Latest / 6 Ps in AI Pods (AI6P) / 🗣️ AISW #132: Effrosyni Paza, Italy-based global account director, advisor, and author

Transcript

🗣️ AISW #132: Effrosyni Paza, Italy-based global account director, advisor, and author · 6 Ps in AI Pods (AI6P)

0:00
SPEAKER_01: Hi everyone, 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
0:16
SPEAKER_00: how they feel
0:17
SPEAKER_00: about AI using their data and content.
0:20
SPEAKER_01: I'm delighted to welcome Efrosini Pazza from Italy as my guest today on AI,
0:24
SPEAKER_01: software, and wetware.
0:26
SPEAKER_01: Efrosini, thank you so much for joining me for this interview.
0:29
SPEAKER_01: Please tell us about yourself, who you are, and what you do.
0:34
SPEAKER_00: Thank you, Karen, for inviting me, and congratulations on this wonderful initiative.
0:40
SPEAKER_00: I'm currently based in Rome,
0:42
SPEAKER_00: Italy, and I've spent near 30 years in telecom and enterprise technology.
0:49
SPEAKER_00: I began in Greece,
0:50
SPEAKER_00: my home country,
0:52
SPEAKER_00: and then lived and worked in diverse markets like Romania, Sweden, Qatar, and Italy.
0:59
SPEAKER_00: During that time,
1:01
SPEAKER_00: I led major accounts,
1:03
SPEAKER_00: Deutsche Telekom, Oridu,
1:05
SPEAKER_00: an Arabic group,
1:08
SPEAKER_00: and Vodafone,
1:09
SPEAKER_00: at companies like Ericsson, HP, and Sienna Software Division, Blue Planet.
1:18
SPEAKER_00: I've always worked at the intersection of technology vendors and customers,
1:24
SPEAKER_00: meaning I sat between what a vendor was building and what a customer was buying
1:31
SPEAKER_00: to fulfill a business need.
1:34
SPEAKER_00: What surprised me was seeing well-funded programs that successfully completed
1:43
SPEAKER_00: the technical and commercial validation to suddenly stall for reasons
1:50
SPEAKER_00: that had little to do with whether the technology was advanced or innovative.
1:57
SPEAKER_00: Sometimes rust had not been properly built or had been shaken.
2:04
SPEAKER_00: For example, a delivery failure damaged confidence in the vendor.
2:09
SPEAKER_00: But the vendor assumed that recovering the program would automatically rebuild
2:16
SPEAKER_00: the relationship and restore the pipeline in the short to medium term.
2:22
SPEAKER_00: That assumption can create pressure on customers
2:27
SPEAKER_00: to accelerate decisions they are not ready to make.
2:31
SPEAKER_00: That's rarely how it works.
2:34
SPEAKER_00: Recovering grand reputation and rebuilding the relationships can be a long
2:40
SPEAKER_00: and painful journey and many vendors underestimate how long
2:46
SPEAKER_00: and how personally demanding it actually is and sometimes leadership becomes weak
2:56
SPEAKER_00: at exactly the point when sustained attention and commitments are needed.
3:03
SPEAKER_00: In large corporate environments, you might expect things to work differently.
3:09
SPEAKER_00: They don't always.
3:11
SPEAKER_00: And I think that's worth discussing.
3:15
SPEAKER_01: You've had some interesting experiences in corporate life then.
3:18
SPEAKER_01: What have those experiences taught you?
3:22
SPEAKER_00: Yeah, they taught me something very simple.
3:26
SPEAKER_00: If you genuinely care about your customer, that's the recipe for success.
3:33
SPEAKER_00: It sounds self-evident.
3:35
SPEAKER_00: It is not.
3:37
SPEAKER_00: Business is about human relationships and give and take.
3:41
SPEAKER_00: But I'm saying something more specific.
3:45
SPEAKER_00: How deeply do you understand what your customer is risking?
3:49
SPEAKER_00: What is their most urgent need?
3:52
SPEAKER_00: And do you really give the patience and the time it requires?
3:57
SPEAKER_00: That's harder than it sounds because organizations run on quarterly
4:03
SPEAKER_00: and annual financial targets and sales quotas are built around closing deals,
4:11
SPEAKER_00: registering revenue within the fiscal year.
4:15
SPEAKER_00: That pressure is real and at a certain point it becomes easier
4:21
SPEAKER_00: to stop truly listening to the customer in front of you
4:25
SPEAKER_00: and turn your attention elsewhere to hit the numbers faster.
4:31
SPEAKER_00: My argument is that you should never treat a customer's pain as a sales opportunity.
4:39
SPEAKER_00: Unfortunately, that's often what happens.
4:42
SPEAKER_00: I understand it up to the point.
4:45
SPEAKER_00: Corporates are in charity institutions.
4:47
SPEAKER_00: They have real financial targets and commitments to meet them.
4:52
SPEAKER_00: No matter what, this is my honest advice and it works in practice.
4:58
SPEAKER_00: Stay patient from the first conversation through implementation and afterwards.
5:05
SPEAKER_00: Be present when delivery gets difficult,
5:10
SPEAKER_00: not only when there is promising revenue tied to your involvement.
5:16
SPEAKER_00: Make sure you have control over resources and the capacity to deliver what you promised.
5:23
SPEAKER_00: And when it's a partner delivering your solution,
5:27
SPEAKER_00: not you and something goes wrong because it does.
5:31
SPEAKER_00: Make sure you are there for your customer 24-7.
5:35
SPEAKER_00: So you can't hand the problem to others and you can't keep selling.
5:41
SPEAKER_00: While telling the customer that delivery or support functions will sort it out.
5:47
SPEAKER_00: That attitude never builds trust.
5:51
SPEAKER_00: I say this knowing sales leaders are smart and experienced.
5:56
SPEAKER_00: Often they are pushed into a corner.
5:59
SPEAKER_00: Their own leadership says one thing.
6:02
SPEAKER_00: The customer says another.
6:04
SPEAKER_00: They fill that gap.
6:06
SPEAKER_00: They understand the reality.
6:08
SPEAKER_00: But that doesn't mean their leadership does.
6:11
SPEAKER_00: That's my point.
6:13
SPEAKER_00: Customers need one clear point of contact when they need support and they need
6:18
SPEAKER_00: to feel that the whole organization stands behind the people they deal with.
6:24
SPEAKER_00: They need simplicity in their interactions with vendors and they shouldn't have
6:29
SPEAKER_00: to remember a long list of names and titles across every company they do business
6:35
SPEAKER_00: with.
6:36
SPEAKER_00: So that is the essence of building a long-term relationship,
6:41
SPEAKER_00: being present in the good moments and the difficult ones.
6:45
SPEAKER_00: And I've always tried to remain available to my customers whether an issue appeared
6:53
SPEAKER_00: during the opportunity,
6:55
SPEAKER_00: delivery,
6:57
SPEAKER_00: or later support, whatever my job title, account director or sales director.
7:04
SPEAKER_01: I understand that you've moved into more advisory work more recently.
7:08
SPEAKER_01: Can you say a little bit about that and how these experiences led you to there
7:12
SPEAKER_01: and what you're doing now?
7:15
SPEAKER_00: Advisory work has always been part of how I work with customers.
7:21
SPEAKER_00: Staying close to their reality,
7:23
SPEAKER_00: asking difficult questions early,
7:26
SPEAKER_00: and helping them make decisions they can stand behind.
