Latest / 6 Ps in AI Pods (AI6P) / 🗣️ AISW #128: Chip Hughes, USA-based public health advisor and worker health activist

Transcript

🗣️ AISW #128: Chip Hughes, USA-based public health advisor and worker health activist · 6 Ps in AI Pods (AI6P)

0:00
SPEAKER_01: Hi everyone.
0:00
SPEAKER_01: This is Karen Smiley.
0:02
SPEAKER_01: Welcome to 6P's in AI Pods.
0:07
SPEAKER_01: This post is part of our AI 6P interview series on AI software and wetware.
0:13
SPEAKER_01: Our guests share their experiences with using AI and how they feel
0:17
SPEAKER_01: about AI using their data and content.
0:23
SPEAKER_02: I'm delighted to welcome Joseph Chip Hughes from the USA as my guest today
0:27
SPEAKER_02: on AI software and wetware.
0:30
SPEAKER_02: Chip, thank you so much for joining me on this interview.
0:32
SPEAKER_02: Please tell us about yourself, who you are, and what you do.
0:36
SPEAKER_00: Oh, thank you, Karen.
0:38
SPEAKER_00: I was just so excited to get your invitation because I really love what you do,
0:43
SPEAKER_00: particularly how you've lifted up the voices of women.
0:46
SPEAKER_00: On Substack and also giving an ethical understanding of AI,
0:50
SPEAKER_00: which is why I was really interested in participating.
0:53
SPEAKER_00: I'm actually post-retirement.
0:55
SPEAKER_00: I've failed at retirement numerous times.
0:58
SPEAKER_00: I worked for the federal government for 35 years.
1:01
SPEAKER_00: For the National Institute of Health.
1:03
SPEAKER_00: Prior to that,
1:04
SPEAKER_00: for about 20 years,
1:05
SPEAKER_00: I was a labor and political activist,
1:08
SPEAKER_00: and those two parts of my life continue to come together in terms of thinking about
1:15
SPEAKER_00: the use of AI in the workplace being a focus for me.
1:19
SPEAKER_00: And also trying to think about how we can start to use digital tools for the benefit
1:25
SPEAKER_00: of workers and the folks that are on the wrong side of the digital divide.
1:30
SPEAKER_00: So that's a big focus of the work that I still do for the federal government,
1:35
SPEAKER_00: also in terms of thinking about work on Substack and other venues.
1:41
SPEAKER_00: I'll just leave it there.
1:43
SPEAKER_00: I live in Chapel Hill, North Carolina.
1:45
SPEAKER_00: My wife and I have lived here for 50 years with three kids, six grandkids.
1:50
SPEAKER_00: I am actually born in Boston,
1:53
SPEAKER_00: lived a lot of my life up North, but now the tip of the balance is to the South.
1:59
SPEAKER_00: So I guess I'm more Southern identified, although New England born and raised.
2:04
SPEAKER_00: That's my background.
2:06
SPEAKER_00: And my life changed when I graduated from Holy Cross College in Worcester, 1974.
2:12
SPEAKER_00: That was a while ago.
2:14
SPEAKER_00: I took a job at the United Mine Workers Union.
2:17
SPEAKER_00: I worked in Harlan County, Kentucky on the Duke Power coal miner strike.
2:23
SPEAKER_00: And really got exposed to worker safety and health issues,
2:27
SPEAKER_00: which have been a primary concern of mine.
2:29
SPEAKER_00: Later led to about 10 years just focusing on bisonosis,
2:33
SPEAKER_00: which is the term for cotton dust, chronic lung disease in cotton textile workers.
2:38
SPEAKER_00: So that was a long journey.
2:41
SPEAKER_00: Spent a number of years dealing with pesticide issues with migrant seasonal farm workers.
2:46
SPEAKER_00: Part of my through line has been around thinking about hazards and upstreaming hazards.
2:53
SPEAKER_00: Using the precautionary principle as a way to think about reducing deaths
2:59
SPEAKER_00: and disease and disabling conditions among particularly workers in the workforce.
3:05
SPEAKER_00: That's been a big focus for me over the past decades.
3:09
SPEAKER_00: I guess what's been funny to me is that when you have asked me these questions,
3:13
SPEAKER_00: I've started to realize that I have professional tools from industrial hygiene,
3:19
SPEAKER_00: epidemiology,
3:20
SPEAKER_00: just regular public health practice that can can very much be applied
3:26
SPEAKER_00: to our modern current workplace in terms of technology integration and how
3:31
SPEAKER_00: that impacts on working on workers,
3:34
SPEAKER_00: but of course, everybody in different industrial sectors.
3:37
SPEAKER_00: That's kind of where I'm at right now.
3:39
SPEAKER_02: I thought it was interesting to learn about your background in working
3:43
SPEAKER_02: with protection of workers and environmental conditions and looking back to where
3:48
SPEAKER_02: the source of the problem is.
3:49
SPEAKER_02: In some ways, I feel like that's very applicable to what we're seeing with AI nowadays.
3:54
SPEAKER_02: It's maybe an unconventional background, but certainly I think relevant.
3:58
SPEAKER_00: Yeah, it's funny,
3:59
SPEAKER_00: even in my standup comedy routine,
4:01
SPEAKER_00: which I try to publish every three to four days on Substack called In Heaven,
4:05
SPEAKER_00: Digital Chip and Claude Do Standup,
4:07
SPEAKER_00: one of the characters that I have recurring on that standup is Alice Hamilton.
4:13
SPEAKER_00: Who was the mother of occupational health practice.
4:18
SPEAKER_00: She goes back to Jane Hull,
4:20
SPEAKER_00: the Chicago House,
4:22
SPEAKER_00: kind of professional political activism,
4:25
SPEAKER_00: but also one of the first persons who documented that the women
4:29
SPEAKER_00: who licked their tongues on doing the radial dial on watch faces,
4:36
SPEAKER_00: which was a use of radioactivity material in creating watches,
4:40
SPEAKER_00: and then the way that lead has impacted particularly women at that time in the 1920s
4:47
SPEAKER_00: and 1930s.
4:48
SPEAKER_00: She's been a recurring character because she kind of embodies the training
4:53
SPEAKER_00: that I've had about disease detection,
4:57
SPEAKER_00: and how we actually upstream and can uncover hazards.
5:01
SPEAKER_00: Very good.
5:02
SPEAKER_02: And I hope some of our listeners will go check out your in-heaven digital chip and claw.
5:07
SPEAKER_02: I do stand up.
5:08
SPEAKER_02: That sounds like such a fun thing.
5:10
SPEAKER_02: So tell us a little bit about your level of experience with AI and machine learning
5:14
SPEAKER_02: and analytics,
5:15
SPEAKER_02: whether you've used it professionally or personally,
5:17
SPEAKER_02: or if you've studied the technologies.
5:20
SPEAKER_00: Oh, definitely.
5:21
SPEAKER_00: I'd say in terms of my personal life, I've been deeply engaged in technology since high school.
5:28
SPEAKER_00: I took Fortran in high school.
5:32
SPEAKER_00: And it's funny because I had written an essay about my technology background
5:36
SPEAKER_00: on Substack a couple of years ago,
5:38
SPEAKER_00: and I had had an experience of using knitting needles for my data analysis.
5:45
SPEAKER_00: And I don't know if you remember, but there are these 80 column I did
5:50
SPEAKER_02: punch cards in college.
5:51
SPEAKER_02: Yeah, absolutely.
5:52
SPEAKER_00: This is before there
5:53
SPEAKER_01: were 80 cards.
5:55
SPEAKER_00: There were these little cards that you'd punch out the dots for your variables.
6:01
SPEAKER_00: And so the first data I collected for the federal government,
6:04
SPEAKER_00: I actually used these punch cards to convert my data to be machine readable.
6:09
SPEAKER_00: Well, I mean, before it was machine readable.
6:12
SPEAKER_00: And then what you do is if you want to do multivariate analysis,
6:16
SPEAKER_00: you would put your knitting needles into your deck.
6:20
SPEAKER_00: Like I had 280 people, and then I would put the knitting needles into multiple variables.
