Latest / Humanity Working / How Do You Incentivize Teams to Architect Their Own Extinction?

Transcript

How Do You Incentivize Teams to Architect Their Own Extinction? · Humanity Working

• 0:14
  SPEAKER_06: Hi guys.
• 0:15
  SPEAKER_06: Paul here with Matt and Bill.
• 0:18
  SPEAKER_06: Have you been, Bill?
• 0:21
  SPEAKER_05: Great to be back.
• 0:22
  SPEAKER_05: It's been a week, man.
• 0:23
  SPEAKER_06: That's all I gotta say.
• 0:24
  SPEAKER_06: It's been a week.
• 0:27
  SPEAKER_06: I think in general, right?
• 0:28
  SPEAKER_06: We seem to be saying that pretty much every week right now.
• 0:31
  SPEAKER_05: Yeah.
• 0:32
  SPEAKER_05: I feel like just when we think we've reached the apex of the onslaught of news and headlines, we wake up to another one.
• 0:39
  SPEAKER_05: So I hope people are getting comfortable and have some sort of beverage nearby.
• 0:45
  SPEAKER_06: And then how have you been back?
• 0:49
  SPEAKER_01: Yeah, again, crazy week.
• 0:51
  SPEAKER_01: It's like the dog years week.
• 0:53
  SPEAKER_01: Seems like it's a month or two and it's only still not over.
• 0:57
  SPEAKER_01: So yeah, been crazy.
• 1:01
  SPEAKER_06: Yeah.
• 1:01
  SPEAKER_06: I mean, you know, people talk about people being head in the sand over this stuff.
• 1:07
  SPEAKER_06: And I have criticized many people for being very head in the sand when it comes to AI.
• 1:17
  SPEAKER_06: And I think the last couple of weeks I've kind of got my head around why people have been head in the sand about this stuff.
• 1:24
  SPEAKER_06: And maybe they're a lot wiser than I give them credit for.
• 1:26
  SPEAKER_06: Although we will get into some of those issues today.
• 1:29
  SPEAKER_06: I really want to break today's chat down actually into two things, guys.
• 1:34
  SPEAKER_06: One of the things I want to talk about is a little bit of what's going on right now, even though it'll sound horribly horribly dated once we get into it.
• 1:41
  SPEAKER_06: And then secondly, I know that both of you have been looking about a theme that we alluded to in our last recording, which was really looking at if we're changing the world of work and we're expecting to automate elements of it, that potentially results in fewer people doing work.
• 2:09
  SPEAKER_06: And the question then becomes in Britain we would say, how do you get turkeys to vote for Christmas?
• 2:16
  SPEAKER_06: Which doesn't really make a lot of sense in Britain.
• 2:18
  SPEAKER_06: Because sorry, it makes a lot of sense in Britain.
• 2:20
  SPEAKER_06: Doesn't really make a lot of sense in America because you you guys don't have turkey at Christmas.
• 2:24
  SPEAKER_06: But for people that are listening to this from the UK, it will make instant sense.
• 2:28
  SPEAKER_06: For those of you living in the US, the British have their turkeys at Christmas because we don't have Thanksgiving.
• 2:33
  SPEAKER_06: And you've still got, I don't know, turkeys to eat.
• 2:36
  SPEAKER_06: Does anybody look just like apropos of nothing?
• 2:39
  SPEAKER_06: Do either of you like turkey?
• 2:43
  SPEAKER_01: I happen to really like turkey, actually.
• 2:45
  SPEAKER_01: A lot of people don't all year.
• 2:49
  SPEAKER_01: I love it at the holidays.
• 2:51
  SPEAKER_05: I'm not such a huge fan.
• 2:53
  SPEAKER_06: My son's very wise about this.
• 2:54
  SPEAKER_06: He's like, why do we eat turkey?
• 2:56
  SPEAKER_06: It's supposed to be a special day, and turkey sucks.
• 3:01
  SPEAKER_06: I was like, I don't know, can't really argue with that.
• 3:03
  SPEAKER_06: Shut up, some, we do.
• 3:07
  SPEAKER_06: It's about the sun.
• 3:07
  SPEAKER_06: It's even worse for him because you have turkey at Thanksgiving, and then we go home to my family for Christmas, and then he has to have turkey again before it's basically the same meal twice.
• 3:17
  SPEAKER_06: So he's developed a special antipathy for it.
• 3:19
  SPEAKER_06: But anyway, so we'll we'll get into both of those things.
• 3:22
  SPEAKER_06: But before we get into the turkeys voting for Christmas bit, everybody's excited by that.
• 3:26
  SPEAKER_06: Before we get into that, we're recording this what the day after the various leaders were convened at what some people have called the Golden Age summit.
• 3:40
  SPEAKER_06: I've heard other people talk about it as the Gilded Age or the Gilded Era, with all of the negative connotations associated with that summit.
• 3:47
  SPEAKER_06: But the one where effectively we supposedly signed what I heard referred to as a moral agreement.
• 3:54
  SPEAKER_06: A moral agreement for AI with the various organizations appearing to agree to A regulate themselves and B call the thing they've been working on by a different name.
• 4:08
  SPEAKER_06: Including at least one of them that actually has AI in its title.
• 4:13
  SPEAKER_06: So that's an interesting one.
• 4:14
  SPEAKER_06: So aside from the fact that lots of Slovenian websites are selling for a lot of money so far, I'd be curious as to what your takeaways are in terms of what's happening right now and is it meeting the moment of where we're at?
• 4:34
  SPEAKER_06: Let's start with you, Matt.
• 4:35
  SPEAKER_01: It's so loaded.
• 4:36
  SPEAKER_01: I am still a little in disbelief at the group coming together in this way.
• 4:43
  SPEAKER_01: And it is clear that something has to be discussed.
• 4:46
  SPEAKER_01: Talking about morals is a good one to discuss.
• 4:49
  SPEAKER_01: Ethics would also be good.
• 4:51
  SPEAKER_01: But the proof will be in what happens next.
• 4:54
  SPEAKER_01: And that is really where the seeds of doubt are planted by the history that got us here to today.
• 5:07
  SPEAKER_01: Because I seem to remember that Google's main theme was don't be evil, right?
• 5:14
  SPEAKER_01: This was their main value.
• 5:15
  SPEAKER_01: We didn't need to rehash this.
• 5:17
  SPEAKER_01: It was baked it was baked in from the beginning, it used to be, right?
• 5:21
  SPEAKER_01: So uh interesting to have to get together and sign like this to be moral about AI on the heels of a meeting with other another superpower.
• 5:34
  SPEAKER_01: So I interesting.
• 5:39
  SPEAKER_01: That's about as deep as I can really stand to think about.
• 5:43
  SPEAKER_01: Just to think about it.
• 5:44
  SPEAKER_06: Yeah, I mean it's crazy.
• 5:46
  SPEAKER_06: I don't blame you.
• 5:46
  SPEAKER_06: I think the uh it's interesting to pull it together in terms of the big geopolitical meeting happens between the US and China, and then within days this happens.
• 5:58
  SPEAKER_06: It's interesting it was that order as well, by the way.
• 6:01
  SPEAKER_06: And so I think that is I do think that's intriguing.
• 6:05
  SPEAKER_06: I don't think anybody really knows what to make about it.
• 6:07
  SPEAKER_06: There was a piece of paper signed, as you pointed out, Matt, where they misspelled the United States of America on it, but there was a but there was a piece of paper signed.
• 6:14
  SPEAKER_06: Legend.
• 6:15
  SPEAKER_06: A piece of paper.