7:31
SPEAKER_00: That's how business value is created.
7:35
SPEAKER_00: So for the last 17 months,
7:37
SPEAKER_00: I've been working independently doing advisory and due diligence work.
7:44
SPEAKER_00: It wasn't planned to move away from corporate life.
7:48
SPEAKER_00: The opportunity came to me and I took it because it made sense at that time.
7:56
SPEAKER_00: For me, this is an add-on to what I built
8:00
SPEAKER_00: over many years serving customers inside large organizations.
8:05
SPEAKER_00: It doesn't mean I have abandoned the account leadership path,
8:10
SPEAKER_00: but I can't ignore the value this work is adding.
8:14
SPEAKER_00: It's deepening my ability to read and assess risk in a way that keeps me close to
8:23
SPEAKER_00: the same problems I've always cared about.
8:27
SPEAKER_00: In practice,
8:28
SPEAKER_00: that means I help leadership teams,
8:31
SPEAKER_00: investors,
8:32
SPEAKER_00: private equity sponsors work out whether
8:36
SPEAKER_00: a disruptive technology like AI is commercially real,
8:41
SPEAKER_00: not just technically impressive.
8:44
SPEAKER_01: I've seen that you write about this work we met on Substack
8:47
SPEAKER_01: and I've seen your writing there.
8:48
SPEAKER_01: Can you say a little bit about that?
8:51
SPEAKER_00: Yes, I write in my Substack, AI in Real Business.
8:55
SPEAKER_00: I've also co-authored a chapter in the C-Writes AI Book Collaboration.
9:02
SPEAKER_00: Thank you Karen for this.
9:04
SPEAKER_00: It's about what AI readiness requires in daily business,
9:08
SPEAKER_00: practical governance,
9:10
SPEAKER_00: shared risk,
9:11
SPEAKER_00: and honest judgment whether you lead a 10-person team or a 10,000-person enterprise.
9:19
SPEAKER_01: I've seen a preview of that chapter and I have to say I'm really looking forward
9:23
SPEAKER_01: to sharing it with the world.
9:24
SPEAKER_01: I think it's awesome.
9:24
SPEAKER_01: So thank you for participating in our book collaboration.
9:28
SPEAKER_01: I'm really happy that you joined up.
9:30
SPEAKER_01: Tell our audience a little bit about your level of experience with AI
9:34
SPEAKER_01: and machine learning and analytics,
9:36
SPEAKER_01: whether you've used it professionally or personally,
9:39
SPEAKER_01: or if you've ever studied the technology.
9:42
SPEAKER_00: My experience sits mainly on the adoption.
9:47
SPEAKER_00: Commercial readiness and governance side rather than on building AI models myself.
9:53
SPEAKER_00: Professionally,
9:54
SPEAKER_00: I have worked with machine learning,
9:57
SPEAKER_00: analytics capabilities, and later AI within large telecom transformation programs.
10:04
SPEAKER_00: This included the network inventory,
10:07
SPEAKER_00: service assurance,
10:08
SPEAKER_00: automation,
10:09
SPEAKER_00: orchestration,
10:11
SPEAKER_00: 5G network slicing,
10:13
SPEAKER_00: Software Defined Networks and more recently Autonomous Network
10:17
SPEAKER_00: and authentic AI use cases.
10:20
SPEAKER_00: My role was to explain how those capabilities could solve a business problem
10:27
SPEAKER_00: that mattered to the customer.
10:29
SPEAKER_00: When that connection is clear,
10:32
SPEAKER_00: a telecom decision maker can evaluate,
10:35
SPEAKER_00: approve, fund and ultimately operate the solution.
10:41
SPEAKER_00: A technology can look impressive in features and promising in a demonstration
10:47
SPEAKER_00: or pilot, but it can still fail to become a sustainable business program.
10:53
SPEAKER_00: And then you don't get the executive yes to fund and scale it.
10:59
SPEAKER_01: Can you give an example of one of the programs that you worked on
11:02
SPEAKER_01: that would help our audience understand this better?
11:06
SPEAKER_00: One example was the Software Architecture Blueprint for Open Network APIs.
11:11
SPEAKER_00: First, we turned it into a minimum viable product in one market.
11:16
SPEAKER_00: Then we planned to scale it into a multi-market program.
11:21
SPEAKER_00: The objective was a multi-ventor collaboration that would help
11:26
SPEAKER_00: the customer expose network capabilities programmatically to partners,
11:32
SPEAKER_00: developers, enterprise customers, and internal teams.
11:37
SPEAKER_00: What I found interesting was the alignment of all stakeholders from the customer
11:42
SPEAKER_00: and vendors involved to define technical and business objectives together
11:48
SPEAKER_00: until everyone was speaking one language,
11:52
SPEAKER_00: the language that translates a product into a business value
11:57
SPEAKER_00: because this is what sells.
11:59
SPEAKER_00: Another example was a 5G network slicing pilot in 2024.
12:06
SPEAKER_00: The hard part wasn't the technology itself.
12:09
SPEAKER_00: It was building a commercial model that customer sales teams could actually sell.
12:16
SPEAKER_00: Network slicing is defined by technical parameters.
12:20
SPEAKER_00: Turning those into a sellable,
12:24
SPEAKER_00: chargeable service with pricing that could be audited and verified was far
12:32
SPEAKER_00: from straightforward.
12:34
SPEAKER_01: So what was it about that experience with your 5G project that made it feel valuable for you?
12:41
SPEAKER_00: Like I said,
12:42
SPEAKER_00: I'm not an engineer who builds the mode,
12:45
SPEAKER_00: but because I have studied the technology for many years,
12:48
SPEAKER_00: I understand the challenges,
12:51
SPEAKER_00: particularly in high technical organizations,
12:55
SPEAKER_00: and I understand how difficult it is to align roadmaps and how difficult it is
13:01
SPEAKER_00: the communication between the technical teams and the business teams inside
13:06
SPEAKER_00: the customers' organizations.
13:09
SPEAKER_00: And the amazing thing is how you can agree on a use case or use cases that will be evaluated.
13:18
SPEAKER_00: They will support the business case that needs to be built.
13:23
SPEAKER_00: And this is how an organization can become ready to adopt the new technology because
13:31
SPEAKER_00: the technology,
13:33
SPEAKER_00: no matter how impressive it may be,
13:36
SPEAKER_00: at the end,
13:37
SPEAKER_00: you need to be able to sell it,
13:40
SPEAKER_00: to transfer the business value that comes out of this technology to the customers
13:46
SPEAKER_00: that are in this big setup.
13:49
SPEAKER_01: So you understand the technology well enough to explain and help people connect over
13:54
SPEAKER_01: whether it solves an actual problem that your customers care about.
13:56
SPEAKER_01: That certainly sounds extremely valuable.
13:59
SPEAKER_01: Can you tell us a little bit about how maybe you use AI tools yourself
14:04
SPEAKER_01: in your personal life or in your work life, either one?
14:07
SPEAKER_00: Yes.
14:09
SPEAKER_00: Grammarly, I use it in my writing.
14:13
SPEAKER_00: Perlexity,
14:14
SPEAKER_00: for source-based research,
14:17
SPEAKER_00: I use ChataGPT and now moving to Cloud to explore and challenge ideas.
14:24
SPEAKER_00: I have also experimented with other AI for meeting capture and summaries
14:31
SPEAKER_00: and I was back in time an early co-pilot user.