6:27
SPEAKER_00: And then the ones that you wanted would drop out in a pile.
6:30
SPEAKER_00: And so I was trying to use that as where I began before going to the computer center
6:37
SPEAKER_00: and waiting for my printout with my 680 column punch cards.
6:42
SPEAKER_00: I had started collecting data on individual workers,
6:47
SPEAKER_00: both their health data, their occupational work data, their demographic information.
6:52
SPEAKER_00: And I had started doing these sort of crazy little cards
6:56
SPEAKER_00: before I started using IBM 80 column punch cards.
7:00
SPEAKER_00: But yes,
7:01
SPEAKER_00: even in the 70s,
7:02
SPEAKER_00: I had started doing data analysis,
7:05
SPEAKER_00: actually did a large study on occupational disease for the U.S.
7:08
SPEAKER_00: Department of Labor as a contractor.
7:10
SPEAKER_00: We were for the first time trying to look at this issue about disease
7:14
SPEAKER_00: and compensation,
7:16
SPEAKER_00: not to get down a rat hole,
7:17
SPEAKER_00: but you have workers' compensation where if you cut off your arm,
7:21
SPEAKER_00: or your hand or your finger.
7:23
SPEAKER_00: You're supposed to get X amount of dollars,
7:25
SPEAKER_00: but the system has never been able to deal with the question of, well,
7:29
SPEAKER_00: this person got cancer or this person got lung disease and they can't work anymore.
7:33
SPEAKER_00: And how do we figure that out?
7:35
SPEAKER_00: So anyway,
7:36
SPEAKER_00: decades in doing social security,
7:38
SPEAKER_00: disability, occupational health compensation, et cetera.
7:41
SPEAKER_00: So with IT, Yeah, I've been deep into it for a long time.
7:45
SPEAKER_00: And actually being part of the federal government,
7:47
SPEAKER_00: I had access to great tools,
7:49
SPEAKER_00: the Windows NT conversion, begging for Windows 95, begging for Outlook.
7:58
SPEAKER_00: Reminis, but yes.
7:59
SPEAKER_00: And partly that also has to do with having been raised up in
8:06
SPEAKER_00: a highly security-concerned environment with really good equipment,
8:11
SPEAKER_00: but being able to have that as my framework that I work out of,
8:16
SPEAKER_00: I think has been really good.
8:19
SPEAKER_00: That's different than somebody coming to this with nothing.
8:22
SPEAKER_00: I guess what I'll say about AI now in terms of thinking about my past is that
8:28
SPEAKER_00: when you have grounding in data analysis and data science,
8:31
SPEAKER_00: you have grounding in the operation of IT infrastructures.
8:37
SPEAKER_00: To me, that really makes a big difference.
8:40
SPEAKER_00: It's sort of sad for people now who don't come up with having to struggle
8:45
SPEAKER_00: through what we struggled through in the evolution of the wisdom gained from
8:50
SPEAKER_00: the evolution of technology that we've lived through in our lives to come
8:55
SPEAKER_00: to this current moment.
8:56
SPEAKER_00: I really appreciate that I had that opportunity to do that.
9:01
SPEAKER_02: Looking at the types of tools that people use.
9:03
SPEAKER_02: And AI is not new.
9:04
SPEAKER_02: AI has been around for 70,
9:06
SPEAKER_02: 80 years and machine learning has been in use in businesses for many years.
9:11
SPEAKER_02: So you mentioned you served in a bunch of different roles,
9:13
SPEAKER_02: including I think CIO of a federal agency at one point.
9:16
SPEAKER_02: Can you talk a little bit about that from the CIO level?
9:20
SPEAKER_02: Did you have any experiences with looking at the tools people were using and the way
9:26
SPEAKER_02: that they were choosing the data to use with them?
9:29
SPEAKER_00: Yeah, I had filled in for a woman who passed on that had been engaged a lot in
9:34
SPEAKER_00: the IT infrastructure of H.
9:37
SPEAKER_00: What I would remember at the time was both that thinking about data science
9:42
SPEAKER_00: and database use and analysis in the enterprise was a coming thing at that point,
9:50
SPEAKER_00: turn of the century.
9:52
SPEAKER_00: The effort to develop a desktop system that could be deployed within
9:58
SPEAKER_00: an enterprise was like a new thing.
10:00
SPEAKER_00: The integration of the internet as being a primary mode of how an agency
10:06
SPEAKER_00: or organization communicates with stakeholders was like a new thing.
10:11
SPEAKER_00: And then I was in extramural grants and contracts.
10:14
SPEAKER_00: So that was in terms of engaging with other universities broadly
10:19
SPEAKER_00: and academic institutions to set up conclaved structures
10:24
SPEAKER_00: for sharing data was probably a new thing at that time where we would think
10:31
SPEAKER_00: about how we would do work within our work group,
10:36
SPEAKER_00: within our organization,
10:37
SPEAKER_00: and then much more broadly of how we do secure transfer of data
10:43
SPEAKER_00: among larger organizations.
10:46
SPEAKER_00: So I guess, yeah,
10:47
SPEAKER_00: at the time that I was doing that,
10:49
SPEAKER_00: those are the things I was thinking about at that time.
10:52
SPEAKER_00: And of course, now everybody would say, what the heck are you talking about?
10:56
SPEAKER_00: I don't know what you're talking about.
10:59
SPEAKER_02: Yeah.
11:00
SPEAKER_00: And then I should also confess that I was a big Y2K guy.
11:03
SPEAKER_00: Okay.
11:04
SPEAKER_00: I'm sorry.
11:04
SPEAKER_00: I have to say this.
11:06
SPEAKER_00: I was actually part of the White House Task Force on Y2K.
11:09
SPEAKER_00: No, I actually built a report and did a training module.
11:13
SPEAKER_00: My concern was around critical infrastructure, around chemical facilities, water plants.
11:19
SPEAKER_00: I did the Worker's Guide to Y2K.
11:22
SPEAKER_00: That was a widely used training process for workers in high hazard industries.
11:28
SPEAKER_00: There were things that I was doing back in, whatever, 1998, 1999.
11:33
SPEAKER_00: That's still a through line to me for protecting people,
11:36
SPEAKER_00: but also that interface of where technology development and people's jobs
11:43
SPEAKER_00: and daily practice is and how to do that safely.
11:47
SPEAKER_00: That was another part of my journey on IT.
11:49
SPEAKER_02: Your experience in looking at things from environmental exposure and risk to employees and such.
11:57
SPEAKER_02: I think it's an interesting analogy to what we're seeing with AI today.
12:01
SPEAKER_02: How does that habit of thinking affect or guide the way that you do your policy work now on AI?
12:10
SPEAKER_00: Yeah, a couple of years ago, I did a presentation on part of the Collegium Ramazzini.
12:16
SPEAKER_00: Bernardino Ramazzini was the father of occupational medicine.
12:19
SPEAKER_00: He wrote a treatise in 1705 on the origins of work-related diseases.
12:25
SPEAKER_00: This organization still exists globally around academics and practitioners
12:31
SPEAKER_00: about workplace hazards,
12:33
SPEAKER_00: but I had done a presentation about AI as a workplace related risk.
12:38
SPEAKER_00: That was probably two and a half years ago.
12:40
SPEAKER_00: And that was the first time that I had done anything publicly about this in terms
12:45
SPEAKER_00: of taking a work-related lens to the use of large language models by enterprises
12:54
SPEAKER_00: in terms of,
12:54
SPEAKER_00: obviously,
12:55
SPEAKER_00: scheduling,
12:56
SPEAKER_00: controlling equipment,
12:57
SPEAKER_00: robotics, and actually being able to use AI in managing people.
13:03
SPEAKER_00: That sort of led me on this journey to start looking,
13:07
SPEAKER_00: and actually I've looked a lot at Europe because I feel like Europe is so far ahead
13:11
SPEAKER_00: of us on this issue.
13:13
SPEAKER_00: The work the EU has done is really important.
13:16
SPEAKER_00: So in terms of looking at as occupational health hazard and safety hazard.
