AI in Learning & UX Design: Augmenting, Not Replacing

Mike Brown at UXDX EMEA. Video: https://youtu.be/FNIDrJgO_Fg

Readable transcript: edited from the recording's captions for readability (fillers and false starts removed, punctuation and section headings added). Wording is the speaker's own. Timestamps are positions in the video. Names marked [?] could not be verified against the audio.

Leading design through change at big organizations

[00:00:09] Thank you. Hello everyone. Thrilled to be back here at UXDX speaking again. Quick question: is anyone hungry? Yeah. Well, not long to go now, I promise. You just have to get through one more talk and then it's lunch. But today I'm going to be talking to you about AI and learning and UX design.

[00:00:25] There we go. I've been building and leading design organizations for quite a while now, at the variety of companies you can see here, and right now I've started at Barclays and it's a bit of a change. My background has mostly been in UX and product design, but the organization that I lead now within Barclays is focused on learning design. I still have a number of UX designers, but the vast majority of the designers in my team are learning designers focused on learning for colleagues, which is quite a nice change from a really heavily commercial imperative.

[00:01:07] But if I look across all this experience at all these different companies, they're obviously quite different businesses, quite different products, but one thing that's similar is they're all quite big. These are all big, complex organizations, which brings a lot of opportunity. You can really do things at scale and deliver quite big things. But for anyone that's worked in a large organization, you probably know all the disadvantages and the difficulties you can have as well, getting things through bureaucracy, politics, and often a lot of tech challenges you have in a really large organization.

[00:01:39] But one thing that's really consistent about all this experience, and why I'm lingering on this, is that in almost every one of these organizations when I was there, we were going through a really big period of change and transformation. Quite often that was a digital transformation, moving things into a new way of working, also agile transformation. So I've got a lot of experience in leading teams, with teams really being at the forefront and helping to take companies through change. And I think this is important because right now we are all really at the start of massive changes. I think, hopefully, it's apparent to everyone that we're really just at the start of what AI is going to do to how we work and to how we live as well.

AI won't take your job, but someone using it better might

[00:02:14] First, just a question for everyone in the audience here. Can you raise your hand if you've ever heard this statement before: AI is going to take our jobs. Anyone heard this before? I hear this all the time, in the media, on LinkedIn, from plenty of colleagues. And I think it comes down to fear. It comes down to people. People are worried about the pace and the scale of the changes in AI, and they're understandably worried.

[00:02:39] But I think the reality is actually closer to this: you won't lose your job to AI, but to someone who uses AI better than you do. This is a quote from Jakob Nielsen. I'm sure most people in the audience are familiar with him, a very famous UX practitioner and writer. And he now seems to write almost exclusively about AI and how it's going to affect design and UX. He's really bullish about the opportunities, as I am. But he makes a great point that it's incumbent on all of us to adapt and to learn how to use AI, because this is the future that we're all heading towards.

[00:03:12] It's very early days so far. There haven't been that many studies. There hasn't been that much time. But almost everything that has been done so far, all the measures that we have, show that it's going to lead to really big benefits. Studies across a whole range of different industries have shown that for most workers, when they start integrating AI into their daily workflow, their productivity goes up by 66%. You can see there are huge financial benefits to the wider industry and companies as well. And then there are some figures on the right-hand side in my area at the moment, which is learning and HR. The numbers all add up.

[00:03:43] So the opportunities for incorporating AI are enormous. But what's the key to realizing them? Well, in our experience, it comes down to measurable gains. There are many applications of AI that go well beyond personal productivity. I think most people in the audience have probably used ChatGPT and started to find some things that make things faster. But I think the key is really understanding what works for your business in a business context, and particularly the parts of your business that you could apply AI to, and then measuring those effectively so that you can actually realize the benefits.

Setting up an AI innovation squad in learning design

[00:04:14] Last year I started in my new role at Barclays. I've only been there for about a year and a half. When I came in, my organization was spread out all across the UK. I'm based in Glasgow at the moment, but they're spread out everywhere. So very quickly I started organizing a series of workshops all across the UK to, first of all, meet my team, but then collaboratively work on what's our strategy, how are we going to solve some of the big problems that we had. And one of the focuses that we had was working out how we're going to start testing and incorporating AI into our processes, because it's a really big focus at Barclays, and particularly in learning as well.

