[00:00:04–00:00:31]
Good morning everybody. Welcome to 1508. Nice to see you all here today. We're gonna bunch of people interested in this subject. Maybe not the most sexy subject on earth, responsible AI, but it's really something that a lot of people care about. And so I've tried to put some slides together and talk together today, and hopefully we can have a great discussion in the end.

[00:00:32–00:00:52]
My name is Lisbeth and I'm a senior transition designer here at 1508. I've been working with the UX for many years, as and when I started to be part of our work, I started to think that this is something that I need to know about. I need to know what this can do for me and for my work.

[00:00:52–00:01:18]
But at the same time, I also think about what are the potential pitfalls of using this. So when I think about AI and using AI, this is the place I would like to start. Today. I have these two emotions, these two feelings about it. On one side I have this like wow, this new technology, it could do all these great things.

[00:01:18–00:01:45]
It can help me be more efficient and I can speed up and it can take off a lot of the load. It can do all the boring stuff for me, and I can do all the important things that I really like so that it can really do something for me. And on the other side, I have this a devil or not a devil, but someone on my other shoulder saying like, you need to be responsible because what about all the pitfalls?

[00:01:45–00:02:11]
What about the climate impact of what you're doing? What about these skilling? Will I lose the skills I have if I use AI? What about bias in the models? And what about all the slop out there that is being created right now and is flooding the internet? So I have these two emotion every time people are talking about use AI for your work, and you can probably some of you recognize this feeling.

[00:02:11–00:02:34]
So I have this inner dialog often playing. And I also kind of feel like there is this when we talk about it in LinkedIn and everywhere else. And I say we need to be responsible when we use AI. I feel like I'm the break, like people saying like, whoa, now you're slowing us down. Then we can't be that innovative.

[00:02:34–00:03:00]
We can't do all the nice quality stuff that we can do with AI if we just move fast and do all this. So I feel like I'm the break when I say we need to be responsible. But I have a hypothesis, and that's the one I want to show to you today that there is a positive relation between responsible use and quality efficiency, and also possibly preserving your craft.

[00:03:00–00:03:30]
So I will try to show you a case. I will show you some anecdote, a study that kind of talks about this. And then I would also really love to hear your experiences with this after my talk. But before we dive into all this, I think it's really relevant that we talk about what is responsible AI, because there are also a lot of different definitions going around there, and people are talking about different things when they talk about responsible AI.

[00:03:30–00:03:55]
So how I usually see it is that there are two different sides of responsible AI, and the first one is the one that most people think about when they talk about responsible AI. So that's that's how the AI is built. So that is how the designers and the engineers at Claude or OpenAI designed the LLMs. So they have the responsibility to build the models responsibly.

[00:03:55–00:04:20]
So that is one way to talk about responsible AI. The other way to talk about it is am I using AI responsibly when I use the tools that they created? So when I go out and use Claude, if I use Figma, make, am I responsible in my own use of of of doing this? And today, even though the other part is very interesting too, we can talk about that in another moment.

[00:04:20–00:04:52]
Booster, I will talk about the second one. So it's about using AI to to do something in your work. So it's every time I open up Claude or Figma make or a Notion AI or Figjam and using their magic things in there. Every time I use one of these AI tools and that's several times a day. So there are a lot of points in time where I need to be responsible when using this.

[00:04:52–00:05:14]
All right. And then when we talk about responsible use of AI, there are a lot of different subjects and pitfalls and things that are wrong with AI that we need to think about to be responsible. So that could be some of these. For example, bias are the bias in the model verification. So can I trust what the model is telling me what the output?

[00:05:14–00:05:42]
I get the facts correct. There's also a big chunk about confidentiality safety legislation. I can't say that — legislation, laws. So what can I paste in as context when I use the tools for example. And then there are things about intellectual property. Will when I do this, will I copy someone else's work. And there are more, for example, homogenization.

[00:05:42–00:06:06]
So will everything look the same? What I do look the same as the rest of everything that is produced. Transparency. Should I reveal to everybody that I used AI for doing what I did and then the big one environment. What does this cost in energy and water? When I when I use the tools. And also a lot of people are starting to talk about this dependency.

[00:06:07–00:06:27]
So if I use AI a lot, will I then start to lose my own craft because I get so dependent on using AI for this, and there are probably a lot more. Maybe one of you are sitting and thinking about what about this obvious one? Anyone I missed? There are probably more because we could talk about this the whole day.

