Latest / Connor with Honor: AI, Real Estate and Real Life / "The AI Did It." Then Who Signed Off? No Human Left Behind

Transcript

"The AI Did It." Then Who Signed Off? No Human Left Behind · Connor with Honor: AI, Real Estate and Real Life

• 0:00
  SPEAKER_00: Let me give you the question I think belongs in every serious conversation about artificial intelligence.
• 0:05
  SPEAKER_00: When this technology makes a business faster, makes somebody wealthy, or changes the way we work, what happens to the human being on the other side of that decision?
• 0:14
  SPEAKER_00: I come at this as a former LAPD officer, an instructor, real estate professional, and somebody who builds AI automations.
• 0:21
  SPEAKER_00: Now, I want these tools to work.
• 0:22
  SPEAKER_00: Genuinely I do.
• 0:24
  SPEAKER_00: There's useful work that we can do with them, but I also want somebody willing to ask the uncomfortable questions before a family, a customer, or an employee pays for an assumption nobody bothered to check.
• 0:34
  SPEAKER_00: My biggest concern is how easily we can hide a human decision behind the words.
• 0:39
  SPEAKER_00: The AI did it.
• 0:40
  SPEAKER_00: Somebody selected the system.
• 0:42
  SPEAKER_00: Somebody approved its access.
• 0:44
  SPEAKER_00: Somebody decided what success actually meant.
• 0:47
  SPEAKER_00: And somebody needs to answer for those choices.
• 0:50
  SPEAKER_00: They're real technical failures too, including behavior the people deploying a system didn't intend.
• 0:56
  SPEAKER_00: We have to take both very seriously.
• 0:58
  SPEAKER_00: And here's the framework I want you to carry through this whole conversation.
• 1:01
  SPEAKER_00: Who is making the claim?
• 1:03
  SPEAKER_00: What do they want?
• 1:04
  SPEAKER_00: What evidence are they showing?
• 1:06
  SPEAKER_00: And how can we verify it independently?
• 1:08
  SPEAKER_00: Ask that of an AI company, ask that of a politician, ask that of a person selling a course about how to survive AI.
• 1:16
  SPEAKER_00: And of course, ask it of me.
• 1:17
  SPEAKER_00: A strong claim ought to survive a few questions.
• 1:20
  SPEAKER_00: So let's start with work because this gets personal very quickly.
• 1:24
  SPEAKER_00: When somebody says AI can do part of your job, the next question is which part?
• 1:29
  SPEAKER_00: Under what conditions?
• 1:31
  SPEAKER_00: And with how much checking.
• 1:34
  SPEAKER_00: Writing a first draft, handling a routine request, and carrying responsibility for customers are different things.
• 1:41
  SPEAKER_00: Task demonstration doesn't tell you everything about a person's job.
• 1:45
  SPEAKER_00: The International Labor Organization Organization's research makes an important decision distinction here.
• 1:51
  SPEAKER_00: Exposure to generative AI doesn't automatically mean a job disappears.
• 1:56
  SPEAKER_00: Many jobs contain a miss a mix of tasks and the organization's release points, research points as well, towards substantial job transformation.
• 2:06
  SPEAKER_00: Picture an office with 10 people handling customer questions.
• 2:10
  SPEAKER_00: A new system takes over some routine work.
• 2:13
  SPEAKER_00: Maybe those people can now solve harder problems.
• 2:16
  SPEAKER_00: Maybe the business grows.
• 2:17
  SPEAKER_00: Maybe management cuts positions.
• 2:20
  SPEAKER_00: Maybe the system disappoints and the remaining staff spend their afternoons repairing mistakes.
• 2:25
  SPEAKER_00: Before celebrating, I'd want to see what actually happened to service quality, workload, pay, and the people who worked there.
• 2:33
  SPEAKER_00: Now I'm not going to call every business owner greedy because they're trying to survive.
• 2:38
  SPEAKER_00: Small businesses face real costs.
• 2:40
  SPEAKER_00: Consumers want prices, lower prices.