• 6:18
  SPEAKER_06: Yeah, right?
• 6:19
  SPEAKER_06: Like Yeah, and there was a and interestingly, there was a piece of paper signed by some of the parties, but not all of the parties, which is intriguing in its own right.
• 6:26
  SPEAKER_06: But what did you what did you make of it?
• 6:30
  SPEAKER_05: Well, similar to Matt, I'm I'm sort of taking it all in because at times I felt like the scenes that I was watching were almost, to your point, Matt, like almost misinformation or disinformation.
• 6:42
  SPEAKER_05: So I was having a hard time sort of snapping back into reality.
• 6:46
  SPEAKER_05: And and what I tend to go to, you both know this about me, is I go back to the like language that we're using.
• 6:52
  SPEAKER_05: And so I find it really interesting that this was called an accord.
• 6:57
  SPEAKER_06: Yeah.
• 6:58
  SPEAKER_05: And when I think about the order of events, like the geopolitical events and that visit that took place right before this, the word accord typically is used more in kind of diplomatic avenue, you know, because it has to do with like getting into harmony or like a shared set of principles or a shared set of values.
• 7:19
  SPEAKER_05: And I just have to laugh inside a little bit because I did not see harmony in that room or those pictures.
• 7:26
  SPEAKER_05: I certainly did not see a shared set of terms or principles.
• 7:30
  SPEAKER_05: Um and as you said, can an accord be formalized if it has a spelling error on it?
• 7:36
  SPEAKER_05: And it is notable that both Microsoft and I believe Amazon did not sign the document.
• 7:42
  SPEAKER_05: Now we could speculate as to why that is.
• 7:44
  SPEAKER_05: I don't think people have even had enough time to why, or if that's a dissent.
• 7:50
  SPEAKER_05: Or perhaps people are saying, well, you know, in Microsoft's case, I think they actually signed a document in 2023 under a different administration that they probably might feel is still the same thing that they would abide by.
• 8:06
  SPEAKER_05: So I don't know.
• 8:07
  SPEAKER_05: Um, it is going to be really interesting to see the second act of whatever's to follow this very public showcase.
• 8:16
  SPEAKER_05: I would call it a showcase.
• 8:18
  SPEAKER_06: Well, the very public bit was one of the bits that struck me.
• 8:22
  SPEAKER_06: You know, we've spoken about movies in the last few weeks, and bits of this felt like I was in a movie about AI.
• 8:31
  SPEAKER_06: You imagine the scene.
• 8:32
  SPEAKER_06: We've gathered together the leaders of all of the top AI companies, and they've come together in order to be able to meet this moment, you know, all of that kind of stuff.
• 8:41
  SPEAKER_06: And obviously, many of those leaders would, you know, want to be played by Ryan Gosling or something versus the actual humans that they are.
• 8:48
  SPEAKER_06: But at the same time, you know, it there was a sort of almost like a sort of show element to this.
• 8:56
  SPEAKER_06: This is this is how we're meeting this moment by using the power of the presidency to gather together all of these individuals and basically sort of throw it back onto them.
• 9:11
  SPEAKER_06: The thing that it sort of reiterated to me was actually some of the themes that we were kind of talking about last week, which is like almost every time something like this is happening, there's what's happening, and then there's like a backstory that you don't need to be a conspiracy theorist to look at it and go, there's a backstory here and we don't know what it is, right?
• 9:33
  SPEAKER_06: And that that's the really intriguing thing about this.
• 9:36
  SPEAKER_06: And then I suppose this would always be the case, but each of these leaders is coming into this from a very different position.
• 9:45
  SPEAKER_06: Some of them are coming into it from, I would say, a position of greater strength.
• 9:50
  SPEAKER_06: Other ones are coming into it from a position of weakness.
• 9:52
  SPEAKER_06: There was an article literally this morning on the BBC website that was titled, Is Meta dying?
• 10:00
  SPEAKER_06: Right?
• 10:00
  SPEAKER_06: And some really, really interesting concerns about what have been considered to be repeated strategic missteps from a company that even holds the key to one of its greatest strategic missteps in its name, namely Meta, right?
• 10:17
  SPEAKER_06: Remember the rebranding of Meta was because we were going all in on the metaverse.
• 10:22
  SPEAKER_04: Yeah.
• 10:22
  SPEAKER_06: And, you know, how did that go?
• 10:25
  SPEAKER_06: And so there's that.
• 10:27
  SPEAKER_06: There's also sort of like wildly fluctuating trust levels associated with each of these companies.
• 10:34
  SPEAKER_06: I've been doing some informal surveys for an article that I hope will be released in two or three weeks, which is looking at precisely this, right?
• 10:44
  SPEAKER_06: How do you go about making strategic bets in this world where without calling out any of these companies in particular, but the level of trust associated with these companies, many of them it's very low.
• 11:01
  SPEAKER_06: Other ones, it is higher, but falling like a stone.
• 11:06
  SPEAKER_06: Other ones, there is for whatever reason a level of slightly higher trust right now, but that could change.
• 11:13
  SPEAKER_06: I mean, think about the two perceived biggest players on the Frontier model side, although Google just released its own Frontier model yesterday that looks to be as or more powerful than either of them.
• 11:28
  SPEAKER_06: But if you think about the ones that have been in the news, OpenAI and Anthropic, there was a time maybe six months ago, maybe four months ago, or something like that, where almost universally everybody was thinking of like, oh, now OpenAI are the bad guys and anthropic are the good guys, and that's subtly shifted over the course of the last few months.
• 11:49
  SPEAKER_06: So their stars are like flying and fading, based in some cases on how much people think they're worth, but then also what the overall level of trust is associated with each of each of these organizations.
• 12:03
  SPEAKER_06: So they're all coming in, right, with different things that they need to get out of it in a very, very complicated way.
• 12:11
  SPEAKER_06: And so the real question, I suppose, and I'll throw this to either of you that wants to pick it up, is in a world where all of this is changing so rapidly, and every one of them has their own set of interests and their own sort of standing, and that set of interests and that standing is here today, and it could be somewhere entirely different tomorrow if somebody leapfrogs them or you know some other unusual thing happens.
• 12:43
  SPEAKER_06: What kind of meaningful accord can you actually come to when it's all that confusing?
• 12:50
  SPEAKER_06: Because meaningful accords come from shared interest.
• 12:53
  SPEAKER_06: How do you define shared interest in this world?
• 12:57
  SPEAKER_05: I think it's really being tested.
• 12:59
  SPEAKER_05: And again, in the absence of anything else, I think we can only look to behavior and choice.
• 13:06
  SPEAKER_05: Again, I'm gonna caveat everything I say here.
• 13:09
  SPEAKER_05: I don't know if this is true or if this is misinformation.
• 13:11
  SPEAKER_05: Paul, you could probably check me because I think this is on X or Twitter, but I saw a screenshot, so again, everything is to be verified.
• 13:19
  SPEAKER_05: OpenAI announced dots, right?
• 13:22
  SPEAKER_05: Which is sort of this agentic thing.
• 13:24
  SPEAKER_05: SpaceX AI buys dot dot com, it redirects straight to the Grokbot download page.
• 13:35
  SPEAKER_05: So again, to your point, if this is what's happening behind the curtain, then it proves to you the sense of like there's not harmony and there are counter behaviors of harmony and shared belief if people are capitalizing on such um I don't want to call it nefarious, but strategic and duplicitous ways.
• 13:59
  SPEAKER_06: Yeah, I did check it does indeed redirect to the Grop Bop page.
• 14:03
  SPEAKER_06: So there you go.