[00:06:27–00:07:11]
There are a lot of different things that can go wrong when using AI, and a lot of pitfalls, so it is a bit difficult to be responsible, responsible if you have to think about all these things. Whenever you use a tool, every time you open Claude up, I need to think about all these different things and probably more things. So what I have found very useful is to find a model that helps me group it a bit and think about it in a different way. And I don't know if you know this model, it's called the Four D's and it's created by two professors, but anthropic is starting to use it as well in their courses. So so they have kind of adapted it and worked together with the two professors.

[00:07:11–00:07:41]
And the model looks a bit like this. We have adapted it to our own design of course, but it's it's about four different aspects of using AI. And it's for to to both be efficient and effective. But it's also to be safe and ethical. So they kind of bunch it all together in this model. But let's dive into the for this just to have this framework ready when we talk about the cases later on.

[00:07:41–00:08:03]
And the first one I like to, to start with when I describe this model is description, because description is about how you prompt to get the best behavior out of the AI. And that is probably the thing that most of us think of when we think about being responsible with AI, because that is where you start. Like you go to the AI, you prompt it to get something out.

[00:08:03–00:08:29]
And what is in this prompt? That's what description is about. Then we have delegation. Delegation is actually a step before you go and write something in the prompt. This is where you decide what will the AI do and what will I do. So here you decide whether the AI is good enough for the task and which model you want to use, which tool you want to use.

[00:08:29–00:08:54]
So that's a thought to do before actually prompting. Then there's discernment. And discernment is I see it as kind of an assessment. You do so after you have prompted or got an output from an AI, then you need to do some kind of assessment. You look into what the AI did, the output, but you also look into how it came to this output.

[00:08:54–00:09:23]
So you look into the reasoning of it. Why did it do this and why did it do that. And here you can also see if it's a bit skewed, for example for biases and and other things. The last one is called diligence. And diligence is about taking responsibility for what the AI did. So it's after you're done, you have prompted something, you got an output, you did your discernment, and you are okay with the result.

[00:09:23–00:10:02]
Now you go out and present it and now you need to be transparent about that. You have done this and you own the result that the AI did. So it's about like saying, I'm responsible for this result that I did with AI. Yes. So these are the Four D's and I will try to use them when we talk later on in this in this session, because I think they are really good at getting around all of this, all of these many subjects in one time. So you also know more when to use what if you think about these Four D's.

[00:10:02–00:10:36]
And as I said before, this model was actually created to or this framework was created to get people to be more effective, efficient, ethical and safe. So they kind of bunch both the like the efficiency and the speed, together with the ethical and safe, the responsible part. And I think that's pretty interesting regarding my hypothesis that there is some kind of relation between the efficiency and the quality and also the, the responsible part, because they don't really divide it in the model.

[00:10:36–00:11:06]
They say they kind of say that diligence is the responsible ethical part, but as they describe it. They also describe a lot of responsible things in the other 3DS. So I think it really gets around everything. But of course, not all the different aspects, but some of them at least. All right, let's dive into my first example. This is an imagined example, but it looks pretty much what like something we could do in a normal day.

[00:11:06–00:11:34]
So here I wanted to show you what happens when you prompt something in Figma to create a prototype. So we use make sometimes to get inspired to do other directions or to find new features for, for example, when we do a website. So we could prompt it and say like we want this and this and that for our client, can you make some ideas for this or can you make an idea for this?

[00:11:34–00:12:01]
And usually we would do a prompt that could look something like this. It's not really it's not really lazy. I would say we have some content in it and it does say what we wanted to to, to do, but we are not really diving into all the responsible things when we when we do a prompt like this. And if I prompted this for Figma make, I would get a pretty nice looking website I guess.

[00:12:01–00:12:29]
And this website, I think it looks nice, but it's also pretty generic, right? It's like it has this big hero image that most theaters do, and it has like all the different stuff that a regular theater would have. And it doesn't really help me to get new ideas because it has all the ideas that I already know. So then I thought, why not try to to design the prompt a bit better?

[00:12:29–00:12:54]
You don't have to read it all, but I wrote a long prompt to to to get some of some, some more aspects in it to. So I thought about like, how can I get it to be more unbiased. So for example, to look at more different angles and not to do the classical website. So also not to be too generic and to homogenic with all the other sites.

[00:12:54–00:13:22]
And I did a problem where I said like, do this, don't do that, and listed a lot of different things here. And of course some of it is also you can this defined definition of what is responsible prompt and what is just extra context. And that is a kind of also my point. So when you try to be responsible, when you're prompting, you will think about more things that you need to include in your prompt about what you actually try to get out of, of the model.

[00:13:23–00:13:44]
So when I did this, I kind of thought, okay, it's for the use and I don't want this type of branding, I want more this type of branding. So your thoughts are starting to think more about like what is it actually I need and want when you start to think about these, these things and I know it's not perfect and not beautiful at all, but it's kind of different the results I get.