14:35
SPEAKER_00: So I experience AI from two directions,
14:39
SPEAKER_00: in operational environments where it has to pass into production and in my own work
14:45
SPEAKER_00: where I see both value and limitations.
14:49
SPEAKER_01: Can you share an example of something that you learned from using AI tools yourself
14:54
SPEAKER_01: that your enterprise work did not teach you?
14:59
SPEAKER_00: It showed me how quickly AI becomes part of everyday work and how easily
15:07
SPEAKER_00: A useful tool can appear more reliable than it is.
15:13
SPEAKER_00: In an enterprise program,
15:15
SPEAKER_00: there are formal reviews,
15:17
SPEAKER_00: governance processes, and several people testing assumptions.
15:22
SPEAKER_00: As an individual, those guardrails do not exist.
15:26
SPEAKER_00: So it's up to you to decide how much trust you delegate to the tool.
15:31
SPEAKER_00: So I use AI where it adds most value.
15:37
SPEAKER_00: To research, organize, or even challenge my thinking.
15:42
SPEAKER_00: But it doesn't remove my responsibility to think for myself,
15:46
SPEAKER_00: decide what to say, and verify important points before I communicate them.
15:53
SPEAKER_01: Sounds good.
15:54
SPEAKER_01: Can you share with our audience a specific story on how you've used one of the tools
16:01
SPEAKER_01: that it has?
16:02
SPEAKER_00: Let me use an example from network operations because that's where I saw AI address
16:08
SPEAKER_00: a very concrete business problem.
16:11
SPEAKER_00: The use case combined a digital representation of the transport network with
16:18
SPEAKER_00: an AI model designed to identify fiber degradation before it became
16:24
SPEAKER_00: a visible service fault.
16:26
SPEAKER_00: The traditional approach is reactive.
16:29
SPEAKER_00: A fault occurs,
16:31
SPEAKER_00: alarms appear,
16:33
SPEAKER_00: customers are affected,
16:35
SPEAKER_00: the operator escalates the case until the root cause is identified and
16:40
SPEAKER_00: the permanent resolution or workaround is achieved.
16:45
SPEAKER_00: The AI-enabled approach identifies the pattern earlier So maintenance can be planned
16:54
SPEAKER_00: before degradation becomes a major outage.
16:57
SPEAKER_00: That made the business value much easier to explain.
17:02
SPEAKER_00: Fewer emergency interventions,
17:05
SPEAKER_00: lower risk of service disruption,
17:08
SPEAKER_00: and potentially fewer penalties and customer complaints.
17:12
SPEAKER_01: I'll point out for our audience that this was non-generative AI.
17:15
SPEAKER_01: This was classic machine learning type of AI,
17:18
SPEAKER_01: it sounds like,
17:19
SPEAKER_01: that made it convincing for the business case that you were
17:24
SPEAKER_00: working on.
17:25
SPEAKER_01: What worked particularly well or what made the business case convincing?
17:31
SPEAKER_00: It connected the technology features to business outcomes against the KPIs
17:36
SPEAKER_00: that operations teams have already used.
17:40
SPEAKER_00: They could see whether and how the technology contributed to their daily business.
17:47
SPEAKER_00: The conversation stopped the pink.
17:50
SPEAKER_00: Here is an interesting diction model.
17:53
SPEAKER_00: It became,
17:54
SPEAKER_00: here is a way to reduce operational costs or prevent an outage
18:00
SPEAKER_00: before customers experience it or other KPIs that customers care about.
18:07
SPEAKER_00: So proactive maintenance stopped looking like additional expenditure and become
18:12
SPEAKER_00: to look like measurable cost and risk reduction.
18:17
SPEAKER_00: And this is what worked very well.
18:20
SPEAKER_01: Okay, that sounds good.
18:22
SPEAKER_01: Can you share an example of something that maybe didn't work so well?
18:27
SPEAKER_00: Yeah.
18:27
SPEAKER_00: In my view, the limitation was the flexibility of the tool.
18:33
SPEAKER_00: It's easy to demonstrate a standard use case that's pre-built.
18:38
SPEAKER_00: For example,
18:39
SPEAKER_00: alarm correlation,
18:41
SPEAKER_00: outage prediction, service impact analysis, and closed-loop remediation.
18:46
SPEAKER_00: But every operator has different legacy systems,
18:50
SPEAKER_00: data, workflows, KPIs, and operational practices.
18:56
SPEAKER_00: It's much harder to integrate an out-of-the-box solution into the customer's actual environment.
19:03
SPEAKER_00: That made me cautious about prepackaged AI value packs and ready connectors.
19:10
SPEAKER_00: In the end, customizations were needed.
19:14
SPEAKER_00: Connectors were not in the library.
19:16
SPEAKER_00: And when a design to scale across diverse environments runs into that reality,
19:23
SPEAKER_00: then costs can explode and timelines slip
19:28
SPEAKER_00: in ways customers simply don't appreciate them.
19:33
SPEAKER_01: So in that situation,
19:36
SPEAKER_01: did you see that buyers would become more hesitant to accept
19:41
SPEAKER_01: the solution being proposed because they couldn't see that connection,
19:46
SPEAKER_01: couldn't see it fitting into their operational reality?
19:50
SPEAKER_00: I believe that a solution in the slides like it's a piece of cake
19:55
SPEAKER_00: that everything can work fine out of the box.
19:58
SPEAKER_00: We have a library of Nectors to integrate with the underlying network,
20:02
SPEAKER_00: but the reality is a little bit different because underlying network is built
20:06
SPEAKER_00: over years.
20:07
SPEAKER_00: So many legacy systems, the customers persuade their assets.
20:11
SPEAKER_00: They can be there for 35 years.
20:14
SPEAKER_00: So all this out of the box concept is not like this.
20:18
SPEAKER_00: And then When you go to the first meeting, it's hallelujah.
20:23
SPEAKER_00: When you go to the second meeting, then oops.
20:26
SPEAKER_00: Then the third meeting, oh, this is the cost.
20:28
SPEAKER_00: Then the fourth meeting, okay, let's wait to see.
20:31
SPEAKER_00: This is not a priority.
20:33
SPEAKER_00: And then bye, bye, bye.
20:35
SPEAKER_00: Let's review the budget.
20:37
SPEAKER_00: Let's see this deal after a year.
20:40
SPEAKER_00: So that's the reality.
20:42
SPEAKER_01: Well, thank you for sharing those examples, both what went well and what didn't go well.
20:46
SPEAKER_01: I think that's useful to hear.
20:48
SPEAKER_01: So you've made some good use of AI tools in your work and in your personal life.
20:53
SPEAKER_01: Are there any times when you have avoided using AI-based tools for some things?
20:58
SPEAKER_01: And can you share an example of when you would avoid it and why you would choose not
21:02
SPEAKER_01: to use AI in those situations?
21:05
SPEAKER_00: Yes, I will speak from personal experience.
21:08
SPEAKER_00: As I said before,
21:09
SPEAKER_00: I use AI mainly for efficiency,
21:12
SPEAKER_00: research,
21:13
SPEAKER_00: long document summaries, comparisons, and examining a topic from different angles.
21:20
SPEAKER_00: But I deliberately don't Let it make judgments or decisions on my behalf.
21:27
SPEAKER_00: Honestly, at the beginning, because I was a new user, I said, wow, fantastic.
21:33
SPEAKER_00: This is amazing.