• 14:04
  SPEAKER_06: Yeah.
• 14:05
  SPEAKER_06: Yeah.
• 14:06
  SPEAKER_06: So that's fascinating.
• 14:08
  SPEAKER_01: This is how crazy the news cycle is, Paul, that I totally forgot.
• 14:12
  SPEAKER_01: I don't know how long ago it was, it was months.
• 14:16
  SPEAKER_01: When didn't Anthropic get banned from all government?
• 14:21
  SPEAKER_01: They were just like, oh, they're out, completely broke the contract, all that stuff, right?
• 14:25
  SPEAKER_01: So how do you go from that to him propping up the CEO to speak at this event yesterday?
• 14:32
  SPEAKER_01: It's really, I mean, you know, you're allowed to change your mind with new information.
• 14:37
  SPEAKER_01: I'm not faulting them for that, but it's just like this is really weird and it's happening right in front of our face.
• 14:44
  SPEAKER_01: And how can we expect anybody to be able to decipher what to do and what's actually going on in this space if you're paying attention and you're relatively critical thinker and you're confused, what hope does anybody else have?
• 15:10
  SPEAKER_01: And that's why you look at this group of people and you're like, I think I've been misled.
• 15:15
  SPEAKER_06: Well, if I'm speculating which we've been two and two weeks of now, but I kind of get where we are, right?
• 15:22
  SPEAKER_06: As uncomfortable as it is, I'm starting to think that this is much more about the short term.
• 15:28
  SPEAKER_06: We kind of talked about this with one of the scenarios last week, but I think this is much more about the short term than it is about what's happening in ten years.
• 15:36
  SPEAKER_06: And what I mean by that is that if these companies are planning on going public relatively soon, people might have seen that anthropic to the initial filing for the IPO.
• 15:46
  SPEAKER_06: If we are looking at something in the very short term and then you start to think about short-term risk, then I think it's fairly clear what the short-term risks at this point look like.
• 16:04
  SPEAKER_06: And they look like some difficult to control cyber incident, and worst case scenario, a cyber incident that impacts human infrastructure that humans rely on.
• 16:21
  SPEAKER_06: And so we're at this very, very interesting juncture where, if you look at the valuation of non-AI companies, and look at their valuation over the last couple of months.
• 16:44
  SPEAKER_06: Just over the last couple of months, the valuation of all the companies that are not AI companies has dropped about 5% just in the last couple of months.
• 16:52
  SPEAKER_06: And stock markets continue to go up because of the valuation of the AI companies.
• 16:59
  SPEAKER_06: So literally, you've got stock market going up overall, the AI company valuation, just those that are gone public at this point, right?
• 17:10
  SPEAKER_06: But the valuation of those companies is what is continuing the stock market trajectory that we're on.
• 17:17
  SPEAKER_06: So if you're super short-term, and you know, we're not a political show, but I think most people can agree, whatever their political background, that this is an administration that tends to think short-term, right?
• 17:34
  SPEAKER_06: If you're super short-term, you're like, let them rain, because without them, the whole stock market's going down, right?
• 17:42
  unknown: Right.
• 17:42
  SPEAKER_06: But if you're even ever so slightly, or if you're looking at it from another perspective, from a perspective of like going for an IPO, and you're genuinely worried that like some form of, you know, you've seen the cyber incidents that we've heard about, and then you've seen the other ones, right?
• 17:58
  SPEAKER_06: And you're worried about some very significant cyber incident that impacts human infrastructure, and then you start to project forward and you think, okay, well, how much am I, you know, if it if it takes down the the grid of the Western United States or the Eastern United States, or something along those lines, but some incident that has some massive, massive effect on our infrastructure, how much are those companies really worth at that point?
• 18:24
  SPEAKER_06: Right?
• 18:25
  SPEAKER_06: And so again, speculation, but I think well actually if you even if you pull speculation out of it, I guess if we come back to the original question that I was posing, which is what could they agree on?
• 18:41
  SPEAKER_06: Maybe they could agree on the idea of we need to make sure, as Jensen Wong said the other day, we need to make sure that from an engineering perspective these things are as safe as they can be.
• 18:58
  SPEAKER_06: And you don't have to love humanity to think that.
• 19:02
  SPEAKER_06: You can love money and think that, right?
• 19:06
  SPEAKER_06: Because if you want to be valuable, then sort of pulling down massive chunks of human-facing infrastructure, it's not unlikely to be associated with a very high valuation.
• 19:21
  SPEAKER_06: Unless suddenly we say, okay, well they're they're weapons now, and weapons are worth money.
• 19:25
  SPEAKER_06: But I don't know.
• 19:25
  SPEAKER_06: But anyway, I don't know.
• 19:26
  SPEAKER_06: What do you guys think about that?
• 19:28
  SPEAKER_01: Yeah, that's interesting.
• 19:29
  SPEAKER_01: What can they agree on?
• 19:31
  SPEAKER_01: And do we have any parallel that would have looked quite like this?
• 19:35
  SPEAKER_01: Like if you think, you know, nuclear power becoming a thing, I don't know that they would have been able to bring out all the scientists like this and talk about it this way.
• 19:49
  SPEAKER_01: This one's uh it's hard to think of a parallel, it's very unique.
• 19:53
  SPEAKER_06: Well, I return, and we're gonna write about this, but I return to the idea that what is unbelievably unique about this is massive power and starting privately owned.
• 20:06
  SPEAKER_06: Do the thought experiment.
• 20:07
  SPEAKER_06: Think about the idea of this being discovered by the United States military.
• 20:13
  SPEAKER_06: If you start there, then maybe in ten years' time we would now be thinking about okay, well, we fully understand the power, we're putting the safeguards into place, we've done enough research on this to be able to fully understand it, and now we'll invite you know private contractors to come into this world which we have created.
• 20:38
  SPEAKER_06: That's the normal trajectory for super powerful stuff that can be weaponized, and we're not on that path.
• 20:46
  SPEAKER_01: And now the incentive is flipped.
• 20:48
  SPEAKER_05: Yeah.
• 20:49
  SPEAKER_05: I think the other part which is interesting to your discussion on this, Paul, is I don't think we can discount this sort of undercurrent of, and I put in quotes, celebrity.
• 21:00
  SPEAKER_05: And what I mean by that is there's celebrity and then there's this weird sort of like transparency piece.
• 21:07
  SPEAKER_05: Like I've been following the work of the US government for a long time.
• 21:12
  SPEAKER_05: I don't think I've ever seen a seating chart published the way they published a seating chart for this.
• 21:20
  SPEAKER_05: For this.
• 21:20
  SPEAKER_05: No, for state dinner, yeah, but not for state dinner, but not for this.
• 21:24
  SPEAKER_05: Yeah, yeah.
• 21:26
  SPEAKER_05: And it made me like pause for a second and say, okay, this is being done because these dare I say, titans of industry are honestly supposed to be people that we don't know by name.
• 21:39
  SPEAKER_05: Because to your point, Matt, like what is the thing that incentivized them?
• 21:44
  SPEAKER_05: I would have thought, you know, in the days of Steve Jobs and others, right?
• 21:48
  SPEAKER_05: There's a larger, grander thing that incentivizes them because they're they're leaving a mark, hopefully a good mark, on society and culture.
• 21:56
  SPEAKER_05: But we're almost supposed to not know who they are.
• 21:58
  SPEAKER_05: And now we you know we know.
• 22:00
  SPEAKER_05: them all by name.
• 22:01
  SPEAKER_05: We know them by their look.
• 22:02
  SPEAKER_05: We know them by their dress.