[00:13:44–00:14:09]
So here, instead of having a big hero image on the front page, they actually put the program on the front page. And another thing is that in the description, it puts in this section about what happens when you get there, what to expect from this, because I wrote in my prompt, the response prompt that that it's for first time users that don't really know what a theater is.

[00:14:09–00:14:36]
So it starts to put in more ideas for making this more specific for me. Yes. And it is also really nice to go into the reasoning of Figma Make. So my diligence here I go, my discernment. I go into the assessment and look at like, why did it do this? And here I can see that it's actually thinking about some of the of the things I wrote. So why is it actually not having a hero image in the top?

[00:14:36–00:15:00]
And also why does it do this note about what to expect when you get there. And we are also seeing this in other areas. So I talked to one of our developers, Victoria, and uses Claude to code sometimes, and she told me that.

[00:15:00–00:15:31]
Because she uses a lot of time to actually do a really good prompt in the beginning, she does she do spend a lot of time on the prompt and on doing skills and doing context before actually coding things, because she does that, then it saves her time in the end. So being responsible and actually putting in all these safeguards and thinking about how the code will look like in the end actually saves her time afterwards, because else in that she would have to go through all the code and change a lot of different things afterwards.

[00:15:31–00:15:55]
So at least in her work it gets her to be more efficient. But I also see that it's kind of you need to do the manual work at some point. It can be before and it can be after. And I think what I usually see is that if you spent the time before and not after, then the quality will also be better of what you get the first time.

[00:15:55–00:16:20]
But sometimes it's not enough just to do a really good prompt, really good context, really good description, and to do this assessment afterwards. And I have brought a little case from, from the real world that we are working on. So we are writing a book with one of our customers and it's an AI playbook. So it's about AI.

[00:16:20–00:16:48]
And of course then when we started to do this, we thought we need to to experiment a bit on how to use AI for writing a book. So we we did all the good things. We thought to set up a really good project in Claude. So we inputted a lot of different things in the product instructions about how it should behave in different situations.

[00:16:48–00:17:19]
For example, always note when you don't have a source for this and and you need to to write all where are you missing facts and all the different things that it didn't know it should highlight for us. So we wrote a long description here in the project instructions. We also had a lot of context. So we did a whole Notion board where we put in all the different documents that we got from the customer that it should look into, and we did all these things around it.

[00:17:19–00:17:49]
We also built a skill for Claude that could write in the tone of voice of our client. So we did a lot of legwork to actually get it to write something good. And then we prompted it also a very long prompt to to do our first chapter just to see what came out of it. And we were aware that this might not be something that we could use, because we know AI sometimes is not the best tool to write stuff, because it just writes things that are not really correct.

[00:17:49–00:18:12]
Right? Or like sounds correct, but isn't really correct. So we tried and then we looked at it and what happened was that at first we we were a bit impressed. We were like, okay, it is actually writing good. It's writing like our client. It is using the sources the way we wanted it to use the source. That is correct.

[00:18:12–00:18:33]
What is writing? So let's try to use this as our first draft. Let's see if this is something that we could work with. So we started to go in and do our assessment of the chapter and read it through. Try to to I put in a lot of notes and we spent a couple of days doing this, trying to get the chapter right.

[00:18:33–00:18:55]
And then after these two days we had a meeting in the team, like a bit lost in this. It's kind of like, we can't really get this chapter right. It sounds right. It has all the right things in it, but it's not really what we wanted. It kind of. And we are lost all the the red thread is lost during this chapter.

[00:18:55–00:19:18]
We need to to do this all over. So we ditched it and we started all over. And yeah, that's of course a decision that is pretty difficult after two days. But what happened was that after we did this and started to think about when do we use our Claude, then if we're not using it for writing chapters, when do we use it?

[00:19:18–00:19:42]
And then it was actually pretty fast to get it written the chapter, because then afterwards we started to use Claude More as a research assistant and as an editor. So we asked the question, for example, dive into the material and look if you can find where they set this and that, because I can't remember where it is. So it helped me find it in the material.

[00:19:42–00:20:24]
Then I could go in some of the material myself and figure out, like, what did they write? Also, we did the red thread ourselves. We decided like, we want this and this and that in our chapters, and then we could use it for writing small pieces of content and also rewriting some of our content. But we were the editors. And the problem with this case was that we didn't think too much about delegation. So we thought like, let's try to give AI the. Or maybe we experimented with delegation, but what we learned was that we needed to delegate in another way. So we should shouldn't give the AI all these different tasks. We should do them ourselves.