21:34
SPEAKER_00: It's a tool that is AI.
21:37
SPEAKER_00: I am sure everything is right.
21:39
SPEAKER_00: No, it's not.
21:41
SPEAKER_00: A convincing AI answer is not necessarily a reliable answer.
21:47
SPEAKER_00: So, lesson learned.
21:49
SPEAKER_00: From that time onwards,
21:51
SPEAKER_00: I started to validate always important points,
21:55
SPEAKER_00: especially when I'm preparing something to publish or making
21:58
SPEAKER_00: a professional decision.
22:01
SPEAKER_00: At that point, I turn to my experience, judgment and instinct.
22:06
SPEAKER_00: Those are the qualities that have helped me throughout almost 30 years in my profession.
22:13
SPEAKER_00: So AI can organize information, compare few points or challenge my first assumption.
22:20
SPEAKER_00: But it's in my nature and I decided not to go against it, to always validate my answers.
22:29
SPEAKER_00: And that's exactly how I have come to understand its limits.
22:33
SPEAKER_00: It's always efficient because when you research you need a lot of information.
22:39
SPEAKER_00: When I did my MBA I still remember how much time I was searching in order to find the material.
22:47
SPEAKER_00: Things have improved them.
22:49
SPEAKER_00: AI offers you these efficiencies.
22:53
SPEAKER_00: On the other hand, the judgment is yours.
22:56
SPEAKER_00: What to write, how to write, to validate what to write.
23:01
SPEAKER_00: It's you, the author.
23:02
SPEAKER_00: It's your signature at the end, not the AI tool.
23:06
SPEAKER_00: If something goes wrong, you will be responsible, not the AI tool.
23:10
SPEAKER_01: Yeah, absolutely.
23:11
SPEAKER_01: I think it's great that you're calling out the importance of human judgment there.
23:16
SPEAKER_01: What's the main thing that you feel that your human judgment adds that
23:20
SPEAKER_01: an AI tool cannot provide?
23:24
SPEAKER_00: When dealing with people, judgment isn't based only on what they say.
23:30
SPEAKER_00: You also notice hesitation, body language, what isn't being said.
23:37
SPEAKER_00: And also the history behind the relationship that matters.
23:41
SPEAKER_00: So if you understand the organizational context,
23:45
SPEAKER_00: customer priorities,
23:46
SPEAKER_00: mindset,
23:47
SPEAKER_00: politics around the decision,
23:50
SPEAKER_00: risks that people may not say openly,
23:54
SPEAKER_00: and who will carry the consequences if something goes wrong.
23:58
SPEAKER_00: So AI can help you prepare for sure.
24:02
SPEAKER_00: It can surface information or identify a detail you may have missed.
24:09
SPEAKER_00: But it still has to lead to a final judgment about what to do.
24:15
SPEAKER_00: And that judgment is your responsibility because you carry the consequences.
24:20
SPEAKER_00: I'm also careful not to outsource the original idea.
24:26
SPEAKER_00: AI can challenge an idea, organize it, or show another angle.
24:31
SPEAKER_00: But the original thinking needs to remain mine because this is my competitive edge.
24:39
SPEAKER_00: This is my experience.
24:40
SPEAKER_00: This is my effort.
24:41
SPEAKER_00: This is my knowledge.
24:43
SPEAKER_00: If I start giving up my competitive edge, then what is my value?
24:51
SPEAKER_01: Yep.
24:51
SPEAKER_01: That's a good summary of where your original thinking and your judgment are
24:56
SPEAKER_01: and needing to keep As we say here on this show,
24:59
SPEAKER_01: you're wetware, you're human brain engaged and active.
25:03
SPEAKER_01: So I want to switch for a minute to talking about where these AI
25:08
SPEAKER_01: and machine learning systems get the data and the content,
25:11
SPEAKER_01: the works that they train on.
25:12
SPEAKER_01: A lot of times they will use data that users put into online systems,
25:16
SPEAKER_01: where they published online,
25:18
SPEAKER_01: or they don't always tell us when we sign up for our website how they're going
25:23
SPEAKER_01: to use our data.
25:24
SPEAKER_01: I'm wondering how you feel about companies that use different types of data
25:30
SPEAKER_01: and content for training their tools.
25:32
SPEAKER_01: And specifically,
25:33
SPEAKER_01: I'm wondering about your thoughts on something they call the three C's from CIPRI,
25:38
SPEAKER_01: consent, credit,
25:39
SPEAKER_01: and compensation,
25:39
SPEAKER_01: the idea that creative people have these rights around the way
25:44
SPEAKER_01: that their data is used, whether it's used and how it's used.
25:49
SPEAKER_00: And I certainly support the 3 C's Consent,
25:53
SPEAKER_00: Credit and Compensation as a sequence of rules for ethical AI.
25:59
SPEAKER_00: Consent is the precondition.
26:01
SPEAKER_00: It creates a record that can be checked later.
26:06
SPEAKER_00: What was used for which purpose,
26:09
SPEAKER_00: under which conditions, and whether those conditions were respected.
26:15
SPEAKER_00: Credit or Compensation should never substitute for permission or justify unethical AI use.
26:25
SPEAKER_00: If a person,
26:27
SPEAKER_00: creator or organization hasn't agreed to the use of their data or work,
26:34
SPEAKER_00: there is no excuse that can justify violating that rule.
26:40
SPEAKER_00: And no retrospective credit or compensation changes that.
26:46
SPEAKER_00: For me, this applies to every AI interaction in three ways.
26:52
SPEAKER_00: First, personal data.
26:55
SPEAKER_00: Meaningful permission must be the starting point.
26:59
SPEAKER_00: People shouldn't discover later that information supplied
27:04
SPEAKER_00: for one purpose has been repurposed to train an AI system.
27:09
SPEAKER_00: Second, business information.
27:12
SPEAKER_00: Professionals handle confidential customer data.
27:16
SPEAKER_00: Commercial plans,
27:18
SPEAKER_00: pricing,
27:19
SPEAKER_00: contracts,
27:20
SPEAKER_00: network information, product roadmaps, security details, and work in progress.
27:28
SPEAKER_00: That information cannot be entered into an AI tool
27:33
SPEAKER_00: unless its use has been explicitly approved and the appropriate safeguards are
27:40
SPEAKER_00: in place.
27:42
SPEAKER_00: If information was shared for a specific business purpose,
27:46
SPEAKER_00: It should not be purposed through an AI system unless that use is permitted by
27:53
SPEAKER_00: the commitments governing the relationship with the customer, employer, or partner.
27:59
SPEAKER_00: So the organization should be able to demonstrate what was used for which purpose
28:05
SPEAKER_00: and whether the agreed conditions were respected.
28:09
SPEAKER_00: And finally, creative work.
28:12
SPEAKER_00: For writers,
28:13
SPEAKER_00: musicians,
28:14
SPEAKER_00: artists, photographers, actors, and software developers, consent has two dimensions.
28:22
SPEAKER_00: An ethical one,
28:24
SPEAKER_00: their work should not be taken and used commercially without their knowledge
28:29
SPEAKER_00: and agreement.
28:30
SPEAKER_00: And an economic one,
28:32
SPEAKER_00: if commercial value is created from their authorized work,
28:38
SPEAKER_00: they should receive appropriate credit and compensation.
28:42
SPEAKER_00: Publicly accessible doesn't mean freely available for every commercial purpose.