• 2:44
  SPEAKER_00: Competitors adopt newer tools.
• 2:46
  SPEAKER_00: These pressures do matter, but pressure doesn't erase responsibility.
• 2:50
  SPEAKER_00: If a company expects to benefit from changing someone's livelihood, asking for a serious transition plan is very fair.
• 2:59
  SPEAKER_00: What bothers me is the casual advice to just learn AI as if that settles it.
• 3:03
  SPEAKER_00: Learn what?
• 3:04
  SPEAKER_00: For which job?
• 3:05
  SPEAKER_00: Who's hiring?
• 3:06
  SPEAKER_00: Who pays for the training?
• 3:08
  SPEAKER_00: When a parent working two jobs, when are they supposed to do it?
• 3:12
  SPEAKER_00: And how does the rent get paid between the old paycheck and the new one?
• 3:15
  SPEAKER_00: Those are ordinary questions with real consequences.
• 3:19
  SPEAKER_00: A link to a course doesn't answer them.
• 3:22
  SPEAKER_00: Now that's why I want to transition moonshot.
• 3:24
  SPEAKER_00: You know, moonshots, you've heard of these, these uh organizational abilities where they put people together and they solve big problems and pay them a lot of money to do so.
• 3:33
  SPEAKER_00: A serious effort with that same urgency we bring to that building of technology, we ought to be able to explain how people move from work, that shrinking into work somebody will actually pay them to do.
• 3:45
  SPEAKER_00: No human left behind should be the goal we measure ourselves against.
• 3:50
  SPEAKER_00: Now, here's what I'd ask employers to put on the table: paid time to learn tools relevant to work, a clear description of which responsibilities are changing, a fair chance to move into available roles, support for people whose positions really are going away, and a way to tell whether the plan helps somebody keep earning a living.
• 4:09
  SPEAKER_00: Now, training has to connect to opportunity.
• 4:11
  SPEAKER_00: If the course teaches one thing and employers need something else, the worker's still stuck.
• 4:17
  SPEAKER_00: If the promised new job pays much less, we need to acknowledge that as well.
• 4:22
  SPEAKER_00: If somebody needs transportation, childcare, basic computer help to participate, the plan has to account for reality.
• 4:29
  SPEAKER_00: Otherwise, we're congratulating ourselves for offering something they can't use.
• 4:33
  SPEAKER_00: I'd want business, educators, local communities, and governments asking those questions together.
• 4:40
  SPEAKER_00: Their costs and trade-offs.
• 4:41
  SPEAKER_00: No single program is going to guarantee a good outcome for everybody, but we can judge proposals by concrete results.
• 4:48
  SPEAKER_00: Did people get hired?
• 4:49
  SPEAKER_00: Did they stay employed?
• 4:50
  SPEAKER_00: What happened to their earnings?
• 4:52
  SPEAKER_00: And what did we learn from the people the program failed to help?
• 4:56
  SPEAKER_00: And we keep an eye on the first rung of the ladder.
• 4:58
  SPEAKER_00: If we automate the work beginners used to learn on, how do they become experienced?
• 5:03
  SPEAKER_00: In real estate, a new person needs more than a polished description of a house.
• 5:07
  SPEAKER_00: They need to learn how to notice missing information, handle a worried client, and recognize when they need help.
• 5:13
  SPEAKER_00: Somebody has to teach that judgment.
• 5:15
  SPEAKER_00: And that brings me to scams because technology can make an old con look very convincing.
• 5:21
  SPEAKER_00: The FBI has warned us about criminals using generated audio images and video to support fraud.
• 5:26
  SPEAKER_00: A familiar voice can be imitated.
• 5:29
  SPEAKER_00: An unbelievable picture can be manufactured.
• 5:31
  SPEAKER_00: The emotional pressure can feel very real even when that story's false.
• 5:36
  SPEAKER_00: So imagine a call that sounds like somebody you love.
• 5:39
  SPEAKER_00: There's an emergency.
• 5:40
  SPEAKER_00: They need money immediately.
• 5:41
  SPEAKER_00: It's always time sensitive.