[00:20:24–00:20:50]
All right. So now I've talked a bit about like efficiency and quality when using AI responsibly. But what about de skilling. Because that's also something people are talking a lot about right now. Like are we losing our skills when we use AI? And there are a lot of different studies, scientific studies about this coming out right now, and some of them are in better quality than others.

[00:20:50–00:21:12]
Some of them are pretty criticized also, so it's also a bit difficult to find something that is like this is proof of it. And maybe it's also because it's happening really slow. So it's not something that you can just see. But in this study, they said that they could actually see it after three months. And the title of it is a bit medical.

[00:21:12–00:21:38]
So I tried to translate it to normal language. So what happened in this study was that they found out that doctors using AI tools to detect, detect potential cancer tumors risk losing their skills. And how they did it was that they had for endoscope centers, that centers where they do examinations with cameras in the bowels. They went into.

[00:21:38–00:22:04]
They had like these four centers in Poland. They had a new AI, two that they could use for their work. So this AI two could help them detect these cancer tumors or potential cancer tumors. And they had the tool for three months. And then after three months, they gave all the data about the the detection rate to to these scientists.

[00:22:04–00:22:37]
And what is notable here is that sometimes they use the AI tools and sometimes they didn't. So the same doctor could have one session where he used the tool, and another session where he didn't use a tool. And what the scientists did was that they looked at the data from before they got the tool, and then they looked at the data from when the scientists didn't use the tool but had access to it sometimes, and then they could see that the detection rate, it fell from 28.4%, where they could detect these cancer tumors to 22.4 when they didn't have the tool, so they had access to it.

[00:22:37–00:22:59]
But at the times when they didn't use the tool, so their conclusion was a bit weak, because it was only four centers in Poland, and it was more like a historical study than an actual real scientific study. But it was something that came out in the media and people were talking about this. And this killing is not a new thing.

[00:22:59–00:23:40]
These skilling is not something that we only heard about with AI. It's something that, for example, in the aviation business or for the pilots they have seen for a long time, back in 2013, the authorities in US, I think they're called FAA. They came out with some, some like guidelines to all the pilots out there because they had looked into data about use or like the flights, and they saw that when the when pilots are having all these automation, all these digital systems, they kind of lose their ability to fly manually on the planes.

[00:23:41–00:24:10]
So when something happens and the systems fail and the pilots need to use it manually, they are getting worse. So they came out saying like, you need to have some rules for how to handle this in 2013. And then so they said set of rules and guidelines. But then in 2017, five years later, they saw that it is still a problem, even though they set up rules and guidelines in the aviation business, they needed to have something more practical.

[00:24:10–00:24:36]
You need to go into simulators and practice manual flight, because else you won't be able to handle planes when the systems fail. So these are kind of some rules that now is applied to all pilots in the US that they need to go to simulators and practice sometimes. And when I see this, I kind of think about what AI is doing to us.

[00:24:36–00:25:00]
Maybe it's kind of the same thing here that we're seeing right now. We are trying to make all these guidelines and rules for how to use AI responsibly. But do we also need some practice in not using it sometimes? So when we look at this model you can see that the that maybe it's about the delegation part again.

[00:25:00–00:25:25]
Maybe sometimes even though the AI can do something for you, you need to say like I need to practice my skill. I need to say for this exact job, I want to do it myself, even though I know the A, I can do it because I want to practice. That's at least what I think could help. But another thing is that right now and today we are talking a lot about us.

[00:25:25–00:25:51]
Me, you using tools alone and our responsibility to use it responsibly. But this is if we see this as a scale, if I need to, to do something in the world to make AI more responsibly, it's a pretty heavy lift. If I if I want to change the tip of the scale. So what could we do if we wanted to tip it more?

[00:25:51–00:26:13]
We could do legislation, or we could go to the AI builders and OpenAI Figma and say, like when you're building all these tools for us, please put in some defaults so we will be more responsible in our work. And that is, of course, something maybe out of scope for our room. Maybe it is in scope, then think about doing it.

[00:26:13–00:26:53]
But what could be in scope for us is to look inside our organizations. How can we, in a more systemically way, try to make some defaults for the employees so they work more responsibly with AI? Yeah. So what can we do? Because right now what we see is that a lot of organizations, they have a lot of rules when they build products where AI is part of it. So there they have like guardrails and the legislation or not legislation, but like internal guidelines on how to create products where AI is part of it because then they are responsible for the product.

[00:26:53–00:27:18]
But when it comes to using AI tools, they are more like this is up to you how to use these tools. You just need to be responsible for your product so they are putting it all on the individual users. The output of and the result they get from from from the AR. So what can we do? A good place to start is of course to do some AI guidelines and principles.