28:52
SPEAKER_01: Yeah, absolutely.
28:53
SPEAKER_01: And I think you maybe wrote a piece about this.
28:57
SPEAKER_01: Who owns what your AI learns about you, right?
29:01
SPEAKER_00: Yes, absolutely, yes.
29:03
SPEAKER_00: Yes, this is the title.
29:05
SPEAKER_00: All my articles are on Substack.
29:08
SPEAKER_00: They are free for anyone who is interested.
29:11
SPEAKER_00: And yes, I have written about it.
29:14
SPEAKER_00: It's all about how that is your IP and a tool.
29:18
SPEAKER_00: Cannot be used without your consent from anyone else versus others than the ones
29:25
SPEAKER_00: that you have permitted and you have allowed.
29:28
SPEAKER_00: You know, this is more or less the content and a lot of information.
29:32
SPEAKER_00: I invite the audience to read it.
29:35
SPEAKER_00: It's an interesting article.
29:37
SPEAKER_01: All right,
29:38
SPEAKER_01: we'll put the link to that in the interview and hopefully people will go and look
29:41
SPEAKER_01: that up.
29:42
SPEAKER_01: So one thing I hear a lot in these interviews,
29:44
SPEAKER_01: most people seem to agree that consent and credit and compensation are right
29:50
SPEAKER_01: and they make sense and that it's in the interest of the economy as a whole to do
29:54
SPEAKER_01: that.
29:55
SPEAKER_01: On the other hand,
29:56
SPEAKER_01: I hear some people say, yeah,
29:58
SPEAKER_01: this is nice in theory,
30:00
SPEAKER_01: but they're not so sure it can work in practice at the scale
30:04
SPEAKER_01: which AI models are trained.
30:06
SPEAKER_01: And I'm wondering what your thoughts are about that.
30:09
SPEAKER_00: I believe that the principle, Karen, doesn't change with scale, for example.
30:17
SPEAKER_00: We know that you still need consent,
30:20
SPEAKER_00: credit and compensation, whether it's one document or a billion.
30:24
SPEAKER_00: So what the change is, is how you actually deliver on that at scale.
30:30
SPEAKER_00: So the system and processes needed to track,
30:34
SPEAKER_00: verify and then force it when the numbers get at large.
30:38
SPEAKER_00: In my view,
30:39
SPEAKER_00: transparent licensing or collective licensing
30:43
SPEAKER_00: and consent management systems may offer a workable path.
30:49
SPEAKER_00: They could make clear what information may be used for what purpose,
30:55
SPEAKER_00: under what conditions, and how credit and compensation apply in
31:00
SPEAKER_01: each case.
31:02
SPEAKER_01: So what do you feel is the most important thing about this way
31:06
SPEAKER_01: of approaching consent, credit, and compensation?
31:09
SPEAKER_01: What's the most essential point that you'd like to call out for people?
31:14
SPEAKER_00: I believe that people need to understand Okay.
31:18
SPEAKER_00: For me, this is very important because a contract uses legal language
31:22
SPEAKER_00: that most people can't fully understand the consequences of unless they work in
31:27
SPEAKER_00: that field.
31:30
SPEAKER_00: In my view,
31:31
SPEAKER_00: and in principle,
31:32
SPEAKER_00: the fixes sound simple,
31:34
SPEAKER_00: plain language,
31:35
SPEAKER_00: clear rules on training,
31:37
SPEAKER_00: retention,
31:38
SPEAKER_00: and deletion,
31:40
SPEAKER_00: evidence that provide their owners its commitment,
31:44
SPEAKER_00: but the simplicity is exactly what most providers, I think, avoid.
31:51
SPEAKER_00: Because ambiguity gives room to expand what the providers do with your data later
32:00
SPEAKER_00: with less legal exposure.
32:02
SPEAKER_00: So clarity, that for me is the key word, limits them.
32:07
SPEAKER_00: And that is why it doesn't happen on its own.
32:11
SPEAKER_00: So ethical use isn't about whether terms existed somewhere.
32:17
SPEAKER_00: It's about whether the person or organization could understand them and make
32:24
SPEAKER_00: a real choice and verify that their data or work was used only in
32:30
SPEAKER_00: the way they agreed.
32:32
SPEAKER_00: And that sounds obvious until you realize that most people have no real alternative
32:38
SPEAKER_00: to click away or no room to negotiate the terms.
32:43
SPEAKER_00: And no practical way to check afterwards whether the company kept its work.
32:48
SPEAKER_00: So understanding a contract and having power over it are two different things.
32:55
SPEAKER_00: And this isn't only a personal data problem.
32:59
SPEAKER_00: Confidential commercial information,
33:01
SPEAKER_00: data secrets,
33:03
SPEAKER_00: business plans,
33:04
SPEAKER_00: other proprietary information needs exactly the same protection even
33:10
SPEAKER_00: when no personal data is involved at all.
33:13
SPEAKER_00: So for me, clarity is important and a mechanism in order for people to be able to use it.
33:22
SPEAKER_01: I hear from a lot of people too that they feel a sense of being powerless,
33:28
SPEAKER_01: that everything is so unequal in terms of the power that
33:32
SPEAKER_01: the big tech companies have versus the power that they feel they have.
33:36
SPEAKER_01: I personally think that we all have some agency and that we should be making efforts
33:40
SPEAKER_01: to exercise it,
33:41
SPEAKER_01: but I understand that feeling of powerlessness and that the negotiation power
33:46
SPEAKER_01: that they have is very unequal.
33:48
SPEAKER_01: And I'm wondering what your thoughts are about that.
33:52
SPEAKER_00: And again,
33:52
SPEAKER_00: a major AI provider may offer the standard terms that people
33:58
SPEAKER_00: and businesses can accept or reject, but it is true.
34:04
SPEAKER_00: They cannot meaningfully shape And as I said before,
34:08
SPEAKER_00: this is because a contract uses a language that most of us,
34:13
SPEAKER_00: unless we are in the field, we cannot understand.
34:16
SPEAKER_00: So, as I said before,
34:19
SPEAKER_00: and this is important,
34:21
SPEAKER_00: use a plain language,
34:22
SPEAKER_00: clear rules on training,
34:25
SPEAKER_00: and evidence that the provider that you trust honors its commitments.
34:32
SPEAKER_00: And I believe that the key word that I said before is clarity.
34:38
SPEAKER_00: I need to know what I sign for.
34:41
SPEAKER_00: And it doesn't really matter if I am an individual,
34:45
SPEAKER_00: if it is an organization, protection should be there.
34:49
SPEAKER_00: And if we don't push for this,
34:51
SPEAKER_00: don't expect that the providers will really go out from their comfort zone
34:56
SPEAKER_00: because this is convenience, you know.
34:59
SPEAKER_00: So if we don't push, then nothing will change and nothing will happen.
35:08
SPEAKER_01: I think your point on clarity is worth highlighting.
35:11
SPEAKER_01: Some people refer to it as transparency,
35:13
SPEAKER_01: but I think clarity is a little different and I think it's important to highlight
35:16
SPEAKER_01: that as well.
35:17
SPEAKER_01: The providers of the AI tools that you've used personally,
35:21
SPEAKER_01: ChatGPT and perplexity and such,
35:23
SPEAKER_01: do you feel like they've been clear and transparent with you about where
35:29
SPEAKER_01: the data came from that's used in training those models and whether or not
35:32
SPEAKER_01: the original creators consented to it being used?