[00:04:47] We have a whole innovation chapter, a stream within our larger learning and development organization, which is largely focused on trying to incorporate AI and realize the benefits there. And what we actually found was that learning design and design could be the tip of the spear and one of the first parts of the business that could start testing this. I'm not sure how familiar everyone here is with learning design. It's a relatively new thing for me. I've been learning a lot over the last year. But the best way I can describe the learning designers in my team is they're halfway between a copywriter or content designer and a product designer.

[00:05:23] They largely produce interactive learning modules. There are lots of different kinds of outputs as well, but that's one of the main things that we produce. And well more than half of that process is working on the scripts, working with stakeholders, getting the scripts and the copy right before it then moves into the interactive design elements at the end. So in all that part of the journey, all that part of the process at the start, which is really focused on copy, we could imagine all these opportunities to bring AI into that process. AI exists now. From the experiments where we'd just been playing around with ChatGPT, we thought we could actually start using this in a business context.

[00:05:55] What we did is we built up an AI innovation squad, a specific squad with a few designers, a scrum master and a product manager, to start testing these in detail. We looked at specific use cases, got a wide number of people feeding in, came up with a huge number of use cases that we could do, then narrowed them down to the very best ones, then rated them for feasibility and importance, and then set about testing them in 2024.

[00:06:22] We started small initially. These are some of the learning design use cases that we started testing last year. Some of them seem quite small. They're more based around process improvements rather than massive changes. But that was deliberate, because what we wanted to do, at least at the start, was to be working on things that we could test and deliver ourselves. There are a lot of probably much more exciting use cases, but all of those would have involved integration with existing systems, further development, collaboration with other departments. What we wanted to do was concentrate on things that we could actually test and deliver within our teams.

[00:06:55] With each of these, what we did is we first established a benchmark as well. We determined how long it would take a normal designer to do one of these tasks, and we rated that for quality as well. We had a team of experts rating the output, and then comparing that and giving similar ratings to all the outputs from the various AI tools. And then we tested them across a range of different tools as well: ChatGPT, Claude, Copilot and the learning-specific AI tool Hive [?]. I won't go through all of it, but I'll take you through a couple of the use cases which had some quite interesting results.

Test question generation

[00:07:29] The first one was test question generation. Has anyone in the audience completed a mandatory training module before in their business? Yeah, everyone just loves those. Lots of fun. Our team at Barclays is responsible for most of those globally. We put out hundreds of those every year. It's a really key part of the output and the outcomes from our team. And a really important part of that is the test questions at the end, which can be a pain for everyone to do. But so much of the training that we produce at Barclays is to answer specific regulation. Banking and finance is a highly regulated environment.

[00:08:03] Testing is usually part of that. We have to prove that people have gone through testing. Passing a test is one way that we can point to and say, okay, people have completed this, but it's also a really good way to show that people have engaged with the content. We're not always successful with this, but if the test questions are good enough, it's very hard for anyone to answer them and pass the test without actually having gone through and understood the content as well. So it's a really important part of our process, and for each module which our designers would produce, it generally takes the best part of a day, about six and a half hours, to create a really good set of questions and answers which can properly test those modules.

[00:08:37] What we found when we integrated AI as part of this process is that in each case we saved around four hours. And we actually saw an uptick in quality as well. Not a massive one, but an uptick in quality across the board. It was remarkably similar across all the different tools as well. Now, four hours might not sound like a lot, but over a year that's roughly 400 hours for my team, depending on an average year. And that's just one tiny little part of the process. The entire process for a module is usually a few months of work. We're constantly rolling, doing different ones at different times, but that's just one small part of the thing. In one small part of the process, we already saw really big benefits there. And there are lots of other ones where we're seeing similar gains as well.

Summarizing content, and why language models can't count

[00:09:18] But this was a really interesting one: summarizing content. Obviously we work with a huge amount of content, particularly when we're updating modules from year to year and quarter to quarter. A big part of the job is negotiating with the stakeholders on what needs to change. We might want to bring a new learning methodology in. We might want to improve some of the content, introduce something new. And at that point, we have to take what's already there and summarize it. It can often be a bit of a fraught process. People have different opinions, different views on how things should be summarized.