[00:27:18–00:27:42]
And this is a copy of our principles right now. It's a working document, and it's a good place to start because it gets us to talk about like what is responsible in our business, how do we want to work? What what boundaries do we want to set? And when that set, I think it's pretty difficult to to use it for actual practical work.

[00:27:42–00:28:09]
I don't know how many of you have have AI guidelines in your business or organization. Yeah, that's not many. And how many of you are using those guidelines? Every time you are using an AI tool and know all the rules in that AI guideline? Yeah, it's yeah, yeah, you are good. So one yeah, it's difficult because you don't think about all these rules.

[00:28:09–00:28:32]
Every time you pick up Claude or Figma make it's difficult. Of course we can put in like practice and good advice and all these stuff, but it's difficult. And so what we need to do is that we need to set some better default for our employees. And we are trying to experiment a bit here in 1508 on how to do this.

[00:28:32–00:29:07]
And one of the things we found was really helpful is to create some better systems or some reusable systems. So for example, Victoria, the developer, I talked about, she when she works with Claude, she, she has this whole package of skills and documents that she uses every time she starts up a new project. So she has these tests that go in and check for different things afterwards, and all these different description of what it needs to do and don't do.

[00:29:07–00:29:47]
So she can reuse this every time she starts up a new project. So she just has to start like all over and like what is responsible for this project. So she has this package that she can just apply every time she goes into a project. And that is, of course, something that she can share with the rest of us. So we can also use this. This is very much for coding, but of course a colleague said they can use this as well, and maybe we could build more like these for also when we do UX and design and other stuff. So think about like creating some packages or reusable skills or systems that the employees can, can use and reuse.

[00:29:47–00:30:10]
Another thing that we are thinking a bit about is creating a reusable practice. So what we did was that we divided in our crafts in here and then had a workshop where we talked about like, where do we want to use AI, where don't we want to use AI for different tasks, and how will we use it when we decide we use it?

[00:30:10–00:30:37]
So actually talking in specialized groups about the different tasks we have and how to use AI for this in the most responsible way. So that's the way that you can kind of put some frames on it. Of course, it's still difficult to actually when I do facilitation, then I go in and how do I do that. So you still need to dive into the details of each and every one of the tasks to talk about.

[00:30:37–00:30:57]
What do I do here, and maybe do some examples of it that you can share with your colleagues. And then the last one is there are a lot of settings that you can do in all these tools on an organizational level, and a lot of companies that don't really go into all these settings because they are pretty nonsense.

[00:30:57–00:31:23]
Some of them, they are pretty hard to understand. And maybe the the one that is the admin doesn't really know what all this can do and what it does. So spend some time, go into the settings, for example on Claude and look at what, how how you can set some boundaries to be more responsible. For example, you can you can go in and tell it which websites are all my employees are allowed to use agents on.

[00:31:23–00:32:06]
So where can agents go and what are the boundaries for agents when using Claude? So it's a really good place to go in and try to figure out what are these settings about. All right. So we have been a lot about around all these different four circles. And most of them are about like description delegation and discernment. And I think all of these three are kind of like the the reason or when you do all this right, then you are more eager to be diligent. And when we do all these frames and we do all these defaults for our employees, then it's also easier for them.

[00:32:06–00:32:52]
It's with less risk for them to be diligent and say, like, I did this and I am responsible for what I did with AI, because in your company you already have some defaults for so. So it's not that many things they need to to take responsibility for. Yeah. So I guess I didn't really prove that it is better quality or more efficient, but we got a bit around some of the thoughts that I have about this subject, and I'm really eager also to hear if you have any cases where you've seen some of these things, or if you have any questions or any insights on this, then let's share it in the room and get a bit smarter.

[00:32:52–00:33:20]
Yeah, anybody who wants to share something. Any questions? Yeah. Was wondering have you considered the bias in the model itself? Yes. Yes, of course it is. What happened with the environment? Yeah. The environment is not really part of that model too much. Of course, you can talk a bit about like when choosing a model or for example, haiku for translation instead of opus.

[00:33:20–00:33:43]
That is a bigger model. You spent less energy on your prompt and that could be part of the model, but it's not really that much in in all the environmental. That's true. Yeah. So you need to think about the biases in the model. Of course. Yeah. Yes.

[00:33:43–00:34:24]
Anybody who have done something of this and where they saw like a result that was better when they thought a bit more about all these things. No. Am I am I totally wrong? Like that's also interesting. Maybe it's just me babbling here, saying like, all this is getting us to do better quality. Is that wrong? Yeah. You have a question? Yeah. I had a thought when when I started working with it, that I actually started to move a little bit too fast, Claude, and then figure it out that I need to start with.