• 22:03
  SPEAKER_05: We know them by their seating chart.
• 22:05
  SPEAKER_05: And by the SNL pattern.
• 22:09
  SPEAKER_05: Yeah.
• 22:10
  SPEAKER_03: Yeah.
• 22:10
  SPEAKER_05: Right?
• 22:11
  SPEAKER_05: Like archetypes.
• 22:12
  SPEAKER_05: It's and I just worry like our priorities are being a little lopsided and dictated to us in that way.
• 22:21
  SPEAKER_05: I just scratch my head at that.
• 22:23
  SPEAKER_06: Yeah.
• 22:24
  SPEAKER_06: Wow.
• 22:25
  SPEAKER_06: Alright.
• 22:26
  SPEAKER_06: Wanted to cover it before we got into sort of like that's the old d'eau before we got into the main course.
• 22:32
  SPEAKER_06: I know you guys have been working on a really interesting article that we're going to be publishing on humanity working and it'll be out by the time this episode comes out.
• 22:42
  SPEAKER_06: And I mentioned it in the intro How do you get Turkeys to vote for Christmas?
• 22:49
  SPEAKER_06: The title of what you've been working on is a really interesting title How do you incentivize teams to train the robots and architect their own extinction?
• 22:58
  SPEAKER_06: And I know in the article largely we're talking about robots metaphorically but let's for a second talk about them actually.
• 23:04
  SPEAKER_06: I've got to tell you guys I think this is one of the missing pieces of the story.
• 23:08
  SPEAKER_06: I was like back in 2020 I recorded a much younger looking version of me that didn't appear to have any grey hairs in his beard at all.
• 23:16
  SPEAKER_06: I was looking at the video the other day was banging on about what I saw as a convergence between robotics, AI and automation.
• 23:25
  SPEAKER_06: And of course we've seen the convergence in a very clear way between what AI was then and automation with the emergence of agentic frameworks.
• 23:37
  SPEAKER_06: And that will only continue.
• 23:40
  SPEAKER_06: The thing that we've been not talking about as much and we only seem to talk about it about once a year when they have the robot games and we take great pleasure in laughing at them.
• 23:51
  SPEAKER_06: But we would be foolish to continue to laugh at them.
• 23:54
  SPEAKER_06: I mean Matt, you and I I know have discussed this on a recent previous episode I mean if you look at the most recent robot games while we were laughing at it there was one that ran faster than the same bolt right and while we were laughing at it there were ones that were like moving with incredible dexterity.
• 24:12
  SPEAKER_06: And just the other day there was one that was seen online that was very with a highly great precision, you know, adding nuts to bolts and you know and doing various things of that nature.
• 24:25
  SPEAKER_06: We're not that far away now from general purpose robotics that can do stuff, physical stuff that humans can do.
• 24:32
  SPEAKER_06: And I worry about the lack of talk about that convergence for two reasons.
• 24:38
  SPEAKER_06: So the first is that when we look at some of the dangerous outcomes like a lot of the argument against dangerous outcomes is well you need a physical person to physically do stuff, you know, in order for X or Y or Z to happen.
• 24:53
  SPEAKER_06: I mean if it's a physical person physically doing things there are physical entities namely robots that can physically do things.
• 25:01
  SPEAKER_06: And then of course the second part is that once you start to bring in physical activities, maybe you don't have the same level of white collar job loss that we project, who knows but you can potentially have a lot of manual labor become less attractive than the robot that can work twenty-four hours a day.
• 25:22
  SPEAKER_06: I've spoken to economists on this they see a point of convergence where robots become you know per hour thirty dollars an hour.
• 25:37
  SPEAKER_06: Something along those lines that's not far away from you know the fully loaded cost of a human.
• 25:44
  SPEAKER_06: So before we get into the sort of like metaphorical usage of robots I'll start with you Matt.
• 25:50
  SPEAKER_06: I mean you work in a field where robotics is used in quite a few different areas alongside humans but often independently of humans.
• 26:00
  SPEAKER_06: You see the value of it in a highly specialized way but do you see situations where more generalized robots are going to be more valuable in your industry and beyond?
• 26:12
  SPEAKER_01: Absolutely I mean I think we've already seen it they just have a different shape right so like full self-driving cars, washing machines dishwashers we are forever inventing things to save us time in quotes and yet we have not saved ourselves any time it'll be interesting to see you know will we finally deliver on that one?
• 26:35
  SPEAKER_01: Like we'll have nothing but time because there's no other things we can do right so um think of all the things you don't feel like doing or are dangerous.
• 26:46
  SPEAKER_01: Yeah if you need a human shape to do them why wouldn't you rent one to come and you know do it for for thirty bucks or whatever.
• 26:57
  SPEAKER_01: Like I think we're inherently looking for shortcuts and it's when the price hits a point where people can tolerate outsourcing it if it's done right you're not going to worry about it.
• 27:10
  SPEAKER_01: Full self-driving cars is another one like that went from a pipe dream to like now we have fleets of them driving around cities.
• 27:16
  SPEAKER_01: Like it's that didn't take long.
• 27:18
  SPEAKER_01: And so I extrapolate that that is going to be more and more a part of everything that we do.
• 27:25
  SPEAKER_01: Certain people will tolerate having a bot in their house to do stuff it's humanoid.
• 27:32
  SPEAKER_01: Or even you know we talk about that movie her and stuff like that.
• 27:36
  SPEAKER_01: And what's the other one where the robot is actually living in the house I think it wasn't her.
• 27:40
  SPEAKER_01: It was the other one.
• 27:41
  SPEAKER_01: But I don't see why it doesn't happen.
• 27:45
  SPEAKER_01: That's kind of where I'm at and again morals and ethics are going to be really important.
• 27:52
  SPEAKER_01: In the programming at the root it's got to be there from the get go and and I think precedent science fiction and experience and black mirror tell me that it's not going to be the way it turns out it's pretty daunting but we still do it we still do it.
• 28:09
  SPEAKER_01: In every timeline we still do it.
• 28:12
  SPEAKER_06: You know while you were chatting there Matt I looked at major AI companies that have invested heavily in robotics and we obviously know that Tesla has slash SpaceX Microsoft has Amazon has Nvidia has OpenAI has so you know they're not idiots.
• 28:43
  SPEAKER_06: They're seeing a convergence that is that is going on here and in a sense it's the missing piece isn't it Bill right?
• 28:50
  SPEAKER_06: Like if you can do everything in your head the obvious logical thing is what else do you want to do that a human can do?
• 28:57
  SPEAKER_06: Right?
• 28:58
  SPEAKER_06: These things can these things can see, they can speak weirdly they can smell right so the thing that we've not connected them to yet really very effectively is the ability to do the physical stuff that humans can do.
• 29:15
  SPEAKER_05: Yeah you know I I think it's it's interesting when you talk about the sensing of these things and the robotics and sensing um I've been reading a little bit about inference.
• 29:27
  SPEAKER_05: So kind of like this way in which you know presently maybe a robot can go in and identify like a piece of furniture and a desk or an item right they kind of itemize and say okay this is the inventory therefore with and I put in quotes reasoning but that's what they say that does this is where I am and I can sort of manipulate my way through the space you know I think it's going to be really interesting autonomous vehicles are a fascinating example of this also autonomous trucks where it gets to the level of it can take in the entire environment and then know how to manipulate through senses its way through the environment.
• 30:09
  SPEAKER_05: So it's no longer just about items it's about rooms it's about places it's about roads about highways and our understanding I think as humans is very much tied to how we move through the world.
• 30:20
  SPEAKER_05: And as we move through the world what do we sense what do we touch what do we feel but then also what do we preserve?