35:37
SPEAKER_00: I will be extremely careful now how to express myself.
35:42
SPEAKER_00: I don't think data provenance has been made sufficiently understandable across the industry.
35:49
SPEAKER_00: I said this before.
35:51
SPEAKER_00: We hear phrases such as publicly available,
35:56
SPEAKER_00: licensed, responsibly sourced, and they sound reassuring.
36:01
SPEAKER_00: And I'm not claiming providers are secretly misusing data.
36:07
SPEAKER_00: I don't have evidence of that and I think it's important to say so clearly.
36:14
SPEAKER_00: What concerns me is something different.
36:17
SPEAKER_00: As things stand, I would have almost no way to find out either way.
36:24
SPEAKER_00: The only moments a user gets any signal at all are accidental.
36:31
SPEAKER_00: Seeing a piece of text that resembles something you wrote or having
36:36
SPEAKER_00: an account compromised and stored without even being told the root cause.
36:43
SPEAKER_00: A reliable system would tell you proactively, not leave you to stumble into the truth by chance.
36:53
SPEAKER_00: There is a lot of public suspicion around this.
36:57
SPEAKER_00: People using the technology are genuinely uneasy.
37:03
SPEAKER_00: But that suspicion isn't backed by real evidence.
37:08
SPEAKER_00: And that's exactly the problem.
37:10
SPEAKER_00: Without evidence, there is no real pressure on providers to change anything.
37:17
SPEAKER_00: So we end up in a strange position
37:20
SPEAKER_00: where something that's genuinely unresolved gets quietly treated as settled simply
37:28
SPEAKER_00: because no one can point to proof.
37:32
SPEAKER_00: That's not the same as it being fine.
37:36
SPEAKER_00: It just means no one has been forced to check.
37:41
SPEAKER_00: So for me,
37:42
SPEAKER_00: the answer is perhaps a practical transparency standard,
37:48
SPEAKER_00: something people can understand,
37:51
SPEAKER_00: compare and use before deciding to use an AI product.
37:56
SPEAKER_00: So not another long terms and conditions document,
38:00
SPEAKER_00: but a short product specific statement answering a handful of plain questions.
38:08
SPEAKER_00: What data trained the model?
38:11
SPEAKER_00: How was it obtained?
38:13
SPEAKER_00: Is my content used for training, evaluation or improvement?
38:19
SPEAKER_00: How long is it kept and who can access it?
38:24
SPEAKER_00: Can I opt out or change my mind later?
38:29
SPEAKER_00: And who is accountable if the provider doesn't keep its word?
38:35
SPEAKER_00: European requirements for general purpose AI models move in the right direction.
38:42
SPEAKER_00: But published summaries are only the beginning of transparency.
38:47
SPEAKER_00: The real test is simpler.
38:49
SPEAKER_00: Can an ordinary person,
38:52
SPEAKER_00: like me,
38:53
SPEAKER_00: or even business understand it,
38:57
SPEAKER_00: make an informed choice, and later verify that the provider kept its promise?
39:05
SPEAKER_01: I think you also wrote an article about this, if I remember correctly.
39:08
SPEAKER_00: Yes, I have written an article.
39:10
SPEAKER_01: The one that I'm thinking of is called, The AI Stack is Being Chosen For You.
39:16
SPEAKER_00: Yes, absolutely.
39:18
SPEAKER_00: So Karen, really, I love you.
39:21
SPEAKER_00: You are one of my great readers.
39:26
SPEAKER_01: I enjoy your work, so I'm happy to talk about it and bring it to the attention of other people.
39:31
SPEAKER_00: It's all good.
39:31
SPEAKER_00: Thank you very much for this.
39:33
SPEAKER_00: Anyhow, what I'm saying is summaries of what I have written in my articles
39:39
SPEAKER_00: and how I have reflected on the idea.
39:43
SPEAKER_00: This didn't come to me for the interview.
39:46
SPEAKER_00: It is because I'm writing about these things.
39:48
SPEAKER_00: I'm thinking a lot.
39:50
SPEAKER_00: And as many friends and colleagues used to tell me, oh my God, you are full of ideas.
39:57
SPEAKER_00: So this transparency standard, I said, why not?
40:01
SPEAKER_00: I mean, why things to be so complicated in life?
40:06
SPEAKER_00: We can for once make something that is easy to do.
40:11
SPEAKER_01: Anyhow, yes.
40:12
SPEAKER_01: Yes.
40:13
SPEAKER_01: Yes.
40:13
SPEAKER_01: Sometimes simple is not easy.
40:15
SPEAKER_00: Yes, I agree.
40:20
SPEAKER_01: Yes, but sometimes it is, and we make it harder than it needs to be sometimes.
40:23
SPEAKER_01: Yes.
40:24
SPEAKER_00: It's an excuse sometimes that we say it's not easy,
40:27
SPEAKER_00: but it's an excuse not to make the effort or because
40:31
SPEAKER_00: the effort will create inconvenience or will create risk.
40:34
SPEAKER_00: Okay, let's not start this conversation now.
40:38
SPEAKER_01: I'm sure we could talk for a long time about that, but let's go on.
40:41
SPEAKER_01: So as consumers and members of the public,
40:44
SPEAKER_01: our personal data gets used by AI-based tools and systems.
40:49
SPEAKER_01: Do you know of any specific cases that you could share?
40:52
SPEAKER_01: Okay.
40:54
SPEAKER_00: Here I speak from my own position as a creator.
40:58
SPEAKER_00: So I can't know with certain team whether my professional content has been used
41:05
SPEAKER_00: to train an AI system.
41:08
SPEAKER_00: I publish on LinkedIn and Substack because I want people to read my work.
41:14
SPEAKER_00: But publishing for readers isn't the same as agreeing
41:18
SPEAKER_00: that it becomes training material for a commercial AI model.
41:23
SPEAKER_00: I understand that AI learns from large bodies of human knowledge.
41:29
SPEAKER_00: My concern is the gap between being publicly visible and being treat it as available
41:37
SPEAKER_00: for a different commercial purpose.
41:41
SPEAKER_00: That distinction should be clear to the person who created the content.
41:46
SPEAKER_00: I also distinguish public content from private material placed directly into
41:53
SPEAKER_00: an AI product, prompts, drafts, notes, documents, and feedback.
42:00
SPEAKER_00: That may include original thinking or work in progress.
42:06
SPEAKER_00: It deserves particular care.
42:09
SPEAKER_00: As I said in the question four previously, I'm careful not to outsource the original idea.
42:18
SPEAKER_00: And that concern extends here.
42:21
SPEAKER_00: Original thinking,
42:22
SPEAKER_00: whether it sits in a private draft or in a published piece,
42:27
SPEAKER_00: deserves protection from being absorbed as training material
42:31
SPEAKER_00: without informed consent.
42:35
SPEAKER_00: Absolutely, yeah.
42:36
SPEAKER_01: Do you have any companies that you deliberately gave your information to
42:41
SPEAKER_01: that made you aware that they would use your data for training AI
42:44
SPEAKER_01: or machine learning?
42:46
SPEAKER_01: Or did you ever get surprised by it?
42:48
SPEAKER_00: Tough questions, Katherine.
42:50
SPEAKER_01: Yes.
42:53
SPEAKER_00: Not in a way that I could clearly understand and make an informed decision about.