[00:34:24–00:34:54]
Design system. Do you have any like, reflections on that, like where to start? And because I thought that my mind was going in like, I want something out of it and in the wrong place. Yeah, I started the maybe a bit too high on the fidelity ladder. Yeah, I think that's a classical. Right. You think now you have this tool, you just want to get going, and then you are like, I don't need to think about all these things now.

[00:34:54–00:35:17]
I just do the prompt and then I see the result and then I can do it over again if I if it's not good enough. Right. But the trap is that then you see the result and then you're like, okay, it's probably good enough. So I would say that you need to think about like before starting and try to be a bit okay with spending time in the beginning.

[00:35:17–00:35:54]
And maybe it's kind of a mind shift mindset shift that you need to think about before you start. Do spend some time, and it's okay to spend some time on delegating and thinking about the prompt. And then it's of course different from project to project where where you set boundary and how much do you want to do it? Because sometimes when you're just doing like a small piece of thing for internal use, it's maybe not that necessary that you create a long prompt and use all this time on it. But if you are having an agent that goes around the web and does a lot of things, then it's much more important that you set up all these things from the beginning.

[00:35:54–00:36:32]
Yeah. So the more like autonomy the AI has in what it does, I think it's more that that could be a rule. Like then think more about what you do in the beginning. Yeah, yeah. I think when I think of the first thing I think of is the environment, and I don't know if there's any way currently to measure how much energy you're draining, I guess, by this constant, if there's any difference and if you have a more detailed, specific descriptive prompt that you're using less energy, or if it makes no difference, I don't think you can do that.

[00:36:32–00:36:56]
No. And that's probably also the problem, right? There are not that much transparency on the energy usage and water usage and how it affects the environment. Sometimes we get these reports that gets to the media, like your prompt is the same as loading your like charging your telephone for two days or something. But you can't really think about that when you're prompting.

[00:36:56–00:37:18]
While it's difficult to connect these to and we can't be sure that it's true, and sometimes they say like so. So Microsoft is saying something and then anthropic is saying something else. So it's also kind of difficult to know. And that is probably also why it's not part of the model, because it's difficult as you to think about the environment every time you're prompt.

[00:37:19–00:37:41]
You can, of course, if you have it in the back of your head every time, then it's then it could be part of delegation, right? So you say, Will I actually use AI for this or not? So think about the task you're doing. Can I do it better and can I do it without AI? And there are also people saying like, you can't be responsible at all when using AI.

[00:37:41–00:38:26]
I know we have people in here saying that as well. So yeah, but but you can of course think about delegation, where to use it and where not to use it. Yeah. Responsible. Yeah. You said responsible. Yeah. Yes. Sometimes I have like a, some kind of like back and forth in the way that I, that I have to interact with the, with AI and you know what I was asking. But it's like sometimes there is all of these new models and I'm, let's say, I'm sure that the the amount of time that I would have to spend together would prompt, especially in a very complex task, is actually even worth it.

[00:38:26–00:38:48]
So it's kind of like there is this extra part of the project that I have some kind of testing that actually can actually do it, or because maybe three months ago you couldn't do it. So it's like, yeah. And then of course it's going to take extra step, double work. Because first I have to prove and proof prove the concept and then to actually do the hard work and stuff.

[00:38:48–00:39:23]
So it's more like it's hard to be also responsible sometimes if you don't know if the tool actually can handle and ask and if you it's worth spending all day structuring a prompt, and then it's probably for nothing if it's if you cannot handle, I don't know if you have some experience. Yeah, yeah that's true. It's changing all the time. So if you're if you think you can't use the tool for something, then in your delegation you won't use a tool for it. But if you have like a thought that maybe I can use this for it, you have in your delegation phase, like you need to try it out to see if it can actually do it. Yeah.

[00:39:23–00:39:47]
That's true. Yeah. And that is of course a bit more responsible or irresponsible because then you use the AI and maybe figure out you can't use the result. But sometimes you need to know to actually be able to delegate the next time also. Yeah. But it's always like the scale and these two like angel devil on your shoulders should I, shouldn't I?

[00:39:47–00:40:12]
And this inner dialog I guess. Yeah. Yeah. That was a bit curious about just thinking about these skill and dependency on the tech. Yeah, I mean, the the image of the plane is very clear. Obviously need to be able to land a plane and systems fail. Yeah. But another mode of transportation cars, it used to be more mechanical.

[00:40:13–00:40:41]
We could all change here. So it has changed at all. Now we're totally dependent on a mechanic if something breaks. Yes. So how do you work with the skilling and CEO? Yes, I think it's something pretty new, but it's something that I've been aware of for some time for myself. Like, I had this hunch that I was actually getting worse at some of my job because I used AI for it too much.

