:40.931 --> 00:02:51.211
A valid derivation, a correct implementation and a useful prediction answer different questions from what is new and what review a claim has received.

29
00:02:51.211 --> 00:02:54.811
A local AI check is not external peer review.

30
00:02:54.811 --> 00:02:59.857
A source page or an attractive video is not an independent replication.

31
00:02:59.857 --> 00:03:06.371
Record the actual provenance and review status rather than inferring them from presentation quality.

32
00:03:07.161 --> 00:03:12.665
Classify this statement: our program reproduced the stated posterior calculation.

33
00:03:12.665 --> 00:03:21.281
Does it establish the sensor’s accuracy in a real robot?

34
00:03:21.281 --> 00:03:25.001
No. It supports that calculation under the supplied model.

35
00:03:25.001 --> 00:03:29.237
The reader’s real accuracy would require relevant measurements.

36
00:03:29.237 --> 00:03:39.601
Likewise, successful task prediction does not by itself establish subjective experience; that is a further claim, not another name for the prediction score.

37
00:03:40.391 --> 00:03:44.151
Being and Becoming now gives us a set of precise questions.

38
00:03:44.151 --> 00:03:47.551
What is the state, and how is it described?

39
00:03:47.551 --> 00:03:52.039
What can be observed, and which distinctions matter for a task?

40
00:03:52.039 --> 00:03:54.892
Which experiments can obtain them?

41
00:03:54.892 --> 00:03:59.592
How do actions change the state, and how does a model keep up?

42
00:03:59.592 --> 00:04:04.232
We have worked bounded examples and marked the next empirical questions.

43
00:04:04.232 --> 00:04:14.652
Use the optional branches when a specific problem calls for their mathematics, and keep each further claim as explicit as the examples here.