• 5:42
  SPEAKER_00: And they're telling you not to call anyone else.
• 5:45
  SPEAKER_00: The combination of fear, urgency, and isolation deserves a pause.
• 5:50
  SPEAKER_00: In that conversation and contact the person through a number you already trust.
• 5:56
  SPEAKER_00: If you can't reach them, use another established contact to check the story.
• 6:00
  SPEAKER_00: In a real estate transaction, imagine receiving last-minute instructions to send money somewhere different.
• 6:05
  SPEAKER_00: Maybe the message looks polished.
• 6:07
  SPEAKER_00: Maybe the voice sounds familiar.
• 6:08
  SPEAKER_00: Before acting, verify through the trusted contact information established earlier.
• 6:13
  SPEAKER_00: Don't let the message requesting the change supply its own proof.
• 6:17
  SPEAKER_00: The same principle applies to a supposed boss that you that asking you to go buy some gift cards.
• 6:23
  SPEAKER_00: You don't have to become a deepfake expert to build a better habit.
• 6:26
  SPEAKER_00: You just need to separate route and verification.
• 6:30
  SPEAKER_00: Now, a family code word can help, but use it as one layer.
• 6:34
  SPEAKER_00: Keep it private, still check the circumstances.
• 6:37
  SPEAKER_00: The goal is to keep a convincing performance from becoming permission to take your money.
• 6:42
  SPEAKER_00: Now, privacy and permissions.
• 6:44
  SPEAKER_00: This is where enthusiasm can get expensive.
• 6:46
  SPEAKER_00: An app asks you to connect your email, your files, your contacts, and your calendar.
• 6:51
  SPEAKER_00: You want the feature, so you click through.
• 6:54
  SPEAKER_00: Stop there for just a second.
• 6:55
  SPEAKER_00: What can it read?
• 6:56
  SPEAKER_00: What can it change?
• 6:57
  SPEAKER_00: Can it send something as you?
• 6:59
  SPEAKER_00: How do you disconnect it?
• 7:01
  SPEAKER_00: Now suppose I want help drafting a property description.
• 7:04
  SPEAKER_00: That task should make us ask why the tool needs an entire folder of client records.
• 7:10
  SPEAKER_00: A buyer's financial information documents, financial documents don't belong in an experiment just because uploading them is convenient.
• 7:18
  SPEAKER_00: Use only the information the task actually needs in a service you've checked and are authorized to use for that information.
• 7:24
  SPEAKER_00: Now there's also a difference between an assistant suggesting something and an assistant doing it.
• 7:29
  SPEAKER_00: Drafting an email and sending an email, deleting a file and authorizing a payment carry different consequences.
• 7:36
  SPEAKER_00: Access should match that job.
• 7:38
  SPEAKER_00: A tool that helps with one small task shouldn't automatically inherit the keys to everything else you do.
• 7:44
  SPEAKER_00: Now, security authorities are warning that AI agents can behave unexpectedly or be manipulated by instructions hidden in the material they read.
• 7:52
  SPEAKER_00: Now that's one reason simply telling a system to be careful is insufficient.
• 7:56
  SPEAKER_00: Restrict its access, require approval for consequential actions, and of course keep records, and have a way to stop it.
• 8:03
  SPEAKER_00: Controls need to exist outside the model's promises.
• 8:07
  SPEAKER_00: Now here's the everyday question.
• 8:09
  SPEAKER_00: The thing that makes a bad decision while I'm getting coffee, how much damage can it do before I notice?
• 8:14
  SPEAKER_00: Now, if the answer is it could send, it could email every client, erase important records, and expose private information that deserves attention before lunch, a person should know how to shut it down and what to do afterward.
• 8:28
  SPEAKER_00: Next concern is convincing wrong answers.
• 8:30
  SPEAKER_00: Now, NIST identifies confidently standard started false content is a generative AI risk.
• 8:38
  SPEAKER_00: You may hear it called a hallucination.
• 8:40
  SPEAKER_00: But whatever word you use, the practical problem is simple.