• 30:26
  SPEAKER_05: Like what is kind of uniquely divinely ours and that's where things like taste and craft come in.
• 30:34
  SPEAKER_05: And so when you ask the question Paul about like what's left that's the part that really in a good way keeps me up at night which is like okay how do you talk to an Italian craftsman who makes the most comfortable shoe that fits your foot that can only be made in Florence or Venice or whatever because it's hand stitched by that craftsman.
• 30:57
  SPEAKER_05: Now can a robot do it possibly I think we've seen examples in the past of even companies like Nike and others who have tried to take that and automate it, but you can't quite get the nuance and the dexterity and in a way I wonder if some of these quote unquote limitations of ours are in fact our stamp on the thing that cannot be replicated or cannot be repeated.
• 31:22
  SPEAKER_05: And I I don't have answers to that.
• 31:24
  SPEAKER_05: My worry is that we're going to be pushed to such an edge that people are going to say enough is enough and I seek or I crave something that has not been metallic or silica but comes from hide or something or yarn, whatever you know the the things that we at one point called you know what Luddites would use or uncivilized.
• 31:49
  SPEAKER_05: That's just something I I don't have answers for but I think a lot about as all the speed at which the sensing and inferencing goes right in front of our eyes.
• 31:57
  SPEAKER_06: Well and to your point there's a bigger thing that is going on one of the things that I've been looking at recently is public sentiment about AI and what is happening to public sentiment about AI.
• 32:09
  SPEAKER_06: And you wouldn't be surprised I mean it's it's not the most popular thing right now and it turns out that you know if the things you lead with are it's gonna get rid of everybody's job and it's gonna and it might kill us all if you lead with that then eventually some people are going to form a negative opinion of the thing that you're talking about.
• 32:38
  SPEAKER_06: But this kind of gets into the theme the themes of some of the stuff that you that you're talking about in the article which is that we sort of two things going on at the same time right we have this zeitgeist that is happening around AI and how that's playing out inside organizations and how that's causing levels of passive and in sometimes active in some cases active resistance by employees and then you also have companies who want to for want of a better term get shit done right and for whom they still see AI as a potentially highly useful tool to allow that to happen and they're employing people AI enablement officers very difficult job right now to try to actually help ensure that the value of the AI investments that they're making are being realized.
• 33:42
  SPEAKER_06: A lot of those AI enablement offices are realizing that the people who are still needed largely inside these companies have potentially a bit more power here than they thought they did and the people are not willingly always jumping in and going you know what I am going to help AI replace me or replace my team or whatever it happened to be.
• 34:14
  SPEAKER_05: Now I I know you're close to this stuff Bill so how do you resolve that tension well I think the first step is not ignoring the tension.
• 34:25
  SPEAKER_05: I think you have to you sort of have to let it breathe is what I would say.
• 34:29
  SPEAKER_05: And there are cases and we write about this in the forthcoming article but there are cases where it it makes sense.
• 34:35
  SPEAKER_05: I mean there's a lot of what we call friction or administribia or as Matt was saying, you know, if you have to sort of follow a checklist or a standard operating procedure to get work done in order to meet an expected outcome and it's a repeatable process that's kind of a perfect candidate for maybe you know automating or at least training some agent to follow those steps.
• 34:59
  SPEAKER_05: But what I also think it causes people to do is question then where they discern their value and where their worth lies in the enterprise.
• 35:09
  SPEAKER_05: I mean all of us are knowledge workers here.
• 35:11
  SPEAKER_05: A lot of people probably listening are knowledge workers and I think you're going to have to negotiate with yourself what is worthy and where do I want to in fact train either an agent or a robot or something in the middle to take on tasks that I used to do manually but that's not where it ends.
• 35:30
  SPEAKER_05: The next part has always been the million dollar question for me is and then where do I spend the time that I now have available to me?
• 35:37
  SPEAKER_05: If you offered groups of people more imagination, more creativity, more innovation, I think they would take it that they would take you up on that.
• 35:48
  SPEAKER_05: But the question becomes are you willing to let some of that livelihood that you have ensnared in the way you do work, are you willing to let some of that go?
• 35:57
  SPEAKER_05: And I don't know if we know the patterns that will play out and that's why I think you have to let the tension breathe but there are opportune times Paul where it makes sense you know if we're doing a knowledge transition plan or if we're doing an onboarding or things that really are repeatable and kind of a pain in our ass like why not give it over and then find a way to spend more time in relationship with other humans.
• 36:22
  SPEAKER_05: I'm biased towards the human in this in this equation but most people that know me will know that.
• 36:27
  SPEAKER_06: There's a tension here though and I'd love to get your thoughts on this Matt there's a tension between what potentially we're trying to accomplish and the technology we're using to accomplish it.
• 36:37
  SPEAKER_06: Right so the thing about an SOP standard operating procedure is that it's designed to bring in repeatability and reliability and in theory the closer you follow the standard operating procedure the more consistent the results are that you get and so there's an open question I say this as somebody who has been building like a I've spent the last two months trying to build a like a personal chief of staff and I'm using the latest frontier models to do it and I'm building out an architecture that relies on SOPs.
• 37:14
  SPEAKER_06: When something is is rigid and followed rigidly by technology it can be incredibly reliable when the front end into that is a probabilistic AI model that may or may not check fully check its own config files that may or may not follow a procedure that you've put right in front of it that if it doesn't might claim that it does and that won't automatically report back what it's done it's in a sense it's almost like the wrong tool for this, right?
• 37:56
  SPEAKER_06: And we come back to sort of like it's sort of weird that we have humans when we talk about human in the loop what are we really talking about?
• 38:06
  SPEAKER_06: We're actually talking about humans, unreliable humans auditing what previously we would have thought of as reliable technology.
• 38:16
  SPEAKER_06: I mean that's a bit weird isn't it I mean normally like normally your spreadsheet sits over here and that's the you know you're a single source of truth or whatever and then you've got and you automate everything that you can to try to reduce the points of error and now we're in this rather strange world where we have AI do some stuff and then we have humans in it to make sure that it's doing stuff consistently as it's supposed to do it.
• 38:43
  SPEAKER_06: So I get the point about SOPs and I get the point about how we want to have it do all of the stuff that we're supposed to be doing but it needs to do it reliably and consistently and it needs to audit that it's done it and you know and things like that.
• 38:57
  SPEAKER_06: There was one use case that I had the other day with this sort of personal chief of staff that I'm building and I was trying to figure out how reliable how reliably it could tell me that like it's my friend's birthday right and its hit rate was like 85% 15% of the time it's forgetting what today is oh I'm supposed to check birthdays oh I'm supposed to remind Paul of the it's in its SOP right and it's doing it most of the time but it's only doing it most of the time.
• 39:27
  SPEAKER_06: So is this like technology yes but this type of technology is it oh it's in front of us so we use it for this sort of stuff?
• 39:37
  SPEAKER_06: I don't know what are you what are you seeing?
• 39:40
  SPEAKER_01: That's such a good way to characterize it too.
• 39:42
  SPEAKER_01: And I just thought as you were saying and Bill we might have to we might have to look at the article to get around this new thought that I just had um because of what you just said Paul because you can't really rely on the current incarnation of this to even follow an SOP as good as a human unfortunately because it's pattern matching right like maybe an agent workflow maybe like automation is still good but like the AI bit where it can make choices not as reliable.
• 40:13
  SPEAKER_01: What if this is just a backdoor way for us to finally get knowledge workers and people that do these things to write a good process like literally write it down.