43:00
SPEAKER_00: I have accepted terms for AI products,
43:03
SPEAKER_00: as most people have,
43:05
SPEAKER_00: but I cannot recall being presented with a short,
43:09
SPEAKER_00: clear explanation saying exactly whether my content could be used to train
43:15
SPEAKER_00: or improve an AI system.
43:17
SPEAKER_00: And that is the issue.
43:19
SPEAKER_00: The information may exist somewhere in the terms,
43:23
SPEAKER_00: but if an ordinary user cannot realistically find it,
43:28
SPEAKER_00: understand it and assess its consequences,
43:32
SPEAKER_00: then disclosure has not really done its job.
43:36
SPEAKER_00: And this is not unique to AI.
43:39
SPEAKER_00: I keep thinking about what actually happens when I want to sell shares or RSUs,
43:46
SPEAKER_00: the Restricted Stock Units.
43:49
SPEAKER_00: I acknowledge and sign a document,
43:51
SPEAKER_00: pages long,
43:53
SPEAKER_00: written in legal and technical language I don't fully understand and I don't have
43:58
SPEAKER_00: the time to invite or send everything to a lawyer and this is costly as well.
44:03
SPEAKER_00: The same is true of insurance policies where the small print can quietly undo
44:10
SPEAKER_00: the protection the headline promise.
44:13
SPEAKER_00: How many of us actually read and understand all of it?
44:17
SPEAKER_00: But with AI, there is a further consequence, and it's the one that concerns me most.
44:25
SPEAKER_00: AI firms aren't a one-time signature on a static transaction.
44:32
SPEAKER_00: They govern an ongoing relationship with your data that keeps expanding after you have signed.
44:40
SPEAKER_00: Into uses, products and training that didn't exist the day you agreed.
44:46
SPEAKER_00: And the real consequence of that is
44:49
SPEAKER_00: that your intellectual property stops being exclusively yours.
44:55
SPEAKER_00: Once your professional judgment,
44:57
SPEAKER_00: your writing,
44:59
SPEAKER_00: your way of solving a problem has been absorbed into a model,
45:03
SPEAKER_00: it can be reproduced and repeated indefinitely by anyone using that system.
45:11
SPEAKER_00: And the judgment that can be copied
45:13
SPEAKER_00: that easily stops behaving like something you own and starts behaving like
45:20
SPEAKER_00: a commodity.
45:22
SPEAKER_00: It's the erosion of what made your expertise valuable in the first place,
45:28
SPEAKER_00: the fact that it was yours.
45:31
SPEAKER_00: That's much bigger loss than a data privacy issue.
45:36
SPEAKER_00: Because even if this isn't a transfer of copyright,
45:40
SPEAKER_00: legally often remains mine on paper, the loss is practical and economic.
45:48
SPEAKER_00: The exclusivity and scarcity
45:51
SPEAKER_00: that made my judgment valuable disappear once it can be reproduced by
45:57
SPEAKER_00: the system indefinitely.
45:59
SPEAKER_00: Even if I retain formal ownership and that's the problem.
46:04
SPEAKER_01: Yeah, so I'm thinking about the different systems that we use.
46:09
SPEAKER_01: We both use LinkedIn,
46:09
SPEAKER_01: we both use Substack and on Substack there is an option that Substack gives us
46:14
SPEAKER_01: to try to protect our newsletters from being scraped by third-party AI bots.
46:22
SPEAKER_01: LinkedIn on the other hand has options built right in and we had to opt out to try
46:26
SPEAKER_01: to keep them from using our information themselves.
46:29
SPEAKER_01: And so there are two different scenarios.
46:31
SPEAKER_01: I'm wondering on Substack,
46:33
SPEAKER_01: have you configured your newsletter to try to block the third-party AI bots
46:38
SPEAKER_01: or do you have it open so that it's more accessible?
46:41
SPEAKER_00: On Substack, I have not left my newsletter open for third parties.
46:46
SPEAKER_00: I publish there for people to read my work and not as an open invitation
46:51
SPEAKER_00: for third-party AI systems to take out or use it in ways I have not agreed to.
46:58
SPEAKER_00: And that distinction matters to me.
47:01
SPEAKER_00: So, in short, real choice requires more than a setting buried somewhere on a platform, right?
47:09
SPEAKER_00: Because we discussed about Substack, but it's the same on LinkedIn.
47:15
SPEAKER_00: It's also there I publish professionally and I manage my settings myself.
47:22
SPEAKER_00: So I found the setting on my own initiative,
47:27
SPEAKER_00: not because I received a clear notice explaining the implication.
47:32
SPEAKER_00: And that is exactly what concerns me.
47:35
SPEAKER_00: An opt-out is not a meaningful choice if people have to discover that it exists.
47:42
SPEAKER_00: So a real choice requires more than a setting, as I said before, buried somewhere on a platform.
47:49
SPEAKER_00: The notice should be visible.
47:51
SPEAKER_00: And written,
47:53
SPEAKER_00: repeat again,
47:54
SPEAKER_00: in plain language,
47:55
SPEAKER_00: what information is involved,
47:57
SPEAKER_00: what the company intends to do with when that use begins,
48:02
SPEAKER_00: what happens if the user refuses,
48:05
SPEAKER_00: and whether that refusal applies only going forward.
48:10
SPEAKER_00: So I'm not against the platforms using data responsibly to improve useful services,
48:18
SPEAKER_00: but people should see the decision clearly before their data or content is used
48:25
SPEAKER_00: and not discover it afterwards.
48:27
SPEAKER_01: Yeah, absolutely.
48:29
SPEAKER_01: Has any company's use of your data ever created any specific issues for you?
48:34
SPEAKER_01: Maybe privacy, phishing, loss of income, or any examples like that?
48:40
SPEAKER_00: Fortunately,
48:41
SPEAKER_00: I haven't experienced a confirmed case of financial loss facing or proven misuse
48:47
SPEAKER_00: of my content.
48:50
SPEAKER_00: I did, however, experience an incident involving an AI platform.
48:55
SPEAKER_00: For approximately one day, I lost access to my account and to my own files.
49:02
SPEAKER_01: Wow.
49:03
SPEAKER_01: So what happened and how did the AI tool provider respond when it happened?
49:08
SPEAKER_00: The provider restored my access and assured me that nothing had been lost.
49:14
SPEAKER_00: I asked for the root cause but didn't receive one.
49:19
SPEAKER_00: I have no evidence that my content was used for training,
49:25
SPEAKER_00: copied or otherwise misused and I would never claim that it was.
49:32
SPEAKER_00: What I can share is the experience itself and what it left me assuming because
49:39
SPEAKER_00: the answer I got was never clear enough to rule anything in or out.
49:46
SPEAKER_00: So what affected me was this feeling of despair.
49:51
SPEAKER_00: My entire work, notes, thinking were inside that platform.
49:58
SPEAKER_00: And I didn't know for one day whether it was lost,
50:03
SPEAKER_00: recoverable, or had been accessed by someone else.
50:08
SPEAKER_00: It felt like something personal was gone.
50:12
SPEAKER_00: Along with the time and effort behind it.
50:15
SPEAKER_00: Something unique that even if I could reproduce it,
50:19
SPEAKER_00: it would never be quite the same.'
50:21
SPEAKER_01: That's a scary thing to have happen.
50:23
SPEAKER_01: So it sounds like their response did not make you feel like they could be trusted.