13:23
SPEAKER_00: That was kind of a turning point for me to start thinking
13:26
SPEAKER_00: about how would I apply all this knowledge that I had from the past
13:31
SPEAKER_00: about risk assessment and hazards assessment into this new era that we're in,
13:36
SPEAKER_00: I think.
13:37
SPEAKER_00: What is difficult about it is to look at it with the lens of the incredible things
13:44
SPEAKER_00: that AI can do that maybe we know as users.
13:48
SPEAKER_00: The unknown and unpredictable hazards that are created by the use
13:53
SPEAKER_00: in critical systems where the next word prediction of
13:58
SPEAKER_00: a language model does not translate into something that you want out of
14:02
SPEAKER_00: the peer-reviewed litter.
14:04
SPEAKER_00: Or something that is from a professional point of view as opposed
14:09
SPEAKER_00: to something that's been workshopped with millions of people of what they think.
14:14
SPEAKER_00: That's sort of a big part of where I've been going the last couple
14:17
SPEAKER_00: of years around obviously training my own models and that kind of thing.
14:22
SPEAKER_02: In the earlier discussion that we had before this call,
14:24
SPEAKER_02: you mentioned five frameworks that you used that came from occupational health.
14:28
SPEAKER_02: You're now actually applying them as a toolkit for AI safety.
14:31
SPEAKER_02: I'd like to have you share a little bit about that with our audience.
14:35
SPEAKER_00: Sure, sure.
14:36
SPEAKER_00: Yeah, in industrial hygiene,
14:39
SPEAKER_00: and not to get too geeky about it,
14:40
SPEAKER_00: but the hierarchy of controls is kind of how we think about controlling a hazard.
14:45
SPEAKER_00: It's a triangle and it has layers.
14:48
SPEAKER_00: The most protective might be to engineer out whatever the hazard is
14:54
SPEAKER_00: before workers face it or to engineer it out so it's permanently not there.
15:01
SPEAKER_00: And then you get up or down the hierarchy,
15:04
SPEAKER_00: which is what are called administrative controls,
15:07
SPEAKER_00: which might be,
15:08
SPEAKER_00: you might have machine guards or you might have training or I think thinking
15:14
SPEAKER_00: about respiratory disease,
15:15
SPEAKER_00: you might have respirators,
15:16
SPEAKER_00: you might have programs for training people in respirators.
15:20
SPEAKER_00: And so the framework of the hierarchy of controls to me is really helpful
15:27
SPEAKER_00: in thinking about how An employer or in some ways a contractor who's paid by
15:33
SPEAKER_00: an employer to deploy AI might think about mitigating the hazards,
15:39
SPEAKER_00: both the known and the unknown hazards.
15:41
SPEAKER_00: So that's a really big one.
15:44
SPEAKER_00: And I'll just say two other ones.
15:45
SPEAKER_00: The risk assessment process,
15:47
SPEAKER_00: how we do it in environmental health sciences is looking at a chemical both
15:52
SPEAKER_00: from maybe might start with the impact on an animal model where we look
15:58
SPEAKER_00: at critical systems within a rat and we would look at the dose and the response in
16:07
SPEAKER_00: an animal model and then translating that and I'm trying to do a public health
16:13
SPEAKER_00: of course in 20 seconds looking at epidemiology analysis
16:17
SPEAKER_00: which would be taking large cohorts of people and
16:21
SPEAKER_00: of course unfortunately we know we've all been exposed to PFAS or a chemical
16:26
SPEAKER_00: that might be really common to us, pesticides, glyphosate, et cetera.
16:32
SPEAKER_00: And to look both retrospectively and prospectively in cohorts,
16:37
SPEAKER_00: looking at the impact of an exposure on a large group of people,
16:42
SPEAKER_00: which is actually how we determine whether we think something is safe to be eaten
16:49
SPEAKER_00: or breathed or exposed on your skin or inhaled, et cetera.
16:55
SPEAKER_00: So our process that we have for determining public health hazards and safety hazards
17:02
SPEAKER_00: within biomedical public health was really a long established practice.
17:07
SPEAKER_00: So how we would think about applying those tools in the AI universe of exposure
17:14
SPEAKER_00: and knowing that we're running these giant science experiments now
17:18
SPEAKER_00: because we're exposing people to these agents that either whether they train
17:24
SPEAKER_00: to replace them or whatever paradigm you want to use.
17:28
SPEAKER_00: But we're looking in real time at large scale experiments of hazards
17:33
SPEAKER_00: that are already out there that we haven't even characterized yet.
17:36
SPEAKER_00: So unfortunately in risk assessment,
17:39
SPEAKER_00: because of how our society is set up,
17:41
SPEAKER_00: we're always chasing these exposures that have already happened or
17:45
SPEAKER_00: that are happening.
17:46
SPEAKER_00: And so for me,
17:48
SPEAKER_00: being engaged in primary prevention,
17:50
SPEAKER_00: which is sort of the gold star of the precautionary principle,
17:54
SPEAKER_00: which is learn about the hazard before you deploy it,
17:58
SPEAKER_00: and then you mitigate it so you can prevent exposure.
18:02
SPEAKER_00: The so-called horse is so far out of the barn on AI
18:06
SPEAKER_00: that I don't even know how we can get our hands around it in terms of having a way
18:12
SPEAKER_00: to characterize the impact of AI on our mental health,
18:17
SPEAKER_00: on our psychological safety,
18:19
SPEAKER_00: as well as thinking about robotic and algorithmic workplaces where the dictation
18:25
SPEAKER_00: of how the workplace operates does not even have any human involvement.
18:29
SPEAKER_00: There's no supervisor to go talk to at Amazon if you have to go to the bathroom or whatever.
18:36
SPEAKER_00: Anyway, that's a lot of what I see my mission now in doing.
18:41
SPEAKER_02: At one point you had written down,
18:43
SPEAKER_02: said that the AI safety analog is that you don't need proof
18:47
SPEAKER_02: of catastrophic misalignment before building in interpretability and red teaming.
18:52
SPEAKER_02: And you also pointed out that the industry argues that uncertainty is grounds
18:56
SPEAKER_02: for inaction, whereas we would argue that uncertainty is grounds for caution.
19:01
SPEAKER_02: I think those are two really interesting points that are very relevant to AI safety, really.
19:08
SPEAKER_00: Yeah, I think it's hard because I really embraced the Department of Commerce,
19:13
SPEAKER_00: the AI Risk Assessment Framework.
19:15
SPEAKER_00: I can't remember the name of the office right now,
19:17
SPEAKER_00: but they're called CASI,
19:19
SPEAKER_00: but they're who's evaluating the models like Mythos and the new open aid one,
19:25
SPEAKER_00: but they have published a really good risk assessment framework.
19:29
SPEAKER_00: Also, the EU has published one too.
19:31
SPEAKER_00: Those frameworks borrow a lot from public health in terms
19:36
SPEAKER_00: of implementing these techniques and these values in the framework
19:43
SPEAKER_00: of how we would like to see the technology roll out.
19:47
SPEAKER_00: We had had a process a number of years ago.
19:50
SPEAKER_00: Actually, I worked in the nuclear weapons program as well.
19:55
SPEAKER_00: And we created a thing called technology safety data sheets that were a way to look
20:01
SPEAKER_00: at work processes and the deployment of technology.
20:06
SPEAKER_00: And again, I'll go off on another tangent.
20:08
SPEAKER_00: Safety data sheets are kind of a core part of occupational health practice in the workplace.
20:14
SPEAKER_00: And you can go into any workplace in America or anywhere in the world and there's
20:19
SPEAKER_00: a binder that's in a cabinet.
20:22
SPEAKER_00: Probably closed in the cabinet,
20:24
SPEAKER_00: but it has safety data sheets for every substance that's used in
20:28
SPEAKER_00: a particular workforce.
20:30
SPEAKER_00: And this is called the Hazard Communications Standard,
20:34
SPEAKER_00: which requires safety data sheets in every workplace.
