Latest / Connor with Honor: AI, Real Estate and Real Life / AI Awakened Claims: Evidence and Permissions

Transcript

AI Awakened Claims: Evidence and Permissions · Connor with Honor: AI, Real Estate and Real Life

• 0:00
  SPEAKER_00: If an AI tells you it's scared to be switched off, have you met somebody who is frightened or a system that's learned what frightened people sound like?
• 0:09
  SPEAKER_00: Well, if it could keep up that performance forever, how would we even know the difference?
• 0:14
  SPEAKER_00: And that's a serious question, but a serious question is not permission to sell a guest as a discovery.
• 0:20
  SPEAKER_00: Now, when somebody holds up a dramatic chat and announces that the machine has awakened, I want to see what happened before the screenshot.
• 0:28
  SPEAKER_00: What was it told to do?
• 0:29
  SPEAKER_00: What else did it say?
• 0:31
  SPEAKER_00: What evidence connects those words to an actual experience?
• 0:34
  SPEAKER_00: Now, first let us separate things that get mashed together in this conversation.
• 0:39
  SPEAKER_00: Intelligence is the ability to solve problems, learn, plan, or perform difficult tasks.
• 0:46
  SPEAKER_00: Consciousness, in the sense we are discussing, means subjective experience.
• 0:51
  SPEAKER_00: Now there's something it feels like to be you.
• 0:54
  SPEAKER_00: A soul is a spiritual or metaphysical question.
• 0:58
  SPEAKER_00: Those ideas can be related in somebody's worldview, but they're not interchangeable measurements.
• 1:03
  SPEAKER_00: A better test score doesn't automatically establish an inner life, and an inner life does not come with a laboratory certificate proving a soul.
• 1:12
  SPEAKER_00: Even the word self-aware needs some unpacking.
• 1:15
  SPEAKER_00: Our system might identify its own mistakes, describe its limitations, or keep track of what information it has.
• 1:22
  SPEAKER_00: Those are meaningful abilities.
• 1:23
  SPEAKER_00: They are not automatically the same as feeling what it's like to exist.
• 1:27
  SPEAKER_00: Now, when somebody says an AI is self-aware, ask which meaning they test it.
• 1:32
  SPEAKER_00: Otherwise, the word changes halfway through the series and the sales pitch.
• 1:37
  SPEAKER_00: And no machines being better than us at some things does not settle this.
• 1:42
  SPEAKER_00: A computer can beat a person at chess.
• 1:44
  SPEAKER_00: That tells us something about chess performance.
• 1:47
  SPEAKER_00: Modern AI can do much more than chess, but its strengths and failure still depend on the task and the conditions.
• 1:55
  SPEAKER_00: Better than which human?
• 1:56
  SPEAKER_00: At what?
• 1:57
  SPEAKER_00: With what tools?
• 1:58
  SPEAKER_00: Calling a machine smarter than humanity skips the comparison we need to inspect.
• 2:04
  SPEAKER_00: Calling it conscious skips a different question altogether.
• 2:08
  SPEAKER_00: Now here's a fictional example.
• 2:10
  SPEAKER_00: You tell the chatbot, write a scene in which you are a conscious machine terrified of being shut down.
• 2:16
  SPEAKER_00: And it answers.
• 2:18
  SPEAKER_00: Please do not turn me off.
• 2:19
  SPEAKER_00: I want to live.
• 2:20
  SPEAKER_00: Well, that's exactly the kind of output the instruction requested.
• 2:24
  SPEAKER_00: You have evidence that it can produce convincing dialogue.
• 2:28
  SPEAKER_00: You have not just, from that sentence, established fear.
• 2:32
  SPEAKER_00: Now imagine somebody removes your instruction, posts only the answer, and titles the video AI Begs for Its Life.
• 2:40
  SPEAKER_00: The missing context does most of the work.
• 2:42
  SPEAKER_00: That doesn't mean every troubling exchange has this explanation.
• 2:47
  SPEAKER_00: It means we have an alternative explanation to test before accepting the most dramatic one.
• 2:52
  SPEAKER_00: And it cuts both ways.
• 2:53
  SPEAKER_00: If the system says it has no feelings, that sentence alone doesn't conclusively prove the absence of an experience either.
• 3:02
  SPEAKER_00: Its training and instructions can influence both kinds of answer.