[00:40:41–00:41:02]
So I started to roll back a bit and then do, like I said, like sometimes I want to practice this myself and not let the I do it. For example, write a text. Sometimes I'm not using it for the first draft, then I will write it myself before, and then I use it more as an editor. So sometimes I switch around the roles and play a bit with the delegation part.

[00:41:02–00:41:26]
And yeah, and I think the airplane case is really clear because it's so obvious when they are like losing their skills. Right? The plane will crash. So but here we can't really see what happens when our skills decrease. Right. Because it's it's not something that's crashing. It's just a product getting a bit less quality or, or something else.

[00:41:26–00:41:55]
Right. So it's more a slower steps and not that easy to see I think. Yeah. There was a study coming from MIT saying that 25% of people working in marketing built AI, right? Greatness like beginning of 25%. That's that's a lot. Yeah. And I can feel it myself sometimes. I used to have like this. I can feel in my head when I'm thinking a lot.

[00:41:55–00:42:15]
It's like, I don't know if you have the same feeling, but when I really have the complex tasks, I can feel it in my head that it that I'm thinking. And when I'm using AI, I'm not having that feeling. And it's really clear in that way. And then also when I then get the result from the AI, I'm more lost.

[00:42:15–00:43:00]
As I said in the case. Like I get lost in what it gives me and then I sometimes just give up like, okay, then it's good enough, right? If it's not, if it's not important, then you just say, okay, then it's just good enough. But then you just sometimes also have to say, yeah, I need to start over. Don't you say I for this. Yeah, yeah, yeah. This as a follow up on that. But this feeling and also the responsibility of the, of the companies, I think it's also a question of making some progress in this case in China as well. Like you say, you need more time to practice manually. And do you of human work also to be happy employed employees as well.

[00:43:00–00:43:24]
So I'm just curious about how can you make a business case against like a client or your boss, whatever that says, you can actually do this faster. How come you had to spend another day because you need to practice? Two where is that discussion? Do you know what I mean? Yeah, yeah, it's a it's a difficult question. I don't really have the answer.

[00:43:24–00:43:49]
Maybe one of you have a good answer for this, because it is because it's something you do for you, right. Or for for your employees. But in the long run, I guess the quality of the products we are making will be better if we keep our skill set up to date. Right? But but it's hard to argue to a client right now that you need to spend that extra time on manual work, right?

[00:43:49–00:44:13]
I'm also wondering, like, is it a value proposition that you do responsible AI as a, as an agency or as a company, because it's understood that it takes time to make a skill. So it's like there's a lot of communication around it that maybe has to come with all of this is like, why? Why is it it's a long term things.

[00:44:13–00:44:38]
Well yeah, yeah, yeah, it's a good perspective. And I guess right now the thought could also be that if we spend more time on, on like doing it correct, then we also see better quality. So that could be like an argument saying like being responsible, you're not only being responsible but you're also getting better quality in your product.

[00:44:38–00:45:04]
But but yeah. Yeah. And it's yeah you're right. Yeah. It's difficult. Yeah. I spend a lot of my personal time on trying to develop my skills in Claude, like a lot of so on YouTube tutorials and blogs and whatnot, I can find to start it up. And I noticed everyone in my team have different levels of use of AI.

[00:45:04–00:45:30]
How do you, in your exam team or company, make sure that everyone has some kind of development or the same kind of level, a base level when you're using AI? Yeah, I guess there will always be some people that are more ahead because they have the interest in actually doing it. And that's I only see that as a good thing, because then they can help teach the rest how they can use them it best.

[00:45:30–00:46:01]
So here we have like these Friday meetings sometimes where we have an hour together where we then practice something. So that could be, for example, how to set up Claude the correct way or how to use Figma make and stuff like that. So sharing knowledge about this and sharing cases also on how you did this. But yeah, it's a good idea to do it because then you get everybody on some level of of how to use it and how to use it the best and more responsible way.

[00:46:01–00:46:25]
Because there are also we also see like if you don't do this, then people are using it in different ways and some of them are more responsible than other ways. So that's also a way to kind of get people to do it the correct way. Right? Yeah. Yes. What did you experience? What are your experiences with people with Figma Make and Claude and so on.

[00:46:25–00:46:50]
Do you prefer. Yes, that's a difficult question. And and maybe I'm not the best one to ask because I'm not using it that much, but I know some are using it where they start in Figma. Then they go to Claude and do something, and then they go back and forth between those two. But it's also something that shifts right now.