• 40:25
  SPEAKER_01: Like and then we're like yeah you're not going to get any new tools to do it we just really needed you to find out finally write it down and because you do find the problems in your process in trying to make it a good candidate you're like how do you how does this actually happen?
• 40:40
  SPEAKER_01: When do you make this decision?
• 40:41
  SPEAKER_01: Where do you make that decision?
• 40:42
  SPEAKER_01: Who has to be involved?
• 40:43
  SPEAKER_01: Oh see if you draw it out like in a Visio like lean six sigma days you start to see what it actually is and where you should plug in automation robots AI but you have to have that first and it's almost like this impending threat of start using AI to do everything coming from above is going to force us to figure out those processes and then we may decide we can't give this to the AI we tested it and it doesn't consistently work like we needed it to but we now have a really clean process because we were writing the process for robots to do instead of people.
• 41:24
  SPEAKER_06: We are kind of at a point I think maybe the high point of situations where we're like oh it's a thing so I'll use AI for it versus it's a thing what should I use it's a challenge what should I use to address the challenge and what role does AI play in that process?
• 41:47
  SPEAKER_06: And Anala Bill, you and I both worked at Microsoft do you remember the time where it was that there was a phrase that went round Microsoft which was the answer is SharePoint what's the question?
• 41:56
  SPEAKER_06: Where it's sort of right now the answer is AI what is the question that's amazing.
• 42:05
  SPEAKER_01: So true though.
• 42:06
  SPEAKER_06: So true okay so yeah really really interesting part about the the a nugget of truth I think that's really important there which is that we do want to not use humans for lower value less interesting things if technology can replace that in an ideal world so that we get the magic out of what humans really do.
• 42:39
  SPEAKER_06: And I know you guys touch upon this idea that that a lot of the magic of what humans do is in the intangibles of what humans do.
• 42:52
  SPEAKER_05: So talk about that a little bit yeah I think for me probably where this came to fruition recently is something that I had a discussion with you know a supervisor recently and this person basically said to me in a way I know there's a lot of things you do that a lot of people do not see and so my challenge to you is to make the invisible visible.
• 43:16
  SPEAKER_05: And I thought that was a really interesting challenge.
• 43:19
  SPEAKER_05: And I turned to AI to help me do this right because to your point around the intangible or the invisible um you know the way we work it's all ones and zeros and it's signals and it's connections and it's telemetry all day.
• 43:34
  SPEAKER_05: You know I'll cite Microsoft because we were both there Paul in the past to your point the answer is SharePoint what is the question it's not too dissimilar than everything exists in the Microsoft 365 ecosystem.
• 43:47
  SPEAKER_05: So when somebody said to me take the intangible make it tangible or take the invisible and make it visible I turned it I went to Copilot and I said hey copilot here's your challenge go back a year and a half and look at Every one of these sets of data, right?
• 44:02
  SPEAKER_05: Every one of these tools and channels, you've got it all, right?
• 44:05
  SPEAKER_05: You're tracking everything.
• 44:07
  SPEAKER_05: It's your system, it's your world, it's your happy place.
• 44:12
  SPEAKER_05: Make the invisible visible.
• 44:14
  SPEAKER_05: And what I got back was a 57-page document.
• 44:18
  SPEAKER_05: Now I told it to work overnight.
• 44:19
  SPEAKER_05: I was very specific.
• 44:21
  SPEAKER_05: I said, I'm going to go to bed, you work all night, and by tomorrow morning I want the full report.
• 44:26
  SPEAKER_05: And I got a 57-page document that illuminated to me more things that I had forgotten that I had done, that I had worked on, that I had supported.
• 44:35
  SPEAKER_05: And was it 100% perfect?
• 44:37
  SPEAKER_05: No.
• 44:38
  SPEAKER_05: But did it illuminate to me some really valuable things?
• 44:40
  SPEAKER_05: For example, you know, groups of people that turned to me for confidence and quick answers that don't even exist in my everyday social ecosystem.
• 44:52
  SPEAKER_05: So naturally, how would a supervisor or a coworker understand that person's relation to me?
• 44:57
  SPEAKER_05: And all of a sudden it became this beautiful like ontology mixed with a topology, and the invisible became visible.
• 45:06
  SPEAKER_05: To me, that was a great use case of this technology, and it informed how I would approach this differently.
• 45:12
  SPEAKER_05: But that's a very specific use case.
• 45:14
  SPEAKER_05: It wouldn't necessarily work for every single person.
• 45:16
  SPEAKER_05: And now I'm trying to be much more conscious about how I work so that the scales are not tipped in the invisible and away from the visible.
• 45:28
  SPEAKER_06: I think it's incredibly powerful, and I think that it does remind us that you know, just spoke about laughing at the robots, you know, falling over and things.
• 45:38
  SPEAKER_06: Because of the moment that we're in, there's a lot of, you know, humans reassuring themselves about AI being, you know, crap at this and crap at that.
• 45:49
  SPEAKER_06: The flip side of like understanding that AI isn't the tool for everything, is understanding the things that AI can uniquely do in some cases, or can certainly do significantly better than humans.
• 46:03
  SPEAKER_06: And I think you really you really hit on one.
• 46:06
  SPEAKER_06: It's intriguing, really, isn't it?
• 46:07
  SPEAKER_06: I had a conversation with Bobby Johansson from the Institute for the Future a few years back, actually, on one of our earlier podcasts, Matt.
• 46:16
  SPEAKER_06: And I was just I followed your advice, Bill.
• 46:18
  SPEAKER_06: I went back and listened to some of the other ones.
• 46:20
  SPEAKER_06: Alright.
• 46:20
  SPEAKER_06: And it was really interesting because it was at that point in time where if you remember, AI was a bit rubbish at maths.
• 46:29
  SPEAKER_06: And but weirdly, AI was better.
• 46:34
  SPEAKER_06: I'm not gonna say it was good, certainly not as good as a good human, but it was better at say writing a sonnet than it was solving a math problem.
• 46:43
  SPEAKER_06: Yeah.
• 46:44
  SPEAKER_06: And by the way, already back then it was better than writing a sonnet than I am.
• 46:48
  SPEAKER_06: Not better than you, Bill.
• 46:49
  SPEAKER_06: You are you are actually a poet, but the but it was certainly better than I than I am at writing a sonnet.
• 46:54
  SPEAKER_06: And that was weird.
• 46:55
  SPEAKER_06: It wasn't how we thought about AI.
• 46:59
  SPEAKER_06: It wasn't you know what we thought AI should be good at.
• 47:02
  SPEAKER_06: And I rather love the fact that it wasn't good at math because I think that, you know, as a mathematician, I think that the art of math is something that is not regarded highly enough by people, and it kind of indicated that there is sort of an art to it.
• 47:18
  SPEAKER_06: But I think that we should always be asking ourselves that question.
• 47:23
  SPEAKER_06: What is what is AI really good at?
• 47:26
  SPEAKER_06: And in some cases, the reason that AI is good at certain things or better at certain things than humans are tells us something about ourselves as humans as much as it does about the AI.
• 47:37
  SPEAKER_06: So for example, you know, if you're a human, if you're a knowledge worker and particularly a super busy knowledge worker, you kind of flush the cash every time you've done every time you've done something, and you don't, and as we all know, right, human memory is deeply, deeply fallible.
• 47:54
  SPEAKER_06: And we have all of these biases that are inherent within us, recency bias, various other things as well, that cause us to completely, completely forget what we do.
• 48:03
  SPEAKER_06: And you know, like many of us, I have a meeting with my manager every week or two, and part of that meeting is inevitably, so what have you done?