[00:09:42] We thought AI could really help with this. This would be great. We can just give it to them. It can automatically summarize the content down to the level that we want, and we can start with that draft, which would be so helpful and save us a lot of time. So we gave it to each of the tools. I think the example was we wanted to take a piece of copy which is 1,000 words and take it down to 800 words. We gave it to all the different tools. They came back and said, "Here you are." We read through it. It sounded great. This is perfect. This has worked really well.

[00:10:08] But then we did a word count and found that some of them were more words than before, some were drastically lower, some were around about the same. And every time we did it, even with the same tools, we came back with different numbers. We did a bit more digging. Now, this may be apparent to a few people here, but we found that large language models can't count. The way that they interpret data is they look at patterns. The way that they look at words and phrases and sentences, they don't look at them as individual words. And so at the moment, they're not capable of counting effectively.

[00:10:40] You'd imagine that as these things evolve over time, it's something that'll be reverse engineered or incorporated. I'm sure in the future we'll be able to do this accurately, but right now, what we thought would be one of our most useful use cases is inconclusive. But I think the really important point with this one is that in this case it was inconclusive: the AI was fundamentally not able to deliver the outcome that we wanted, but it said that it did. In each case it came back very confidently, very happily, and said, here you are, perfect, we've reduced it down. Then when we questioned it and went back again, it still couldn't do it. But I think that's something interesting there, because without the checking we might have just taken that verbatim.

Bigger use cases and agentic AI

[00:11:15] These are some of the larger use cases that we had a look at within our area of learning design, which we thought were quite exciting. Virtual coaching: this could be something where, for all the learners, the thousands and thousands of learners across Barclays, we could be giving them specific virtual coaching. Media and asset creation as well. Designers spend a fair bit of their time creating assets and choosing images. If that could be done automatically and we're just checking, that could save us a huge amount of effort. Changing format as well. We're constantly changing documents and learning materials from an interactive module to a document, or backwards and forwards. If those kinds of things could be automated, that could be huge for us.

[00:11:57] And then larger things as well, like using AI to design a full curriculum that people could follow, or even getting into personalization. Could we actually be delivering specific learning for each person, incorporating not just the content that we create? We have a lot of partners, in LinkedIn Learning; we share all of their learning materials with all of our learners. If we could design curriculums that way, that would be fantastic. And then finally, taking it even further to generative UI as well, which I think is a really exciting concept, where the same content, the same material, could be presented differently to everyone in the way that's best for them. If you prefer videos, you might see videos. If you prefer documents, you might see documents. If you have certain needs, you need fonts in a certain size, you need colors in a certain way, that could always be presented to you.

[00:12:43] We just think there are huge opportunities here. But in almost every case within Barclays, there's a lot of work, a lot of development work. We have to change things, integrate with our existing systems and collaborate with other teams. These are the things we're going to be working through over the next year, but they're going to be much bigger. These aren't things that we can do on our own. And these last couple, the ones I was talking about there, will take increases in technology and increases in computing power as well to even get close to realizing them.

[00:13:12] This is what we can imagine happening now. But I believe that we've really only started touching the tip of the iceberg of what's going to be possible, because one thing that my team, my wider organization, started experimenting with is... oh, I seem to have lost some slides. Oh, okay, slides are back. We've started experimenting with using AI agents. There are some parts of the bank which are much further ahead than we are with this one. But this is one area we've just started to look at, and I think there's a huge amount of promise.

[00:13:51] I'm sure some people in the audience are very familiar with this already, but agentic AI, you can see the definition here from IBM: an artificial intelligence system that can accomplish a specific goal with limited supervision. It consists of AI agents, machine learning models that mimic human decision-making to solve problems in real time. So rather than you having to do everything and create all the prompts yourself, you can create an agent which can then work independently.

[00:14:12] I've seen some really fascinating talks around this area in the HR space, and there was a really exciting one from Josh Cavalier [?] talking about the implications for roles within learning and roles within HR, and how they're going to evolve as AI agents become more common. First, you can see here what I think most people think of when they think of agentic orchestration. A human is coming up with ideas, getting some agents working, and then they go and complete those tasks. But very quickly, apologies, very quickly you can see that the agents would start interacting with each other. A lot of things would be happening which don't need human supervision. They don't need humans to be involved.