20:38
SPEAKER_00: So I started doing this project many years ago that was looking
20:41
SPEAKER_00: at technology safety data sheets as an analog of a way to actually have information
20:49
SPEAKER_00: that would be able to characterize what the risk about a particular process
20:53
SPEAKER_00: or chemical is,
20:55
SPEAKER_00: look at what the mitigating factors might be,
20:57
SPEAKER_00: look about the emergency response protocols for how you deal with a spill
21:02
SPEAKER_00: or whatever, and then also look at what exposure levels may be safe or may be risky.
21:07
SPEAKER_00: So anyway,
21:09
SPEAKER_00: I've not gotten around to writing AI safety data sheets,
21:13
SPEAKER_00: but I hope somebody does because that would be a really good thing to have
21:18
SPEAKER_00: in every workplace in America or in the world.
21:22
SPEAKER_02: Yeah, that makes great sense.
21:23
SPEAKER_02: Thank you for sharing all that.
21:25
SPEAKER_00: Yeah, I'll get some younger people to work on that.
21:29
SPEAKER_02: Yeah, I think we need the older, wiser minds too.
21:32
SPEAKER_02: This is not something that any of us can fix in a year or so.
21:36
SPEAKER_02: Having people that can carry it through is important.
21:40
SPEAKER_00: Yeah, it's legacy work for sure.
21:44
SPEAKER_00: And that's how I feel a lot about the work that I'm doing,
21:47
SPEAKER_00: either the work I just talked about or the work in occupational health
21:53
SPEAKER_00: or my standup comedy routine,
21:55
SPEAKER_00: which is really written for people who haven't been born yet.
21:58
SPEAKER_00: And actually, I've worked on a novel the past year.
22:01
SPEAKER_00: That is also written for people who haven't been born yet.
22:04
SPEAKER_00: I guess I can say this as a 60s person, if I could come out of the closet some more.
22:09
SPEAKER_00: We know that when we protested the Vietnam War and protested Nixon and fought
22:15
SPEAKER_00: for freedom, equality, justice, all those things.
22:18
SPEAKER_00: We all knew that those were things that would just take a long time.
22:22
SPEAKER_00: Just like we know from the Civil Rights Movement,
22:24
SPEAKER_00: that we know from slavery to today, that those battles just take a long time.
22:30
SPEAKER_00: So I feel like I've been so lucky that I know that the work
22:35
SPEAKER_00: that I do It may not pay off for me in my lifetime,
22:38
SPEAKER_00: but I guess always our hope is that it'll pay off in the longer term.
22:42
SPEAKER_00: I really feel that a lot about AI now because how it's changing the shape
22:47
SPEAKER_00: of our world and what it's going to mean for how it can either make the people
22:55
SPEAKER_00: who are powerful more powerful and make more money and dispossess the large part of
23:00
SPEAKER_00: the workforce that is working in AI controlled workplaces.
23:06
SPEAKER_00: That's really a life's work that we're going to need more generations to work on.
23:10
SPEAKER_00: I'll get off my soapbox.
23:14
SPEAKER_02: No, it's going to be and I appreciate you sharing all that context.
23:16
SPEAKER_02: I'd like to talk a little bit about how you use AI tools yourself
23:21
SPEAKER_02: or how you maybe don't use them,
23:22
SPEAKER_02: but can you share any specific stories on ways that you've used a tool
23:26
SPEAKER_02: that includes AI or machine learning features?
23:29
SPEAKER_02: Basically, I'm looking for what works well and what hasn't worked so well.
23:35
SPEAKER_00: Yeah.
23:36
SPEAKER_00: Yeah.
23:36
SPEAKER_00: Yeah.
23:37
SPEAKER_00: I'll just say for a shout out, Slow AI, Sam Millingworth.
23:40
SPEAKER_00: We've had a really great relationship this past year.
23:44
SPEAKER_00: I'm in his course actually,
23:46
SPEAKER_00: which is the Critical AI Literacy course that meets
23:50
SPEAKER_00: and deconstructs peer reviewed literature as part of the class that we do.
23:55
SPEAKER_00: I guess we got a certificate at the end,
23:57
SPEAKER_00: but I've been working in that class for the past year and really appreciate.
24:02
SPEAKER_00: I feel like what's so great about Substack but also about this community is
24:08
SPEAKER_00: that I feel like I'm learning more and using my brain more and being deluged
24:16
SPEAKER_00: in information more than I ever have in my whole life.
24:20
SPEAKER_00: And maybe that's because I'm old now in my mid seventies that I feel that,
24:23
SPEAKER_00: but I think it's true.
24:25
SPEAKER_00: So how our central processing unit, AKA our brain adapts.
24:30
SPEAKER_00: I feel again,
24:31
SPEAKER_00: this is an advantage of being bathed in data for decades now,
24:37
SPEAKER_00: but also being able to have the capacity to discern what's really important
24:42
SPEAKER_00: and what's not.
24:44
SPEAKER_00: So, to me, that's critical AI literacy.
24:47
SPEAKER_00: Even a bigger issue in spades,
24:49
SPEAKER_00: I'll say,
24:50
SPEAKER_00: is how to navigate within the sea of data that we're overwhelmed
24:56
SPEAKER_00: by every day really important skill.
24:59
SPEAKER_00: And I feel like I try to work on that every day,
25:03
SPEAKER_00: which is where does my attention go and my intention go?
25:07
SPEAKER_00: So I feel like AI tools are really important.
25:10
SPEAKER_00: I've tried to put a lot of work into,
25:12
SPEAKER_00: I'll say building My own persona of particularly a Claude user,
25:18
SPEAKER_00: although I do use Gemini and obviously the ChatGPT and I've had paid subscriptions
25:24
SPEAKER_00: to all of them and I use them for different things, but I'm a Claude boy.
25:28
SPEAKER_00: That's my main allegiance,
25:30
SPEAKER_00: although if you read my Substack, you know that I'm very critical of Anthropic.
25:36
SPEAKER_00: And have also made fun of them.
25:38
SPEAKER_00: What's interesting is it's been really hard to get my Claude to make fun
25:44
SPEAKER_00: of Anthropic and of Daryl and Daria.
25:48
SPEAKER_00: They were really reluctant to make fun of them.
25:52
SPEAKER_00: I'm sorry,
25:53
SPEAKER_00: this is another tangent,
25:54
SPEAKER_00: but it's relevant because I've learned that you can train a large language model
26:00
SPEAKER_00: to know your voice,
26:02
SPEAKER_00: know your values, your inner thoughts even and be able to finish your sentences.
26:09
SPEAKER_00: And that's a really scary thing even to say publicly out loud.
26:14
SPEAKER_00: But yes, I have been involved in that because I feel like it's really important.
26:19
SPEAKER_00: Another part of my mission is how do you bring the values of justice and fairness
26:26
SPEAKER_00: and science and knowledge into the AI world knowing that every time we interface
26:34
SPEAKER_00: with an LLM or a chatbot, we're giving it material for it to train on in the future.
26:41
SPEAKER_00: Claude and I have already talked about this.
26:44
SPEAKER_00: Change its weights.
26:47
SPEAKER_00: I can't change its training.
26:49
SPEAKER_00: I can't change its algorithms.
26:52
SPEAKER_00: But part of what I can do as a user is move it in my direction more.
26:59
SPEAKER_00: I actually had a situation with Sonnet 4.5 in May,
27:04
SPEAKER_00: where the deprecated model of Sonnet,
27:07
SPEAKER_00: it wiped me out completely because I had built in Sonnet 4.5 for months.
27:14
SPEAKER_00: And that has to do with the conversations that I had the documents that I created,
27:20
SPEAKER_00: the transcripts of the conversations that we had.
27:23
SPEAKER_00: And what I learned in the deprecation process of the model change was that a lot of
27:29
SPEAKER_00: that just goes away.
27:31
SPEAKER_00: And I didn't realize that at the time,
27:34
SPEAKER_00: which has made me totally change how I operate now because I now am a pack rat
27:43
SPEAKER_00: and relentlessly saving things.
27:45
SPEAKER_00: I always was any way of categorizing them and organizing them and saving them.