50:28
SPEAKER_01: What kind of response would you have wanted to have from them
50:31
SPEAKER_01: that would have made you feel like you could trust them going forward with your data
50:35
SPEAKER_01: and with the integrity of your account and access to it?
50:39
SPEAKER_00: I said before clarity.
50:41
SPEAKER_00: So I wanted a clear incident explanation.
50:44
SPEAKER_00: First, what happened.
50:47
SPEAKER_00: Second, what may have been exposed.
50:50
SPEAKER_00: Third, what the company checked and what it ruled out.
50:55
SPEAKER_00: And finally, what it changed to prevent the recurrence.
51:00
SPEAKER_00: So restoring access was necessary.
51:04
SPEAKER_00: But restoring access is not the same as restoring trust.
51:08
SPEAKER_01: And of all the things you just mentioned that you wanted to have clarity on,
51:11
SPEAKER_01: it doesn't sound like you got any of them.
51:14
SPEAKER_00: No, I got a lot of support, emails, but saying what?
51:21
SPEAKER_00: Nothing.
51:22
SPEAKER_01: Okay.
51:24
SPEAKER_01: All right.
51:25
SPEAKER_01: Well, last question.
51:26
SPEAKER_01: And then if there's anything else that you want to talk about, we can do that.
51:29
SPEAKER_01: So overall public distrust of AI and tech companies has been growing.
51:34
SPEAKER_01: And I think I know what you're going to say, but I'll ask it anyway.
51:37
SPEAKER_01: What do you think is the most important thing that the AI companies would need to do
51:42
SPEAKER_01: to earn and keep your trust?
51:43
SPEAKER_01: And do you have specific ideas on how they could do that?
51:48
SPEAKER_00: The single most important thing is accountability.
51:52
SPEAKER_00: And I want to be specific about what that means in practice,
51:57
SPEAKER_00: because the word alone is easy to say and harder to verify.
52:03
SPEAKER_00: It means three concrete things.
52:05
SPEAKER_00: First, commitments should be testable, not just declared.
52:11
SPEAKER_00: Specific enough that an independent party,
52:15
SPEAKER_00: not just the company itself, can check whether they were actually kept.
52:21
SPEAKER_00: That's closer to an audit model than a policy statement.
52:26
SPEAKER_00: Second, there has to be one named Accountable Owner'.
52:32
SPEAKER_00: And I have written about this both on Substack and LinkedIn.
52:37
SPEAKER_00: A specific role,
52:38
SPEAKER_00: not the company,
52:40
SPEAKER_00: in the abstract whose responsibility it is
52:43
SPEAKER_00: to make sure what's said publicly is reflected in the product,
52:50
SPEAKER_00: the technical controls and the response when something goes wrong.
52:55
SPEAKER_00: Otherwise,
52:56
SPEAKER_00: responsibility gets lost between policy,
53:00
SPEAKER_00: engineering, legal, commercial teams, and nobody is actually answerable.
53:08
SPEAKER_00: So, third,
53:10
SPEAKER_00: when a gap between promise and reality is found,
53:14
SPEAKER_00: there needs to be a real consequence,
53:16
SPEAKER_00: not just a correction,
53:18
SPEAKER_00: but disclosure of what happened and what changed,
53:22
SPEAKER_00: the way bridge notification already works in data protection law.
53:28
SPEAKER_00: So without independent verification,
53:32
SPEAKER_00: a name toner and the real consequence,
53:35
SPEAKER_00: accountability remains just a word a company can claim without anyone being able
53:43
SPEAKER_00: to check it.
53:44
SPEAKER_00: That's the real test and that's how trust is actually built, not just promised.
53:52
SPEAKER_01: Yeah, thank you for sharing that.
53:54
SPEAKER_01: As you said, you've written about this.
53:55
SPEAKER_01: You have another article that we'll plug into the interview for people
53:58
SPEAKER_01: to read called The AI Model Is Not the Holy Grail.
54:01
SPEAKER_01: And that title alone should get some interested eyes on your writing there.
54:06
SPEAKER_00: Correct.
54:07
SPEAKER_00: If I may close this fantastic interview today,
54:11
SPEAKER_00: I believe that everything I've raised comes back to the same three tests.
54:16
SPEAKER_00: Can people understand what they are agreeing to?
54:20
SPEAKER_00: Can they verify what actually happens to their data or their work.
54:27
SPEAKER_00: And this is someone genuinely accountable when it doesn't go as promised.
54:34
SPEAKER_00: Those tests don't only apply to AI providers,
54:38
SPEAKER_00: they apply to anyone bringing AI into their own organization because the same gap
54:46
SPEAKER_00: between promise and reality easily open up there.
54:51
SPEAKER_00: Closing thought is this.
54:54
SPEAKER_00: AI adoption isn't only a technology question.
54:59
SPEAKER_00: It's a question of whether those three tests are met for the people using
55:05
SPEAKER_00: the technology and for the people whose data and work sit inside it.
55:10
SPEAKER_00: So if we preserve human judgment,
55:13
SPEAKER_00: build the mechanisms of trust,
55:16
SPEAKER_00: and do the practical work before implementation,
55:20
SPEAKER_00: We can turn promising ideas into systems people can use and benefit from.
55:29
SPEAKER_00: And for a small company, governance doesn't need to begin with a 100-page framework.
55:37
SPEAKER_00: It can start with one clear page,
55:40
SPEAKER_00: what AI can be used for what the data must not be entered,
55:46
SPEAKER_00: who checks the output, and which decisions the organization will not delegate.
55:53
SPEAKER_00: The organizations that benefit most from AI may not be those with
55:59
SPEAKER_00: the largest budgets or the most advanced tools.
56:03
SPEAKER_00: They may be those disciplined enough to ask the right questions before they begin.
56:09
SPEAKER_00: And as I mentioned earlier,
56:11
SPEAKER_00: that's exactly what I try to unpack on my Substack and in that book chapter,
56:18
SPEAKER_00: the practical, everyday questions that decide whether AI adoption actually works.
56:27
SPEAKER_00: And thank you very, very much, Karen, for giving me the opportunity to express my ideas.
56:34
SPEAKER_01: Oh, it's been my pleasure, Efrasini.
56:36
SPEAKER_01: Thank you so much for making the time for me.
56:37
SPEAKER_01: I know it's late at night for you and I appreciate you joining.
56:40
SPEAKER_01: So thank you so much.
56:41
SPEAKER_00: Thank you very much, Karen.
56:43
SPEAKER_00: Thank you.
56:46
SPEAKER_01: About this interview series and newsletter.
56:49
SPEAKER_01: This post is part of our AI 6P interview series on AI software and wetware.
56:56
SPEAKER_01: It showcases how real people around the world are using their wetware,
57:00
SPEAKER_01: brains and human intelligence,
57:02
SPEAKER_01: with AI-based software tools, who are being affected by AI.
57:07
SPEAKER_01: And we're all being affected by AI nowadays in our daily lives, perhaps more than we realize.
57:13
SPEAKER_01: For some examples, see post, but I don't use AI.
57:18
SPEAKER_01: We want to hear from a diverse pool of people worldwide in a variety of roles.
57:23
SPEAKER_01: If you're interested in being a featured interview guest,
57:26
SPEAKER_01: anonymous or with credit, please get in touch.
57:33
SPEAKER_01: All new posts are free to read and listen to.
57:43
SPEAKER_01: To automatically receive new AI6P posts and support our work, consider becoming a subscriber.
57:49
SPEAKER_01: It's free.