[00:14:53] And then the end result of that is you could get to a situation where you could have a huge number of AI agents working independently and delivering on a huge number of tasks, with humans only really involved to set the parameters at the start, to set the guardrails, to set the area in which we all want to work. This is obviously really exciting, and you can see that this could be the way that departments could look in the future, maybe even whole companies. Something is set up at the start, then the AI agents are actually able to deliver everything. You can see why this is of so much interest to people in HR and learning, because this is going to have a huge impact on people, and how do we navigate people through this change? It's something that really is at the forefront of everything we think about when we think about AI.

How fast to move, and how big to bet

[00:15:38] But one of the first questions everyone asks when we talk about this is: when's this going to happen? That was one of the first questions in the talks that I saw. When is this going to be widespread? When is this going to be affecting my company and my business? And no one has a really clear answer. People are very confidently talking about this future, but they can't tell you exactly when it's going to happen, because there are so many things that have to happen. There are so many things that have to be improved before this can be trusted and before this can be widespread.

[00:16:05] This very morning on LinkedIn I was reading an interesting story about a test where some academics created a whole virtual company where every role in the company was an AI agent, and then they gave it business problems to solve. What they found is that the AI agents immediately set about arguing with each other, talking over each other, trying to do each other's jobs, and the end result was that most things weren't done. I think their maximum task completion rate was 24%, which is obviously worse than your worst intern at the moment. But it's on the way and it will happen; no one can tell you quite when that's going to happen yet.

[00:16:43] And I think that's one of the key problems here. One of the key things we have to think about is how fast should companies be moving? How big should their bets on AI be? I'm sure all of us have seen it, probably in your company or with some of the suppliers you work with: AI is probably already implemented. There are AI features available. Almost everyone we work with is proudly offering AI features. In my experience, some of those are useful, but a huge amount aren't, because I think people, in the rush to incorporate AI, haven't really been thinking about whether they're actually solving a problem for their users. Are they actually solving a problem for their business? AI by itself is not going to deliver you huge gains.

Apple Intelligence and the risk of big bets

[00:17:20] When AI first really started picking up steam over the last couple of years, obviously led by all the advances by OpenAI, all the other big tech companies pretty quickly followed suit. They started changing strategy, devoting huge amounts of budget, and bringing out big, impressive products which would be some of their flagships, showing how they were delivering AI and were leaders in AI, with one big exception. There was one big tech company that was quite quiet for quite a long time. And then in September last year, Apple announced Apple Intelligence.

[00:17:51] This was going to be a real step change and quite different from the other ones. Rather than just one product, Apple Intelligence would sit across everything that Apple does. Really, what it was was a grab bag of different features. Some of them we already knew, some everyone was already familiar with, some were just small upgrades. But one of the most exciting ones was Apple Intelligence incorporated into Siri, because that would allow Apple to do things that other companies couldn't do. You could be asking Siri questions, and it could access all the data across all your texts, emails, all the different things within the ecosystem, knowledge which other companies don't have, and it could give you really useful information. It could join those dots and close those gaps. It was really exciting, and I think it put Apple in a really exciting position.

[00:18:29] But as the rest of the year went on and into the next year, nothing was released. These features weren't there. And then Apple very quietly announced in March this year that Apple Intelligence in Siri had been delayed. It won't be available until 2026. What was interesting is they were actually showing demos of it last year at their conferences, which people only realized subsequently were concept videos. Now, Apple hasn't been showing concept videos live since the 80s. This was quite a change. But they'd obviously drastically underestimated the amount of time it would take them to develop this and develop it effectively.

[00:19:02] And I think there's a learning there for everyone: you can make a big bet, but there's no guarantee that you'll be able to deliver it in the time frame that you want. Companies can spend a lot of time and a lot of effort and a lot of budget on their AI solutions and have very little to show for it. This is a survey from the learning design space again, with learning professionals thinking about what they expect to be the biggest impact from incorporating AI. The top two make a lot of sense. That's what I've been talking about as well: creating learning content faster, efficiency gains. Those are the obvious things.

[00:19:37] The third one's really interesting, because people are talking about personalization and adaptive learning. Now, I know from bitter experience in other companies that personalization can be very hard to deliver. It's not as simple as just enabling the personalization opportunities from AI. There are huge integration and development challenges to make it actually work within your existing infrastructure. I think a lot of companies could be making a big bet and spending a lot of time with very little to show for it if they're immediately leaping towards these big bets.