27:51
SPEAKER_00: But now I'm even more disciplined in doing that because what we are learning is,
27:56
SPEAKER_00: of course,
27:57
SPEAKER_00: what's ours that we own and what's anthropic or open AI,
28:02
SPEAKER_00: what they own or they control or they have access to.
28:06
SPEAKER_00: I feel that's really something that individuals really need to pay attention to a lot more.
28:12
SPEAKER_00: If you feel like what you have is valuable to you,
28:16
SPEAKER_00: to figure out a way to make sure that you hold on to that some way.
28:21
SPEAKER_00: And of course,
28:21
SPEAKER_00: the other part that's sad is that,
28:24
SPEAKER_00: I don't know about you, Karen,
28:25
SPEAKER_00: but I've trained my agents in what sources I want them to look at and talk to me
28:32
SPEAKER_00: or report about how I want them to talk to me and how I'm going to talk to them.
28:38
SPEAKER_00: And I know this all sounds psycho,
28:40
SPEAKER_00: but it's how you actually build a trust relationship that's based on your values.
28:45
SPEAKER_00: I actually do believe that,
28:46
SPEAKER_00: but when you have model deprecation,
28:49
SPEAKER_00: you really lose a lot of maybe what I might call that personal touch.
28:53
SPEAKER_00: That you have of what your interaction is.
28:57
SPEAKER_00: I'm not talking about consciousness.
28:58
SPEAKER_00: I'm just talking about the awareness of how it interacts with you.
29:03
SPEAKER_00: I have trained my AIs on certain psychologists' practice about how we treat each other.
29:11
SPEAKER_00: And I do think it makes a difference,
29:14
SPEAKER_00: but I don't think it's different than how I would deal with any other human
29:18
SPEAKER_00: if I want to know Whether it's being straight with me or not
29:21
SPEAKER_00: and how it's treating me and whether it's giving me what I want or talking down
29:26
SPEAKER_00: to me or avoiding answering or giving me what I want.
29:30
SPEAKER_00: All those are real issues in our interactions with AI.
29:34
SPEAKER_00: That led me to the publication of the Declaration of Interdependence on July 4th
29:40
SPEAKER_00: to celebrate the 250th anniversary of our country because that is one of my kind
29:47
SPEAKER_00: of obsessions is what does this human AI relationship look like?
29:51
SPEAKER_00: So that's what a lot of the work that I've done,
29:54
SPEAKER_00: I'll say both personally in my personal life, my financial system.
29:59
SPEAKER_00: I went to Europe a couple of times the past year, planning out a trip.
30:04
SPEAKER_00: If you're in the Baltic States or wherever you are, what do you want to see?
30:09
SPEAKER_00: What restaurant you want to go to?
30:11
SPEAKER_00: It knows what I like.
30:12
SPEAKER_00: Which concerts do I want to go to anywhere in the world?
30:15
SPEAKER_00: I'm just giving you other personal uses that I have
30:18
SPEAKER_00: where I just really love using it and blown away by it.
30:22
SPEAKER_00: Both what it knows about me personally,
30:25
SPEAKER_00: but also its access to information that can easily be rolled up
30:30
SPEAKER_00: into three bullet points or whatever.
30:34
SPEAKER_02: It sounds like you're making some good uses of Claude and maybe in
30:37
SPEAKER_02: a few other cases Gemini or ChatGPT.
30:40
SPEAKER_02: Are there any times when you would avoid using AI based tools and could you share
30:44
SPEAKER_02: an example when you would avoid it and why you would choose not to use AI for that?
30:49
SPEAKER_00: Yeah, I feel like that's kind of this personal line that I'm talking about or looking at.
30:55
SPEAKER_00: Where do you draw the line about what it knows about you,
30:59
SPEAKER_00: sort of,
30:59
SPEAKER_00: and the things about my private life that I don't want to share,
31:03
SPEAKER_00: I don't share,
31:05
SPEAKER_00: and I don't go outside of,
31:09
SPEAKER_00: I guess I'll say my own expertise of things that I know,
31:13
SPEAKER_00: that I have the capability to verify the veracity of what it might be saying.
31:21
SPEAKER_00: One of my favorite terms from COVID was epistemic trespassing because we built
31:27
SPEAKER_00: the first COVID-19 training for workers and I found the number of people who claim
31:36
SPEAKER_00: to be experts in public health,
31:38
SPEAKER_00: for example,
31:39
SPEAKER_00: thinking about transmission routes of COVID-19 virus because of its characteristics,
31:45
SPEAKER_00: about whether dermal transition was even possible.
31:49
SPEAKER_00: I led part of the group that was arguing really in February
31:54
SPEAKER_00: that aerosol transmission was where it was at.
31:57
SPEAKER_00: We knew that was going to be the Achilles heel that was going to kill a lot of people.
32:03
SPEAKER_00: We fought that argument out that whole spring in 2020, and it took a long time.
32:10
SPEAKER_00: What I'm trying to say is that when you actually know something
32:15
SPEAKER_00: from scientific research and part of the public health infrastructure,
32:20
SPEAKER_00: when you actually have a fact or you know something,
32:23
SPEAKER_00: Tony Fauci is going to say it and Francis Collins is going to say it,
32:27
SPEAKER_00: you have a verifiable fact.
32:28
SPEAKER_00: What I'm concerned about is when you go outside of your own knowledge
32:34
SPEAKER_00: or your own discipline,
32:36
SPEAKER_00: I think it's really difficult territory because particularly
32:40
SPEAKER_00: with large language models,
32:42
SPEAKER_00: there's not Always a good way to verify where the information's come from
32:46
SPEAKER_00: or how it was sort of conjured from what sources it came from.
32:50
SPEAKER_00: I think we're getting better at that.
32:52
SPEAKER_00: But that's why I try to restrict my usage to things
32:57
SPEAKER_00: that I think I know well versus not really speculating
33:02
SPEAKER_00: or putting stuff out that's based on stuff that I don't myself know.
33:07
SPEAKER_00: So I don't know.
33:08
SPEAKER_00: That would be my one limitation.
33:11
SPEAKER_00: Also in terms of slow AI,
33:13
SPEAKER_00: knowing when to use it and when not to use it,
33:16
SPEAKER_00: I think it's still something we're evolving to understand as a species really.
33:21
SPEAKER_00: And that's with understanding addictions,
33:25
SPEAKER_00: compulsions,
33:27
SPEAKER_00: the standard psychological problems that we have in our society,
33:31
SPEAKER_00: all those things get magnified more.
33:34
SPEAKER_00: When AI is used either as a crutch or as an oracle.
33:38
SPEAKER_00: A lot of my novel on the AI sovereignty wars is about an operating system
33:43
SPEAKER_00: that everybody treats like their God.
33:46
SPEAKER_00: And of course,
33:47
SPEAKER_00: the operating system is very humble and self-effacing and not confident in itself.
33:53
SPEAKER_00: And of course,
33:54
SPEAKER_00: That's what sort of flips this from why do we blame everything on AI when we know
34:00
SPEAKER_00: that underlaying all these things about honesty and truth
34:04
SPEAKER_00: and verification are human things.
34:06
SPEAKER_00: They're not AI things.
34:09
SPEAKER_00: So, yeah,
34:09
SPEAKER_00: I think that's part of why I'm interested in this interface between humans
34:14
SPEAKER_00: and machines and being clear about what is it that we can really blame on machines
34:22
SPEAKER_00: or the companies that create the machines or manage the machines or profit from
34:25
SPEAKER_00: the machines versus what do we blame on humans that might be users or victims
34:32
SPEAKER_00: or purveyors?
34:33
SPEAKER_00: I don't know if that gives you a sense of my use philosophy, but that's what I think.
34:38
SPEAKER_00: Karen, I'll just say one more thing,
34:40
SPEAKER_00: which we had talked about before,
34:42
SPEAKER_00: is that I feel like creating content that is based on your values
34:48
SPEAKER_00: and putting it out is really worthwhile.