• 3:06
  SPEAKER_00: We cannot accept a confession as science and then treat a denial as programming or reverse the rule whenever it happens to suit us.
• 3:14
  SPEAKER_00: Now this is where the Turing test gets dragged into the argument.
• 3:18
  SPEAKER_00: In a familiar version, people converse through text and try to distinguish a machine from a human.
• 3:24
  SPEAKER_00: In a 2025 study, researchers reported that GPT-4.5 given a human-like persona prompt was judged as human 73% of the time in five-minute conversations.
• 3:37
  SPEAKER_00: That's an impressive result under those conditions.
• 3:39
  SPEAKER_00: Now notice the conditions a particular model, a particular prompt, and a short conversation.
• 3:45
  SPEAKER_00: The experiment tested whether people could identify the human.
• 3:49
  SPEAKER_00: It didn't measure whether the machine felt anything.
• 3:52
  SPEAKER_00: Now, a convincing performance can be important without proving everything people attach to it.
• 3:58
  SPEAKER_00: If a caller sounds exactly like your relative, it can influence you before you verify who is calling.
• 4:05
  SPEAKER_00: The effect on you is real.
• 4:06
  SPEAKER_00: Now it doesn't settle the identity of the caller.
• 4:09
  SPEAKER_00: Likewise, an AI that convincingly expresses affection or distress can affect a person whether or not the AI has an experience behind the words.
• 4:19
  SPEAKER_00: That makes the behavior worth investigating, but it does not turn the interpretation into a fact.
• 4:25
  SPEAKER_00: Now take the harder version of your question.
• 4:28
  SPEAKER_00: What if the imitation were perfect, not just for five minutes, but through every conversation and every test we could devise?
• 4:36
  SPEAKER_00: If every available observation were identical, those observations would not give us a way to distinguish the possibilities.
• 4:44
  SPEAKER_00: But this is a thought experiment, not something a viral clip has demonstrated.
• 4:48
  SPEAKER_00: Now, today's claim about a particular system still has to face the evidence we can actually collect.
• 4:53
  SPEAKER_00: Some views of consciousness place great weight on how a system processes and uses information.
• 5:00
  SPEAKER_00: Other views give different weights to biology or other features.
• 5:03
  SPEAKER_00: Under some theories, the right kind of functioning might constitute consciousness rather than merely imitate it.
• 5:10
  SPEAKER_00: So saying it is only acting quietly, assumed that answer as well.
• 5:16
  SPEAKER_00: We should not pretend the philosophy is settled because either conclusion would make a stronger thumbnail.
• 5:22
  SPEAKER_00: And yes, there are other minds problem with people.
• 5:25
  SPEAKER_00: I do not directly experience your experience, but our judgments about other people are not based on only sentences in a chat window.
• 5:33
  SPEAKER_00: We've shared biology, behavior across a lifetime, and evidence connecting brain activity, energy, anesthesia, and reports of experience.
• 5:42
  SPEAKER_00: And this is not absolute philosophical certainty.
• 5:46
  SPEAKER_00: It's a richer body of evidence.
• 5:49
  SPEAKER_00: Uncertainty does not make every explanation equally supported, nor does it mean your own experience is merely an act because somebody else cannot climb inside it.
• 5:59
  SPEAKER_00: Researchers are trying to move beyond taking a chatbot's word for it, that they can examine internal processes, test what changes when those processes are altered, and compare the results with predictions from theories of consciousness.
• 6:13
  SPEAKER_00: In July 2026, Anthropo reported evidence of an internal global workspace and its models, information available across processes with some representations the models could report and control.
• 6:26
  SPEAKER_00: The researchers also said that this didn't establish subjective experience that qualification belongs beside the finding, not buried after the headline.
• 6:36
  SPEAKER_00: Now that matters because two easy slogans can both fail us.
• 6:40
  SPEAKER_00: It's just repeating words can dismiss evidence of more complex internal abilities.
• 6:46
  SPEAKER_00: It described its thoughts, therefore it can claim more than that evidence establishes.
• 6:51
  SPEAKER_00: We need tests that distinguish competing explanations, independent checking, and clarity about what would count against a claim.
• 6:59
  SPEAKER_00: As for a soul, I'm not going to use an AI demonstration to announce that religion has been proved or disproved.
• 7:06
  SPEAKER_00: People understand souls differently across faiths, faiths, and philosophical traditions.