The bigger risk of doing nothing

[00:20:03] But there could be bigger risks in doing nothing. In our area of learning, I took you through some of the things that we're doing right now. We're just focusing on content updates at the moment, making our content creation process a lot faster. But there is a really inherent risk in just focusing on efficiency and content updates for now, because AI is constantly making the process of content creation faster and cheaper with every advance. Content creation is going to become almost free at some time in the indeterminate future. And if a company has just been spending all their time becoming better and better at content creation, they could very quickly become redundant.

[00:20:40] So in learning, one of the things that we have to be thinking about is that we can't just be getting better and better at producing learning content and sharing it with everyone. We have to actually be thinking about how we're going to be landing this AI change, helping people and organizations adapt, and actually helping organizations thrive in this time of huge change that's coming. Just using AI to make content faster is a very dangerous gamble. I think companies that just sit still and do nothing, or just do the same as they've always done but a bit faster, could find that they're not going to be there at some point soon.

[00:21:10] But at Barclays, we're doing a lot right now. This was a Barclays hackathon that I attended last year, with several hundred people working here. These were just the teams that made it; a lot of people applied to get there. And this was just one site as well. This was in London, but we had this going at all the major global sites all across the world. There were a lot of really impressive teams delivering fantastic solutions, all within 24 hours, incorporating AI and solving business problems.

[00:21:40] We had only 24 hours and were able to make amazing progress and deliver something we were really happy with. Sadly, we didn't win. There were some very, very impressive teams who are doing this full-time, rather than us HR and learning professionals, where this is a bit of a sideline. But what we were able to do is build a working prototype of a virtual coaching agent. This is one of the use cases that we thought would have real promise, and we actually tested it with multiple users throughout this process, multiple leaders in the business, and it worked perfectly. Our initial use case was coaching leaders on how to have difficult conversations with their colleagues, and the feedback across the board said that this worked really well.

[00:22:16] Now, if we could roll this out and implement it across Barclays, that could be saving the business tens of millions each year, because coaching is a big business throughout the business at the moment. If we could do that all virtually, that would give you huge savings. This was done last year, a proof of concept which worked quite well, but it's not even close to going live. And do you know why that is? It's all about risk.

Risk, control and moving towards augmentation

[00:22:37] At Barclays, risk is something that we take incredibly seriously, and I think most financial institutions are the same. We're not in a position where we can just put something live and hope for the best, and gather some information like a small company might be able to do, because the implications for a big financial organization like Barclays are too large. And it's not just in banking. I think this quote is from a few years ago, but I think it's become even more relevant now.

[00:23:05] Not that long ago, disruption was seen as a really positive thing. If you came into an industry, disrupted it, moved quite fast and broke a few things, the outcomes would be better. But in so many industries now we've seen the results of disruption without forethought about the impacts it could have. And I think that businesses and the wider public are more aware now and less tolerant of big mistakes. As we start to integrate AI, especially in big businesses, what we're going to see is some bigger mistakes being made. So it's really important to understand these risks and be aware of them.

[00:23:41] Risk and control is a really big program of learning that we have at Barclays. It's a critical school for the entire bank. Everyone that comes into Barclays has to be trained in risk and control. You can see this is just a small amount of the modules that we produce, training the business in risk and all the different types of risk that we have there. Everyone in the bank has to understand risks, identify them, and then, when they have identified them, put controls in place. It's something that we have to do. It's really important for us because the alternatives can be catastrophic if we don't. In regulated industries like banking there can be fines in the hundreds of millions, reputational loss and business loss with even small mistakes, if we allow a risk to go unchecked.

[00:24:25] Why this is interesting is that in all our testing in the learning design, content and design space, we're yet to see a use case which we would be happy to put forward with no risk. What that does is really affect, in our case, automation. Coming back to the start, when I was talking about the fears that everyone has of AI taking our jobs: in our area we're yet to see something that we could happily automate with no risk. The bank does tolerate some risk; it has to in certain areas. But in our position we wouldn't be happy to go out with known risks without them being mitigated.

[00:25:01] So what we really need to do is move towards augmentation. This is where we can still reap the huge opportunities from AI and still use it as part of our process, but humans remain involved. Humans are there to take things forward. We don't do the rote work at the start, but we can still deliver great outcomes at the end. That means we can be aware of the risks that we found bringing AI in, but humans can be that control. We can be the ones that actually help to realize the huge opportunities from AI.