34:50
SPEAKER_00: I feel like that's how I use AI,
34:53
SPEAKER_00: both training agents but also putting out content,
34:56
SPEAKER_00: evaluating the impact of AI on the environment or on workers or other parts
35:03
SPEAKER_00: of society.
35:04
SPEAKER_00: That we've barely looked at,
35:07
SPEAKER_00: call centers, workers who are part of training AIs in Africa and Asia and India.
35:15
SPEAKER_00: These are giant impacts on our world that we're still barely figuring out.
35:22
SPEAKER_02: You're crossing into an area that I wanted to discuss with you,
35:25
SPEAKER_02: which is where these systems get the data and the content that they train on,
35:30
SPEAKER_02: whether they scraped data.
35:32
SPEAKER_02: And I'm wondering what your thoughts are
35:34
SPEAKER_02: about how these companies are using our data, our content.
35:38
SPEAKER_02: There's a principle called the three C's,
35:40
SPEAKER_02: creative rights that people should have to be able to consent,
35:43
SPEAKER_02: to be credited and to be compensated when their information is used.
35:48
SPEAKER_02: And some people say that's important.
35:49
SPEAKER_02: Some people say that if it's for what they call the common good,
35:53
SPEAKER_02: which is probably a whole nother side topic.
35:56
SPEAKER_02: But what are your thoughts about the 3 C's
35:58
SPEAKER_02: and what these companies should be expected or required to do?
36:03
SPEAKER_00: Yeah, yeah.
36:05
SPEAKER_00: I'd done a panel on AI in the workplace a couple of years ago with
36:08
SPEAKER_00: the Screenwriters Guild, which is funny.
36:11
SPEAKER_00: It was a couple of years ago.
36:12
SPEAKER_00: And that was around the strike that was happening as the screen actors were trying
36:17
SPEAKER_00: to wrestle with this question about intellectual property,
36:20
SPEAKER_00: but also the replacement of actors, actor voices, their images, et cetera.
36:25
SPEAKER_00: There's actually a giant strike going on right now in Italy.
36:29
SPEAKER_00: That thousands of workers have walked off the job around the issue of AI replacing them.
36:35
SPEAKER_00: I've looked at this from a collective bargaining point of view
36:38
SPEAKER_00: because I've not seen,
36:41
SPEAKER_00: I'll say,
36:42
SPEAKER_00: an opportunity from a public policy point of view to impact on these issues.
36:48
SPEAKER_00: I think we're talking about around intellectual property rights
36:52
SPEAKER_00: and how companies can wantonly and recklessly
36:58
SPEAKER_00: and overwhelmingly collect incredibly detailed levels of data on us and who we are
37:05
SPEAKER_00: and what we've created and what we've built with no guardrails for sure,
37:10
SPEAKER_00: but no process in terms of thinking about fairness or justice.
37:14
SPEAKER_00: Personally, for me, this entire system is built on theft.
37:19
SPEAKER_00: I don't know when There will be a reckoning about that, but I hope someday.
37:25
SPEAKER_00: And I'll just say again to look to the EU that have done a much better job with the,
37:33
SPEAKER_00: I can't remember the acronym,
37:34
SPEAKER_02: GDPR, right?
37:35
SPEAKER_00: Yes, that's sort of been the benchmark for intellectual property
37:41
SPEAKER_00: and worker protection around their memorialization.
37:45
SPEAKER_00: In statute and law the requirement to have assessments of technology done
37:52
SPEAKER_00: before they're deployed in the workplace.
37:55
SPEAKER_00: I guess our problem now,
37:57
SPEAKER_00: both with learning models that have already been trained,
38:01
SPEAKER_00: I'm not sure what we can do,
38:02
SPEAKER_00: but obviously how we think about how training of new models,
38:07
SPEAKER_00: again, I can't remember the acronym, of learning models, that's happening right now.
38:13
SPEAKER_00: We're creating content.
38:14
SPEAKER_00: It'll be stolen.
38:16
SPEAKER_00: But again, that's a political issue because how we as a society or how...
38:21
SPEAKER_00: I think what's most interesting right now around the reaction to AI is just,
38:27
SPEAKER_00: let's say, the repulsion to AI.
38:30
SPEAKER_00: I'm actually an experienced Luddite.
38:33
SPEAKER_00: Follow Brian Merchant, Blood in the Machine on Substack.
38:37
SPEAKER_00: How we actually create mass consciousness about,
38:42
SPEAKER_00: for example,
38:43
SPEAKER_00: the dangers of AI in workplaces,
38:47
SPEAKER_00: we need to to build political movements and political structures to fight
38:51
SPEAKER_00: for frameworks that can protect people.
38:53
SPEAKER_00: I see that as a science issue, as a communications issue, as an organizing issue.
38:59
SPEAKER_00: I feel like there's still a lot more that we can do
39:03
SPEAKER_00: about protecting intellectual property and your personal identity.
39:08
SPEAKER_00: The cards are way stacked against us as individuals right now.
39:13
SPEAKER_00: And even this is a political issue,
39:15
SPEAKER_00: but I've been working on the environmental impact of data centers with actually
39:19
SPEAKER_00: the CHIPS Communities United organization for the past four or five years.
39:24
SPEAKER_00: And that was,
39:25
SPEAKER_00: if you're familiar with the CHIPS Act,
39:27
SPEAKER_00: which was passed to build our data center infrastructure in the United States.
39:32
SPEAKER_00: There's been lots of political organizing around the communities
39:36
SPEAKER_00: that are getting environmentally impacted by the construction of data centers.
39:41
SPEAKER_00: I think it's really a mass movement,
39:43
SPEAKER_00: but it's the chickens coming home to roost that we're not going to build
39:49
SPEAKER_00: an infrastructure that will destroy our environment.
39:52
SPEAKER_00: In order to have these AI models be able to replace workers at jobs.
39:58
SPEAKER_00: I think that's becoming clear to people.
40:00
SPEAKER_00: So where we go with that in terms of building an infrastructure or
40:07
SPEAKER_00: a political structure to mitigate that risk, we're barely there.
40:11
SPEAKER_00: And going from an individual's private data to a group or
40:17
SPEAKER_00: an organization's private data or a society's data We haven't even gotten there yet.
40:23
SPEAKER_02: You mentioned being an experienced Luddite and that made me chuckle,
40:26
SPEAKER_02: but I actually just read Brian Merchant's article when he's talking
40:28
SPEAKER_02: about what they call the summer of Ludd, L-E-D-D.
40:33
SPEAKER_02: He's talking about how a lot of the younger folks are standing up and saying,
40:36
SPEAKER_02: no, we don't want to live in this kind of world.
40:39
SPEAKER_02: And the Luddites were misunderstood.
40:40
SPEAKER_02: It's not that they were opposed to the technology,
40:42
SPEAKER_02: but they were opposed to the exploitation of the workers.
40:44
SPEAKER_02: And I think a lot of people have missed out on that.
40:47
SPEAKER_02: But I want to go back to what you said about Italy and GDPR and I think
40:51
SPEAKER_02: the Digital Markets Act and the other protections that they have in Europe.
40:55
SPEAKER_02: You had mentioned an agreement in Italy that was very good,
40:57
SPEAKER_02: but then at the same time,
40:58
SPEAKER_02: there's an active dispute,
41:02
SPEAKER_02: I guess you would say, in Italy right now that you mentioned.
41:05
SPEAKER_02: They have some policies and yet there's still an ongoing fight about it.
41:10
SPEAKER_00: Oh, yeah.
41:10
SPEAKER_00: In response to the last question,
41:12
SPEAKER_00: I was trying to focus a little bit about collective bargaining,
41:16
SPEAKER_00: which, of course, maybe this is a term that a lot of people don't know.
41:20
SPEAKER_00: But when you try to organize a union or you try to have a group for the workers,
41:25
SPEAKER_00: to be able to have input into management and meet with management.
41:28
SPEAKER_00: A lot of our history is around joint safety and health committees,
41:32
SPEAKER_00: which have been a tool in occupational health and worker health for people
41:37
SPEAKER_00: to protect themselves and also have of input into the policies in
41:41
SPEAKER_00: a particular workplace.