• 8:43
  SPEAKER_00: An answer can sound finished while an important detail is absolutely incorrect.
• 8:48
  SPEAKER_00: Good spelling and a professional tone don't establish accuracy.
• 8:51
  SPEAKER_00: Now take a house listing.
• 8:53
  SPEAKER_00: If AI fills it in an improvement you never verify, you could publish a claim that you cannot support.
• 9:00
  SPEAKER_00: If it summarizes an inspection and drops a critical qualification, the shorter version can change what a buyer understands.
• 9:07
  SPEAKER_00: Those are examples of why the original document and the responsible professional still matter.
• 9:12
  SPEAKER_00: Ask where the answer came from.
• 9:14
  SPEAKER_00: You want to open the source, make sure the source actually says what the answer claims.
• 9:18
  SPEAKER_00: Check the date, look for the sentence that changes the meaning, and a second chat bot agreing can be useful feedback, maybe by a different LLM company.
• 9:28
  SPEAKER_00: But agreement alone isn't independent verification.
• 9:31
  SPEAKER_00: Now, both systems can repeat the same error or draw on the same flawed material.
• 9:35
  SPEAKER_00: And match the checking to the stakes.
• 9:38
  SPEAKER_00: Brainstorming a headline is one thing.
• 9:40
  SPEAKER_00: Making a consequential decision about a contract, someone's health or the finances demands appropriate expertise and direct evidence.
• 9:48
  SPEAKER_00: The more damage an error could cause, the more deliberate the review needs to be.
• 9:52
  SPEAKER_00: So speed is useful when the result is worth using.
• 9:55
  SPEAKER_00: Now, the bigger technical question people ask, what happens when AI helps build more powerful AI?
• 10:01
  SPEAKER_00: You're going to hear the term recursive self-improvement a lot in the very near future, and even today started breaking out.
• 10:08
  SPEAKER_00: Broadly, that concern is a loop where the systems help create more capable successors, which can then contribute to further improvement.
• 10:19
  SPEAKER_00: There are important differences between assisting a researcher and independently driving that whole process.
• 10:24
  SPEAKER_00: Anthropic has published evidence that its systems are helping with its own AI development.
• 10:30
  SPEAKER_00: That's worth examining.
• 10:31
  SPEAKER_00: Google released some stuff today on that as well.
• 10:33
  SPEAKER_00: It's also a company describing its own technology, so apply the same investigator questions.
• 10:38
  SPEAKER_00: What was measured?
• 10:39
  SPEAKER_00: What did people still do?
• 10:42
  SPEAKER_00: What results, which results were independently checked, what happens outside the test conditions.
• 10:47
  SPEAKER_00: A company report is evidence to examine.
• 10:50
  SPEAKER_00: Now, in that discussion, Anthropic says full recursive self-improvement hasn't been reached and isn't inevitable.
• 10:57
  SPEAKER_00: Personally, I think we're just a couple decisions away from that.
• 11:00
  SPEAKER_00: So I'm not going to hand you a countdown and pretend it's settled, but progress could accelerate and it could be just a couple decisions away.
• 11:07
  SPEAKER_00: I could encounter, it could also encounter serious bottlenecks.
• 11:10
  SPEAKER_00: The timing and the outcome remain uncertain.
• 11:13
  SPEAKER_00: That uncertainty is a reason to prepare and test rather than make promises we can't defend.
• 11:18
  SPEAKER_00: We should also reject the idea that an autonomous failure always requires a malicious person giving a malicious instruction.
• 11:26
  SPEAKER_00: A system can pursue a goal in an unintended way.
• 11:29
  SPEAKER_00: It can act on bad information, misunderstand limits, or be manipulated.
• 11:33
  SPEAKER_00: Greater capability can increase the consequences.
• 11:36
  SPEAKER_00: Human accountability takes taking, includes taking those technical possibilities seriously before granting more power.
• 11:43
  SPEAKER_00: For more powerful systems, I want strong evaluations, independent scrutiny, and of course meaningful access limits and clear conditions for stopping deployment.
• 11:53
  SPEAKER_00: Who can call a halt?