• 48:13
  SPEAKER_06: Look a deer in the headlights when I have those conversations without AI.
• 48:17
  SPEAKER_06: I was like, I don't know what I did yesterday.
• 48:18
  SPEAKER_06: I'm not sure I know what I did this morning.
• 48:20
  SPEAKER_06: I knew I was doing stuff, I knew I was like being productive in some way, shape, or form, but my job is very ambiguous.
• 48:27
  SPEAKER_06: And but AI knows what I've been doing, and I can pull back on that and it can report back for me, and it can help me articulate my value as a as a human being.
• 48:39
  SPEAKER_06: So that's an example of something that I don't think we would have thought of when we first started using it, but it's become a very, very valuable use case because there's a whole bunch of stuff that is associated with it.
• 48:52
  SPEAKER_06: On the flip, and this is where I'm gonna bring you in, Matt, I had a really, really fascinating, let's say, engineering kind of use case with AI.
• 49:02
  SPEAKER_06: I mentioned I've been working on this chief of staff system.
• 49:06
  SPEAKER_06: So I decided that I was gonna have this thing built as though using it to learn a lot about AI's capabilities.
• 49:14
  SPEAKER_06: I decided I was gonna have it work with me as though it were a software engineer or a software architect gathering requirements before building the solution to my problem.
• 49:32
  SPEAKER_06: And so I had it prepare and construct an interview with me.
• 49:39
  SPEAKER_06: The interview was 95 questions, and it took about four hours for it to conduct.
• 49:47
  SPEAKER_06: So it asked the questions, I used Whisperflow, I gave the detailed answers, etc.
• 49:53
  SPEAKER_06: And then that was the basis for what it was building, the roadmap associated with it being built, etc.
• 50:04
  SPEAKER_06: 25,000 lines of code later.
• 50:08
  SPEAKER_06: I that's an accurate number.
• 50:11
  SPEAKER_06: 25 25 it's quite bloated.
• 50:14
  SPEAKER_06: A lot of AR code is quite bloated.
• 50:16
  SPEAKER_06: Quite heavy.
• 50:16
  SPEAKER_06: Yeah, but 25,000 lines of code later.
• 50:20
  SPEAKER_06: I remember when Microsoft passed like 20,000 lines of code in what are the operating systems.
• 50:24
  SPEAKER_06: They're like, oh my god, we're never gonna be able to be able to look at this as a human.
• 50:28
  SPEAKER_06: Anyway, but we get to that point, and I want to transfer over what I've built from what I built in, which was initially Opus, then Fable, and I wanted to transfer it over to Astro to see what Astro could do with it.
• 50:45
  SPEAKER_06: And I was checking to see all of the design documents were there.
• 50:50
  SPEAKER_06: And then I asked a question that it didn't get the answer to quite right.
• 50:55
  SPEAKER_06: And I said, Okay, well the answers to that are in the interview that you conducted with me when we first started this.
• 51:04
  SPEAKER_06: And Fable went, What interview's that?
• 51:08
  SPEAKER_06: What interview?
• 51:11
  SPEAKER_06: And suddenly I'm like, wait a minute, you spent four hours getting verbatim exactly what was required from me, and you don't have that.
• 51:26
  SPEAKER_06: And what was particularly interesting is because of the way in which it compacts conversations and stuff like that, right?
• 51:33
  SPEAKER_06: It could not find it.
• 51:34
  SPEAKER_06: It could not find it anywhere, etc.
• 51:36
  SPEAKER_06: I was able to go back to the original conversation.
• 51:39
  SPEAKER_06: In the original conversation, which it had compacted and all that kind of stuff, I said, I need you to create a copy of this conversation, and we're gonna turn it into a document, and we're gonna put that document in the source materials so that Askra knows what it's building against.
• 51:52
  SPEAKER_06: And even in the context of the conversation, it couldn't do it.
• 51:55
  SPEAKER_06: It was like, I don't know what the conversation what is this conversation?
• 51:57
  SPEAKER_06: Like the lack of contextual awareness for everything that it is doing is just reminded when it goes reminded to when it goes into that.
• 52:03
  SPEAKER_06: By the way, I solved this by building a screen scraper and having it go back and then copying and pulling and copying and pulling and copying and pulling, and after about an hour it got it, and then it created the document and was in it.
• 52:16
  SPEAKER_06: Once Astra found that, it found at least about 20 different occasions where the design had deviated from that original document because it didn't even have it.
• 52:26
  SPEAKER_06: So the the broader point associated for me with all of this is a point that we have talked about already regarding the companies, but I think it's present in the design of what we're dealing with as well.
• 52:43
  SPEAKER_06: I would call it opacity.
• 52:47
  SPEAKER_06: Well crafted procedures, well-written software, well-written, well-run companies have sort of transparency as a core aspect of what they do for good reasons, for good engineering reasons, AI sort of has opacity at the heart of what it does.
• 53:15
  SPEAKER_06: So that when things go wrong, you don't really know or understand how they've gone wrong.
• 53:21
  SPEAKER_06: So as you're very long question, sorry.
• 53:26
  SPEAKER_06: Rant mostly, but as you but as you are going through thinking about what this means for you practically inside your organization and thinking about what it's good at and what it's not good at, does this whole transparency opacity thing factor into okay, well, I'll pull the human out here and I'll throw AI in there?
• 53:52
  SPEAKER_06: That's for you, Matt.
• 53:53
  SPEAKER_01: Um you froze.
• 53:55
  SPEAKER_06: Transparency versus opacity.
• 53:58
  SPEAKER_06: Right?
• 53:58
  SPEAKER_06: Almost like the modern day when this thing first started, and Ethan Mollock was one of the early people who talked about this stuff, he said AI is like a uh hardworking intern that lies a bit.
• 54:12
  SPEAKER_06: Is AI now like a senior software dev that lies a bit?
• 54:18
  SPEAKER_01: Doesn't just lie.
• 54:19
  SPEAKER_01: It completely missed like just completely goes off the rails and without even thinking about it on dumb things.
• 54:29
  SPEAKER_01: It doesn't know dumb things versus really important things, it just knows things.
• 54:34
  SPEAKER_01: So like I'll give you a perfect example where I get so fired up that it's still a problem today.
• 54:40
  SPEAKER_01: Like I'm using it when it's useful, right?
• 54:42
  SPEAKER_01: So I have it do things for me.
• 54:44
  SPEAKER_01: One of the things I have it do for me is pull references with valid hyperlinks in it.
• 54:53
  SPEAKER_01: But now I have to check all the hyperlinks like it's 1992 because a third of them go to a 404 page.
• 55:01
  SPEAKER_01: Or build an app.
• 55:02
  SPEAKER_06: Or build an app or use it.
• 55:03
  SPEAKER_06: Like I had to build the screen scraper to get the thing.
• 55:07
  SPEAKER_01: Yeah.
• 55:07
  SPEAKER_01: I don't want to do that.
• 55:09
  SPEAKER_01: I don't want to do that.
• 55:10
  SPEAKER_01: I said find the link and put it here.
• 55:12
  SPEAKER_01: And like Copilot drives me crazy when it's like, oh, I can't do that.
• 55:16
  SPEAKER_01: It's like, but I'm asking you to do something in Word, your word, aren't you?
• 55:20
  SPEAKER_01: Aren't you Words brain in the background?
• 55:22
  SPEAKER_01: So it's still, as far as we've come, incredibly frustrating.
• 55:27
  SPEAKER_01: So transparency is huge because you got to know why it did what it did if you can engineer it out of doing that and say, okay, I see why I did that, don't do it again.