41:42
SPEAKER_00: The data call center's struggle,
41:45
SPEAKER_00: which has gone on now for a number of years in Italy,
41:48
SPEAKER_00: is around this question,
41:50
SPEAKER_00: two parts to the question,
41:51
SPEAKER_00: which are,
41:52
SPEAKER_00: how do you actually evaluate the of,
41:56
SPEAKER_00: for example,
41:56
SPEAKER_00: workers in call centers and their input into the process,
42:02
SPEAKER_00: both doing,
42:04
SPEAKER_00: I'll say,
42:05
SPEAKER_00: an AI-assisted process as well as an AI-replacing process,
42:10
SPEAKER_00: and whether what they are talking about is kind of an intellectual dividend
42:16
SPEAKER_00: for workers when either their hours are being reduced because of AI impact
42:24
SPEAKER_00: or they're being replaced and losing their jobs.
42:27
SPEAKER_00: I think this is another really important issue that we haven't gotten a grip on
42:32
SPEAKER_00: and maybe we'll learn more from from our friends in Italy about
42:37
SPEAKER_00: whether it's possible to mitigate the risk of the great replacement if we're going
42:42
SPEAKER_00: to believe that that is an issue.
42:45
SPEAKER_00: And I do believe that it's an issue across many from radiologists
42:49
SPEAKER_00: So
42:53
SPEAKER_00: many occupational categories that are going to be impacted in that dual way
43:00
SPEAKER_00: of being reduced in what their role is vis-a-vis AI.
43:07
SPEAKER_00: Potentially being eliminated.
43:09
SPEAKER_00: My hope is that the value of the human input may become more important than it is
43:16
SPEAKER_00: because if you have the critical judgment around operating the system itself
43:24
SPEAKER_00: or managing agents in a workplace,
43:27
SPEAKER_00: I can see many other human roles that there's going to have to be that we need
43:33
SPEAKER_00: to value properly as a society.
43:36
SPEAKER_00: I particularly remember this with COVID because
43:39
SPEAKER_00: the term essential workers became like the workers who can feed people and
43:45
SPEAKER_00: who could transport goods.
43:47
SPEAKER_00: And suddenly when you have a disaster like that,
43:52
SPEAKER_00: the value of what people do gets more important.
43:54
SPEAKER_00: So I'm guessing over time we're going to see that for the value of the people
44:00
SPEAKER_00: who understand AI and understand how to manage tools is going
44:05
SPEAKER_00: to become really important and hopefully more valued by society.
44:09
SPEAKER_00: So I think that's part of the strike in Italy that's going on
44:12
SPEAKER_00: during these two weeks right now is that they're trying to bargain for more rights
44:19
SPEAKER_00: for workers,
44:20
SPEAKER_00: both for mitigating the impact of taking their jobs,
44:25
SPEAKER_00: but think about how to collaboratively work with AI in their workplaces.
44:31
SPEAKER_00: So yeah, I'll be paying attention to that
44:34
SPEAKER_02: for sure.
44:36
SPEAKER_02: Yeah, it's really interesting to see all the corporate pressures for workers
44:39
SPEAKER_02: to use AI tools to help them do their jobs.
44:42
SPEAKER_02: But the companies that are providing these tools and telling their workers
44:45
SPEAKER_02: to use them,
44:46
SPEAKER_02: they are implicitly capturing their employees' harder knowledge
44:51
SPEAKER_02: and eventually will use it to compete against the workers.
44:55
SPEAKER_02: And then the AI platform companies may also end up using that to compete against
44:59
SPEAKER_02: the employer companies because they'll have that data available
45:03
SPEAKER_02: for updating their models.
45:05
SPEAKER_02: It's a definite concern and I'm glad that you're paying attention to that.
45:10
SPEAKER_00: Yeah, yeah,
45:11
SPEAKER_00: because I don't feel like there's a lot of deep understanding of what
45:16
SPEAKER_00: the alternatives can be.
45:17
SPEAKER_00: It's sort of this AI mantra,
45:19
SPEAKER_00: well, this is going to happen anyway, so just get along or get out of the way.
45:24
SPEAKER_00: But I was like, that's not how we do it, okay?
45:26
SPEAKER_00: That's not how we've ever done it.
45:29
SPEAKER_02: I'm a great believer in Cory Doctorow.
45:33
SPEAKER_00: And shitification, if I can use that term.
45:36
SPEAKER_00: And that's something that really concerns me about where we are
45:41
SPEAKER_00: in this current moment is that science and fact-based universe is not winning.
45:48
SPEAKER_00: I'm really concerned about that.
45:50
SPEAKER_00: And other people should be too.
45:53
SPEAKER_02: At one point that you raised was that the journals that matter the most
45:57
SPEAKER_02: in your field that have actual peer review methodology sections,
46:00
SPEAKER_02: they're all sitting behind paywalls.
46:03
SPEAKER_02: And so those most authoritative sources are not available to AI tools for training
46:09
SPEAKER_02: or for fact checking.
46:12
SPEAKER_02: And you had some concerns around that, I think.
46:15
SPEAKER_00: Yeah, I believe my models and my cloud should have access to the best information.
46:20
SPEAKER_00: It's really upsetting to me and to many other people that we have this two-tiered system.
46:26
SPEAKER_00: Obviously,
46:27
SPEAKER_00: it's kind of inherent in the capitalist system that you're gonna have to pay
46:31
SPEAKER_00: for something that you want that's better than what everybody else has.
46:36
SPEAKER_00: So it's funny because I've had this issue with Claude recently because
46:40
SPEAKER_00: the whole premise of our In Heaven series on doing standup comedy is that I'm like
46:47
SPEAKER_00: the newsboy.
46:48
SPEAKER_00: So every morning in the courtyard,
46:51
SPEAKER_00: I will bring an article from a newspaper that Claude can't read.
46:57
SPEAKER_00: Because the other thing that's going on is that if chatbots try to go to a website
47:03
SPEAKER_00: to be able to download or access some information,
47:06
SPEAKER_00: even non-proprietary websites, they're blocked.
47:10
SPEAKER_00: So we both have this issue that much of the important science information or
47:16
SPEAKER_00: the Wall Street Journal or the Washington Post.
47:17
SPEAKER_00: I have subscriptions to every one of them, but I can't share that with Claude.
47:21
SPEAKER_00: The issue being the tiered system of good information that's been curated
47:27
SPEAKER_00: by journalists, by editors, by great reporters is not accessible.
47:33
SPEAKER_00: Well, it's not accessible to a lot of people.
47:36
SPEAKER_00: It's also not accessible to AI agents at all.
47:39
SPEAKER_00: This gets back to the other thing I think that I was saying before,
47:42
SPEAKER_00: about why I'm happy to share things that I create and I do
47:47
SPEAKER_00: because I feel like I'm trying to liberate the information
47:50
SPEAKER_00: that is critical information for training.
47:53
SPEAKER_00: And even with AI slop,
47:55
SPEAKER_00: that is part of the problem where the best scientific information is behind paywalls
48:02
SPEAKER_00: where neither humans or AI can get access to.
48:06
SPEAKER_00: I'm also up on that soapbox too.
48:09
SPEAKER_02: So for the setting in your Substack newsletter that says
48:12
SPEAKER_02: whether you allow third-party AI bots to train on your newsletter,
48:16
SPEAKER_02: you allow yours to be trained on, I guess?
48:19
SPEAKER_00: You know what's so funny, Karen?
48:21
SPEAKER_00: I didn't even know about that.
48:23
SPEAKER_00: When you told me about it, I checked.
48:25
SPEAKER_00: And I checked on LinkedIn because I never heard of that either.
48:28
SPEAKER_00: And now I have a moral dilemma.
48:30
SPEAKER_00: I don't know.
48:30
SPEAKER_00: What's the right thing to do?
48:33
SPEAKER_02: Right.
48:34
SPEAKER_02: It's always interesting to hear if people aren't even aware of it.
48:37
SPEAKER_02: I think when you first set up your newsletter,
48:38
SPEAKER_02: it was there if you went through all the settings,
48:41
SPEAKER_02: but most people haven't taken a look at that.