[00:25:33] I think I'm just about out of time, but here are some quick takeaways. Measurable gains: the AI opportunities are vast, but must be relevant. Smaller process improvements are often more feasible and can be measured and proven. It's really easy to fall into the trap of thinking too big and too fast at the start. Things will be slower than you think. But I've lost my slides. Anyway, those are my key takeaways. Thank you very much, everyone.

Q&A

[00:26:05] Mike: Oh, we jumped ahead to the feedback as well. I think I've lost control of my slides. I wonder if it's been automated. I think it has. Yeah.

[00:26:14] Host: You hit the nail on the head. Throughout the day there have been a lot of concerns about automation versus augmentation. We even had some stats, I won't try to paraphrase them, that said that most people are going for augmentation in their personal workflow, probably because it's not good enough, but also probably because of fear. There are two shades of fear. I'm not asking you to solve the universe here, but not everyone works in a bank, where risk is so tangible, where there's actual government bureaucracy helping reinforce that. As someone who thinks about this a lot, what strategies or anecdotes can you share with people who are probably worried about what somebody else mentioned earlier: the email or the Slack message from the CEO or founder saying, that's it, fire all these people, use AI for this thing? And the fear that happens there.

[00:27:00] Mike: Well, the first thing I'd say is that this happens in banking a lot as well. With all the testing that we've done, we've proven these gains, and we've said repeatedly that we want to use these gains to bring things into a workflow which you don't have time for right now. But periodically I still get the question: you're showing all these hours that you're saving, can we start cutting down a bit? Do you need all these people? It's a constant thing that happens. But then we have to go back and say, look, we're not in a position to automate. This is just making our work faster.

[00:27:29] In answer to the question, the key thing I come back to, and it comes back to one of my other passions as well, is that these arguments and pushing back on these things are always best done with data. If you have concerns that this won't work, test that. Do some user research. Potentially it's something that might work with quantitative research, A/B testing and those kinds of things. In those kinds of conversations, when people are coming and saying AI can save us this much, we now need to incorporate it, if you know that that's going to cause problems, prove those problems. What I would say is, well, you can never do too much user research, in my mind.

[00:28:02] Host: Okay. It's an interesting one. I'm not sure that this actually applies at Barclays with these kinds of learning modules, but any thoughts about making sure people actually learn things and aren't just using AI to pass the test?

[00:28:16] Mike: Well, it's a constant problem, even internally. Absolutely. I was talking with colleagues about it this morning, because everyone's busy. I'm sure everyone in the room is too busy for learning. I think a lot of the people in our organization, our learning designers, would love to think that people will spend the full amount of time, but people skip through and they go as fast as possible. For us, and we have these arguments constantly, it's always about trying to make things more engaging. Really what we want to do is make people want to learn.

[00:28:48] All the arguments we have are about trying to turn quite often quite dry material into something that's engaging. What's the hook? My designers are real experts in learning methodologies. They're not just thinking about the interactions at the end. They're thinking about how we can optimize this content. How can we bring people in? How can we structure it in a way where people, almost against their best judgment, dive in and engage with the learning? So I think it should be all about carrot and not stick.

[00:29:18] Host: In your top two use cases, if I remember correctly, one of the first ones was creating content and one of the other ones was personalization.

[00:29:25] Mike: Yeah.

[00:29:26] Host: We've actually been using AI for a long time to personalize content to us, and it's led to the destruction of the world in social media. Well, now I'm exaggerating, but people are getting fed the information they want. Is that a concern at all, even in a learning space for a bank, that people are going to start to just be reinforced in the things they believe? Which I think connects with the handling bias question as well.

[00:29:47] Mike: Well, I think social media is probably quite a different context to learning for your business, where we can actually understand this is someone's role, this is who they are, this is how they like to learn. And actually, if we could personalize this, that means that things could be more efficient and more relevant to them. I think the big difference there is the context. We actually have clear requirements. We know what people need to learn for their roles. And if we can have that information about how they'd like to learn, then I think we can get to some really exciting places. So I understand the trepidation, but I think it's quite a different context. I'm not sure in social media everyone's actually signed up for the amount of personalization. Well, they have officially, but have they agreed? Is this the experience they actually want?

[00:30:28] Host: It's also probably an argument in favor of augmentation over automation. The learning objectives are always still defined by humans, although that's probably going to become the most expensive thing out there someday: something made by a person.

[00:30:40] Mike: Exactly.

[00:30:40] Host: Thank you so much. Round of applause.