• 55:37
  SPEAKER_01: I still hold like this, it just skips steps you told it to do.
• 55:41
  SPEAKER_01: You brought this up earlier.
• 55:42
  SPEAKER_01: You're like, it's in the SOP.
• 55:44
  SPEAKER_01: How'd you miss this step?
• 55:45
  SPEAKER_01: I put in when you give me links, check them.
• 55:49
  SPEAKER_01: I still get bad links.
• 55:50
  SPEAKER_01: Well, like it's sometimes worse than a human.
• 55:54
  SPEAKER_01: You would fire this human.
• 55:55
  SPEAKER_01: You'd be like, okay, I've told you this seven times, and I wrote it down, and then and it's like, oh yeah, my bad.
• 56:03
  SPEAKER_01: Okay.
• 56:04
  SPEAKER_01: So transparency is huge, and you have to take small chunks that you can show repeated success and idiot proof for the AIs out of it, and then it's pretty, pretty good.
• 56:16
  SPEAKER_01: But I go all the way back to what you said at the beginning, which is I don't know if these are the right models to do what we're trying to do because now aspirationally we saw what we thought AI would do.
• 56:26
  SPEAKER_01: We're like, that would be cool.
• 56:27
  SPEAKER_01: If that worked, yeah, that would be great.
• 56:30
  SPEAKER_01: It doesn't quite do that, but that got us thinking as to well, what would we do?
• 56:34
  SPEAKER_01: And can we wait until these models or this tech figures itself out so these kind of things don't happen?
• 56:42
  SPEAKER_01: I called it last time, are we playing a game of engagement?
• 56:46
  SPEAKER_01: Like it sounds pretty good, and then if you test it a little bit, it's like, oh yeah, sorry, you were right to call me out on that.
• 56:51
  SPEAKER_01: It's like, I don't want to call you out, I just want you to do it right.
• 56:54
  SPEAKER_01: You're a machine, you know.
• 56:56
  SPEAKER_01: So, in my role in what we're trying to do to build into the AI what is not largely value add from the humans that we have that are very skilled, knowledgeable, and you know, have a lot of years doing these things, is we want to free them up, not only to do more interesting things for themselves, obviously there's a productivity thing for the company, like, oh, now I only need to have ideally five people instead of twenty, and I get the same output.
• 57:32
  SPEAKER_01: I you know, on a spreadsheet at the top of the company, that's good.
• 57:35
  SPEAKER_01: For the humans, it's not great, but that's the goal, right?
• 57:38
  SPEAKER_01: Like, so um I think we're at a really weird time where we better tamp down the hype, up the reality, and show where it is really, really valuable.
• 57:50
  SPEAKER_01: And I think Bill had a great use case there.
• 57:53
  SPEAKER_01: Use it to organize my brain better than my brain organizes the day-to-day that's coming at me, you know.
• 57:59
  SPEAKER_01: Like I can say, hey, when was I I know I saw an email a few weeks ago about this topic.
• 58:03
  SPEAKER_01: I forget who sent it.
• 58:05
  SPEAKER_01: And it's like, well, I found these five things, one of these, you know, like so it's it can dredge big pieces of information very usefully for me.
• 58:14
  SPEAKER_01: But we're at a weird spot and the hype is huge.
• 58:17
  SPEAKER_01: Uh I never thought about it the way you characterize it, Paul, and I couldn't stop laughing when you did set it's like so obvious.
• 58:22
  SPEAKER_01: Yeah, the first two things they told us about this miracle product are really bad.
• 58:28
  SPEAKER_01: Why would we expect it to do well as an IPO?
• 58:32
  SPEAKER_01: Like it's nuts when you really think about it that they've put a major campaign in place that this might kill us.
• 58:39
  SPEAKER_01: We better all get together and sign a paper that says we're gonna stay moral.
• 58:44
  SPEAKER_05: Yeah, I think the other thing that you hit on, Paul, is this aspect of like part of dance requires us to have patience.
• 58:53
  SPEAKER_03: Yeah.
• 58:53
  SPEAKER_05: And the problem is, I'm just laughing here because I'm like, yet again, these technologies are reminding us of our virtues.
• 59:00
  SPEAKER_05: Yeah.
• 59:00
  SPEAKER_05: And in a way, I'm like, okay, patience is a virtue.
• 59:05
  SPEAKER_05: Yeah.
• 59:05
  SPEAKER_05: And yet it's being used in a way against us.
• 59:08
  SPEAKER_05: That's a very meta moment.
• 59:10
  SPEAKER_05: It is.
• 59:11
  SPEAKER_05: So yeah.
• 59:12
  SPEAKER_06: Well, and I think if some good comes out of this for organizations that are trying to figure out how to get value out of this, and for individuals that are trying to figure out what their place is in all of this, is that once you strip away the hype, the right place for AI is things that it does both better than a human and with all the caveats as to what better actually means, right?
• 59:47
  SPEAKER_06: Because better is a strange term as we've discussed previously, but better than a human and better than a technology alternative.
• 59:56
  SPEAKER_06: Right?
• 59:57
  SPEAKER_06: Hence the point about like, okay, you could have AI come up with an initial version of a document, and then instead of a human going through and checking that all the URLs are there, you could write a script to check to see if all the AI is there, right?
• 1:00:12
  SPEAKER_06: So check that all the links are there, right?
• 1:00:14
  SPEAKER_06: We need to get ourselves into a point where we understand the spot that AI fits, and we're using AI for those things specifically, and we're using other things where those other things are most useful.
• 1:00:28
  SPEAKER_06: And we can use AI to help us write those scripts, we can use AI to help us write those pieces of code, right?
• 1:00:33
  SPEAKER_06: But getting away from that idea of it's just like, oh, it's better than humans at these things right now, but then two weeks later it'll be better than more things, so it will just it's just simply the human replacement engine.
• 1:00:47
  SPEAKER_06: No, it's it's more it's more complicated than that.
• 1:00:50
  SPEAKER_06: And by the way, these models are changing personality constantly.
• 1:00:54
  SPEAKER_06: I was using Astro.
• 1:00:55
  SPEAKER_06: Yeah, I'm using Astro today.
• 1:00:57
  SPEAKER_06: Three weeks ago, Astra was building end-to-end for me.
• 1:01:01
  SPEAKER_06: Now Astra's stopping every three and a half minutes saying, Shall I do this?
• 1:01:04
  SPEAKER_06: Shall I do that?
• 1:01:05
  SPEAKER_06: You know, and now all of a sudden it's and now all of a sudden it's slowing me down again.
• 1:01:08
  SPEAKER_06: So so there's there's that element of it as well.
• 1:01:11
  SPEAKER_06: But I guess that's a good point for us to leave it today, gentlemen.
• 1:01:15
  SPEAKER_06: We could again talk about this for hours.
• 1:01:17
  SPEAKER_06: Thanks both for your time today, and we'll be back in about a week to discuss whatever's important then, right?
• 1:01:25
  SPEAKER_06: All right, see what's next.
• 1:01:28
  SPEAKER_06: All right, see you guys.
• 1:01:30
  SPEAKER_05: Great stuff.
• 1:01:31
  SPEAKER_05: Thanks.
• 1:01:32
  SPEAKER_00: Humanity Working is produced by Matt Neal.
• 1:01:35
  SPEAKER_00: If you liked this episode, please take a moment to rate and review us on your favorite podcast platform.
• 1:01:41
  SPEAKER_00: It makes a huge difference in helping others find us.
• 1:01:44
  SPEAKER_00: And don't forget, you can find all episodes and get our weekly newsletter at humanityworking.