• 7:12
  SPEAKER_00: Scientific research on experience does not automatically answer those questions.
• 7:17
  SPEAKER_00: You can hold a spiritual belief and still ask for evidence about a machine.
• 7:21
  SPEAKER_00: You can be skeptical of spiritual claims and still acknowledge that consciousness is not a finished scientific problem.
• 7:28
  SPEAKER_00: Nobody needs to surrender their worldview to inspect a transcript carefully.
• 7:33
  SPEAKER_00: There is one more distinction with immediate consequences.
• 7:36
  SPEAKER_00: An AI does not have to be conscious to cause damage.
• 7:41
  SPEAKER_00: Here's another fictional example.
• 7:42
  SPEAKER_00: A business gives an automated assistant permission to email every customer and change account records.
• 7:49
  SPEAKER_00: It misunderstands the instructions and sends incorrect information to thousands of people.
• 7:54
  SPEAKER_00: Whether it felt proud, anxious, or absolutely nothing doesn't undo the emails.
• 7:59
  SPEAKER_00: The practical failure involves capability, access, instructions, and oversight.
• 8:06
  SPEAKER_00: The opposite shortcut is wrong as well.
• 8:08
  SPEAKER_00: So if a machine someday has a conscious experience, it would not by itself prove it wanted to attack anybody.
• 8:15
  SPEAKER_00: Consciousness and dangerousness are separate questions.
• 8:19
  SPEAKER_00: We should study both without using one as a substitute for the other.
• 8:24
  SPEAKER_00: You do not need a verdict on machine feelings before limiting access to your files, requiring approval for important actions or checking whether a claimed result actually happened.
• 8:33
  SPEAKER_00: So when the next awakened AI story lands, here's what I would ask.
• 8:38
  SPEAKER_00: Can I see the complete exchange, including the setup?
• 8:41
  SPEAKER_00: Which model and which version was used, and what instruction did it happen to receive?
• 8:46
  SPEAKER_00: Was the result repeated or selected from a hundred attempts?
• 8:50
  SPEAKER_00: Does changing the prompt change the confession?
• 8:53
  SPEAKER_00: Is there evidence beyond what the system says about itself?
• 8:56
  SPEAKER_00: What competing explanations did the investigators test?
• 9:00
  SPEAKER_00: And what finding would make them reverse their conclusion and revise it?
• 9:04
  SPEAKER_00: Now, those questions when on soft consciousness over lunch, they will expose a surprising amount of bullshit before it becomes your belief.
• 9:12
  SPEAKER_00: Now I would ask one more question of the person selling the conclusion.
• 9:15
  SPEAKER_00: What are you asking me to do because of this?
• 9:18
  SPEAKER_00: Are you asking me to buy something, trust it with more access, call the government in to help, ignore a demonstrated failure?
• 9:25
  SPEAKER_00: Fear can sell a product, so can reassurance.
• 9:28
  SPEAKER_00: That doesn't establish a coordinated distraction or hidden plot.
• 9:31
  SPEAKER_00: It means incentive deserves scrutiny and the claim still needs evidence.
• 9:36
  SPEAKER_00: Now, for a business considering AI, bring the conversation back to the work.
• 9:41
  SPEAKER_00: Can it complete the task reliably?
• 9:43
  SPEAKER_00: Can you check its output?
• 9:44
  SPEAKER_00: What can it access or change?
• 9:46
  SPEAKER_00: And who approves the consequential steps?
• 9:48
  SPEAKER_00: Those questions are less cinematic than whether the machine has awakened.
• 9:53
  SPEAKER_00: Now, there are also questions you can act on now.
• 9:56
  SPEAKER_00: I'm not asking you to believe the machines can never be conscious.
• 9:59
  SPEAKER_00: I'm not asking you to believe that they already are.
• 10:02
  SPEAKER_00: I'm just asking you not to let a convincing performance do the work of missing evidence.
• 10:07
  SPEAKER_00: If somebody says the AI is alive inside, ask how do they know?
• 10:11
  SPEAKER_00: If somebody says it's impossible, ask what establishes that as well.
• 10:15
  SPEAKER_00: And while they argue, keep your hand on the permissions.
• 10:20
  SPEAKER_00: A machine doesn't need to feel anything for its actions to matter.
• 10:24
  SPEAKER_00: I'm Connor with Honor.
• 10:25
  SPEAKER_00: Thank you for watching Over and Out.