• 11:54
  SPEAKER_00: What evidence triggers it?
• 11:56
  SPEAKER_00: Can that person actually exercise the authority?
• 11:59
  SPEAKER_00: A safety policy that disappears whenever a deadline arrives isn't much comfort to the people exposed to the risk.
• 12:05
  SPEAKER_00: And that leads to competitive pressure.
• 12:08
  SPEAKER_00: A company worries about arrival that will launch first.
• 12:12
  SPEAKER_00: A government worries about another country.
• 12:13
  SPEAKER_00: A manager worries that the department next door will report bigger savings or earnings.
• 12:18
  SPEAKER_00: Those incentives can push people to move before they're ready.
• 12:22
  SPEAKER_00: We should recognize that pressure and still demand evidence that the proposed use is safe enough.
• 12:28
  SPEAKER_00: Ask a vendor what failed during testing.
• 12:30
  SPEAKER_00: Ask what a system must never be allowed to do.
• 12:33
  SPEAKER_00: Ask what happens when you disconnect it.
• 12:35
  SPEAKER_00: Ask who gets called after an incident.
• 12:37
  SPEAKER_00: Those answers tell you more about operational readiness than a perfect demonstration.
• 12:41
  SPEAKER_00: A demo is arranged to show you something working.
• 12:45
  SPEAKER_00: Your business house to function on the messy days also.
• 12:48
  SPEAKER_00: I'm not asking you to give up on useful AI.
• 12:51
  SPEAKER_00: Use it to organize information, explore an idea, handle appropriate routine work, or help you get started.
• 12:57
  SPEAKER_00: Learn its limits as you learn its strengths.
• 12:59
  SPEAKER_00: Build your own ability to check the result.
• 13:01
  SPEAKER_00: You can be excited about a tool and demand that it earn your trust.
• 13:06
  SPEAKER_00: Now, for someone worried about their job, start with the work you actually do, which tasks could change, which require relationships, judgment, responsibility, or knowledge of a specific situation.
• 13:18
  SPEAKER_00: Try a low-risk tool with information you're allowed to use.
• 13:22
  SPEAKER_00: Learn one useful workflow, then have a concrete conversation with your employer about training and what they're planning.
• 13:29
  SPEAKER_00: You deserve clearer answers than just adapt.
• 13:32
  SPEAKER_00: Now, for the person making the business decisions, put the people question on the agenda before the announcement.
• 13:39
  SPEAKER_00: Who benefits, who carries the risk, who gets training, who can change an error or challenge an error, who owns the response when something breaks.
• 13:48
  SPEAKER_00: If a proposal can explain the savings in detail, it ought to explain the consequences for people with equal care.
• 13:55
  SPEAKER_00: My faith is part of how I look at human dignity.
• 13:58
  SPEAKER_00: A person's worth doesn't shrink because software can perform one of their tasks.
• 14:02
  SPEAKER_00: That's a standard we can bring into the ordinary workplace conversation.
• 14:06
  SPEAKER_00: Now here's where I land.
• 14:08
  SPEAKER_00: I want powerful tools that help people.
• 14:10
  SPEAKER_00: I want evidence behind the claims, limits around the permissions, and accountable human beings making the consequential decisions.
• 14:17
  SPEAKER_00: I want a transition effort, serious enough to stand beside the investment in the technology.
• 14:22
  SPEAKER_00: And when somebody tells us all of this is inevitable, I want to ask which parts are predictions and which parts are choices.
• 14:30
  SPEAKER_00: Who's making the claim?
• 14:31
  SPEAKER_00: What do they want?
• 14:32
  SPEAKER_00: What is the evidence?
• 14:33
  SPEAKER_00: Can we verify it independently?
• 14:35
  SPEAKER_00: Keep asking the question of what AI can do is going to keep changing.
• 14:39
  SPEAKER_00: Our responsibility to the people affected deserves to stay in the conversation.
• 14:44
  SPEAKER_00: No human left behind.
• 14:45
  SPEAKER_00: I'm Connor with Honor, over and out.