[00:46:50–00:47:18]
What is better for what? I'm sorry, I don't have the correct answer for me because I'm not using it that much. But but myself, I use Figma, make more for like ideation and not to actually do prototype. Sometimes I can of course take something from Figma make into my own figma file and then edit it. But most of the time it's more for getting inspired to do something in another way than I just thought of.

[00:47:18–00:47:46]
But yeah, there are definitely a different uses of Figma make. Yeah, maybe other people here have some good experiences. Yeah, yeah. Any questions or any. Yeah. There's not a question, just a reflection because I really appreciate and like the 4D model here. But I also understand it when we walk through it I understand it better. But you also said it's written by two professors.

[00:47:46–00:48:11]
It's it's very complex critical thinking. What really represents some also curious as to see where we're at this as a boil down guideline or best practice, or have you thought of how that would translate into not just professionals like courses room, but also into schools, for example, where the whole this part is the new. You won't have a calculator in the future, just have a phone.

[00:48:11–00:48:37]
But it still affects how children and young people develop their. So I'm very curious to see the development of something like that, because as of now, it's it's quite complex because it needs to be. So thank you for showing walking through it because I think that's that's what needed now even for someone like us in this room. So let's start.

[00:48:37–00:49:12]
But yeah, as you also said, it's not perfect. It has biases and not everything is included. So maybe more development of this model could be good. They actually created the model for their students at the schools. So they are using it in school in their theory to help them use AI in the best way. Yeah, yeah. I'm curious when it comes to your ideation of design projects, because it's proven that AI gathers all ideas and previous campaigns and visuals and art direction.

[00:49:12–00:49:55]
And I'm curious if you have, you know, any reservations when going through that process? What do you do to make sure that your design stays unique? Oh, that's a big question. Yes, I guess we we mostly use it for like, of course, we always have our own ideation. So it's more for like pushing us and to see new angles that we didn't think of. Because when you prompt something with an AI, you often get something that looks the same as everybody else's. But sometimes you missed an angle or you missed like something that you could have used.

[00:49:55–00:50:18]
So it's more like getting those extra inputs into your product. That's what is how I use it. But I think it's also different from from person to person in here, how they use it in the creative process I yeah. So it's mainly to to push me outside my box because when I start ideas I have a box and some frames I'm inside but sometimes see something else.

[00:50:18–00:51:11]
And also if I give it more context, for example, then it can look into those documents and find something that I can't remember because I can read all these documents once, but then I can't remember everything, but it remembers for me. So sometimes I'm more like, there was something about this subject, could I use this in some way? And then it helps me bring up those ideas and refresh my memory. So it's also for refreshing memory and and finding stuff for me. Yeah. So it's more about asking the right questions. Yeah. Prompting is very very important here also. Yeah. Yeah it is. Yes. Just comment on I really like how you talk about responsible in a more personal level basically, because sometimes I feel like it's a bit over my head, but like a responsibility to do.

[00:51:11–00:51:37]
So. Yeah, I like where it's like, no, literally like that little small step. And then my question would be like, how do you, as a company, communicate the transparency and use of AI to client? And also, what is your experience so far with clients like Casper Against in using AI, which is. Yeah, it's it's the trust is of course different from clients or client.

[00:51:37–00:52:13]
But what we try to do is that we write it in, in the offers and in our contracts, and then also when we always have a point of departure, it's called. So when we start up a project, we actually before that we have an internal point of like kickoff where we decide, like, how do we want to use AI in this project? How do we see it being used? So the delegation part maybe, and then when we have the point of departure with the customer, we are saying like we expect that we would like to use Claude for this and Figma for this. This is how we would usually do it. How do you feel about this? Would you be okay with us doing this?

[00:52:13–00:52:34]
So we are transparent about that. We want to do this. And then of course we can also say to them, like, I did this with the help of AI, but but we kind of have a rule in here that we don't send like something that AI did without editing it ourselves first and reading through and put our own mark on it.

[00:52:34–00:53:10]
So I would never say like, I did this with Claude and now you can just read this. No. Yeah. We also elaborate on the meeting we had yesterday with the museum, where we on one hand use AI in the product, but that's for the plan. And then the client's customers really experience I don't. So we had a discussion in between when to use AI and externally within the project. Yeah. So we had an entire hour where we discussed like how do we want to use AI in the product that we are doing with you?

[00:53:10–00:53:34]
So for for your users, how will we use AI so that they can get an experience? But we also talked in our point of departure earlier, like how are we doing it in our own project. So the transparency is like everywhere, both on the product level and on how we work level. Yeah. That's true.

[00:53:34–00:53:52]
Yeah. Any other questions or insights? We have a little time for one more. No one okay. Then let's say that's it for today. And thank you for joining. I'll stay here for some couple of minutes so you can just come and ask me questions if you have some. Thank you.