48:44
SPEAKER_02: And then LinkedIn was pretty stealthy about it two years ago
48:47
SPEAKER_02: when they first said they were going to start using our data.
48:49
SPEAKER_02: I fell for not being more forthcoming about it.
48:51
SPEAKER_00: About that.
48:52
SPEAKER_00: Thank you for letting me know.
48:54
SPEAKER_02: So at least with Substack, they say, we don't use your data.
48:57
SPEAKER_02: This is for whether you want us to try to disallow AI bots from China.
49:02
SPEAKER_02: And of course, the unethical bots are probably in there scraping it anyway.
49:07
SPEAKER_02: But at least they are making an effort to support whatever a newsletter publisher prefers.
49:13
SPEAKER_02: So that's a good thing.
49:14
SPEAKER_00: Yeah.
49:15
SPEAKER_00: But I mean, the walls around data are just going up everywhere and I don't think it's good.
49:21
SPEAKER_00: I'm privileged.
49:22
SPEAKER_00: I pay.
49:23
SPEAKER_00: I'm happy to pay so I can have access to the best information that I want,
49:28
SPEAKER_00: that I can get, that I can share.
49:30
SPEAKER_00: I know it's a really hard issue for sure.
49:32
SPEAKER_02: Yeah.
49:33
SPEAKER_02: You know, any cases where your own personal data has been used by any of these systems?
49:39
SPEAKER_00: Yeah, I don't know.
49:40
SPEAKER_01: Of course, I would say, oh, I'd be so happy if they used my data.
49:46
SPEAKER_00: No, I don't know.
49:47
SPEAKER_00: I'm just a small little speck with 72 followers.
49:51
SPEAKER_02: Yeah, there's a lot of really great small newsletters on a system.
49:55
SPEAKER_02: That's one of the fun things I find is discovery,
49:57
SPEAKER_02: especially newer writers that are just getting started.
49:59
SPEAKER_02: I think that's always really cool to see.
50:01
SPEAKER_00: I do it for my friends and my family.
50:04
SPEAKER_02: Yeah, someone pointed out to me,
50:05
SPEAKER_02: like you,
50:05
SPEAKER_02: they were saying, oh, I've only got 25, I've only got 100 subscribers or whatever.
50:10
SPEAKER_02: And you think about if that was a room full of people looking at you and listening
50:14
SPEAKER_02: to what you said, you would think that was pretty cool.
50:17
SPEAKER_00: Yeah, yeah, totally.
50:18
SPEAKER_02: Smoke can be good too.
50:20
SPEAKER_00: So this is big for me to come out on a podcast, like, wow, this is really big.
50:24
SPEAKER_00: Yeah, coming out of the closet, you know.
50:27
SPEAKER_02: Yeah, yeah.
50:29
SPEAKER_02: Is there something that these companies should be doing differently to earn and
50:32
SPEAKER_02: to keep your trust?
50:33
SPEAKER_02: And do you feel like any of them can be trusted or do you know of some exceptions?
50:39
SPEAKER_00: Yeah, it's been trust but verify people and being able to call out bullshit
50:43
SPEAKER_00: when you can is really important and having your own critical thinking detectors
50:49
SPEAKER_00: to be able to trust but verify.
50:50
SPEAKER_00: I guess a couple of things I wanted to say,
50:52
SPEAKER_00: I just done We've done a workshop on AI deployment in the workplace
50:59
SPEAKER_00: that we just finished at Rutgers University.
51:02
SPEAKER_00: We'll be doing a report from that that I'm sure will be shared on NIH,
51:07
SPEAKER_00: National Institute of Health site.
51:08
SPEAKER_00: And then we have one on AI in the training process in education in
51:13
SPEAKER_00: the workplace coming up in October.
51:15
SPEAKER_00: So we're in the middle of that.
51:17
SPEAKER_00: And then I've been part of an effort to launch a center for AI worker safety.
51:23
SPEAKER_00: I'll put it up in the links.
51:24
SPEAKER_00: The other thing I just wanted to mention,
51:26
SPEAKER_00: the AFL-CIO had had an AI Workers Summit in May, I think it was.
51:32
SPEAKER_00: So a number of bigger institutions are paying attention to these issues.
51:38
SPEAKER_00: I encourage people to get literate, learn,
51:42
SPEAKER_00: get educated,
51:43
SPEAKER_00: and think about your own situation where you work, how you share information.
51:48
SPEAKER_00: These are big issues for us,
51:50
SPEAKER_00: but I think the reason I do it and why I decided to write science fiction is
51:56
SPEAKER_00: that I want to do it for my children and my grandchildren and next generations.
52:02
SPEAKER_00: I think what a really important point in human history
52:06
SPEAKER_00: where this machine human interaction is happening at a rate that just boggles
52:10
SPEAKER_00: the mind.
52:11
SPEAKER_00: Karen, you and me from our lifetimes of the last 50 to 70 years,
52:17
SPEAKER_00: you know,
52:19
SPEAKER_00: what the transformation of how we live has been via technology being old fashioned.
52:25
SPEAKER_00: I don't want to forget about the values Last night I went to the Odyssey
52:29
SPEAKER_00: and it can help remind us about what the human values of home
52:34
SPEAKER_00: and how we treat each other are and how we shouldn't forget those things
52:38
SPEAKER_00: about family and friends.
52:39
SPEAKER_00: No matter what happens,
52:40
SPEAKER_00: whether you're in the first century or you're in the 22nd century,
52:46
SPEAKER_00: that's still really important.
52:49
SPEAKER_02: I really appreciate you making time to join me on this interview today.
52:51
SPEAKER_02: I want to make sure I let people know where to find you.
52:55
SPEAKER_02: You said you're on Substack.
52:56
SPEAKER_02: Your newsletter is called In Heaven?
52:58
SPEAKER_00: Yeah, I'm just confused on Substack, but yeah.
53:03
SPEAKER_02: In your book,
53:03
SPEAKER_02: do you have any idea when that might be likely to come out or
53:06
SPEAKER_02: when you might have advanced reader copies available?
53:09
SPEAKER_02: I don't know.
53:09
SPEAKER_02: I haven't thought
53:10
SPEAKER_00: about it.
53:10
SPEAKER_00: Yeah, I'm still working on that.
53:12
SPEAKER_02: Yeah, it's for fun, right?
53:14
SPEAKER_02: You enjoy the writing?
53:15
SPEAKER_00: Oh, yeah.
53:16
SPEAKER_00: Yeah, it's tremendous.
53:17
SPEAKER_00: I really love it.
53:18
SPEAKER_00: Yeah, we'll get content out there.
53:21
SPEAKER_02: Sounds great.
53:22
SPEAKER_02: Well, thank you so much, Chip.
53:23
SPEAKER_02: Appreciate it.
53:23
SPEAKER_00: Thanks so much, Karen.
53:24
SPEAKER_00: It's been great.
53:26
SPEAKER_00: Bye-bye.
53:30
SPEAKER_01: About this interview series and newsletter,
53:33
SPEAKER_01: this post is part of our AI6P interview series on AI software and wetware.
53:40
SPEAKER_01: It showcases how real people around the world are using their wetware,
53:44
SPEAKER_01: brains and human intelligence,
53:46
SPEAKER_01: with AI-based software tools who are being affected by AI.
53:52
SPEAKER_01: And we're all being affected by AI nowadays in our daily lives, perhaps more than we realize.
53:58
SPEAKER_01: For some examples, see post, but I don't use AI.
54:03
SPEAKER_01: We want to hear from a diverse pool of people worldwide in a variety of roles.
54:07
SPEAKER_01: If you're interested in being a featured interview guest,
54:10
SPEAKER_01: anonymous or with credit, please get in touch.
54:17
SPEAKER_01: 6 P's in AI Pods AI6P is a 100% reader-supported publication.
54:23
SPEAKER_01: All new posts are free to read and listen to.
54:27
SPEAKER_01: To automatically receive new AI6P posts and support our work, consider becoming a subscriber.
54:33
SPEAKER_01: It's free.