Human-Centered AI Across Standard Chartered

Emily Yang at UXDX EMEA. Video: https://youtu.be/GMv6K-kJ76E

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.

Standard Chartered: 172 years old and experimenting with AI

[00:00:12] Wow, what an energizer. I hope you're all wide awake. Woohoo. I'm going to share a little bit about human-centered AI today. But first, a quick hand poll. How many of you have heard of Standard Chartered? Raise your hand. Okay, fantastic. Human-centered AI? Okay, fewer hands. I'm being cheeky here. Me? Fantastic.

[00:00:38] Let's start with Standard Chartered. Here are the great innovations of our modern history. You have the steam engine, you have the telephone, as well as, obviously, the Macintosh, Apple. The question is, where does Standard Chartered's founding year fall on this timeline? A small moment to think. A's in the room? Any A's? Okay, a couple of brave souls. B? Any B's? Okay, a lot more B's. C's? Yeah, okay. I didn't know about this until I decided to put this question into the deck, so I'm learning a little bit about my company's history. It is in fact 172 years old this year. Very old.

[00:01:30] So, 172 years old, 85,000 [?] employees. We are in over 50 markets globally, mainly in the developing countries, so the Middle East, Asia, Africa and so on. We do have retail banking as well as consumer banking, with headquarters in London; it was also founded in London. I'm from Singapore, by the way, and it's also the bank that started the first ATM in Singapore back in those days. Very fun trivia.

[00:02:00] And what is an organization that is 170-plus years old doing in the age of AI? We are also experimenting. Towards the tail end of April, the first-generation AI use case that landed in production last year, from the department of work that I led, won SAP's awards for AI excellence and people experience. That's definitely encouragement for our human-centered AI methodology. So yes, we are experimenting in the age of AI too.

Emily's role and journey into human-centered AI

[00:02:32] This is me. I'm Emily. I head up the human-centered AI department as part of Strategy and Talent at Standard Chartered. I'm part of the Responsible AI Council, where we look at AI use cases, and also part of the AI learning hub. I'm the founder as well as the product owner of our AI literacy program within.

[00:02:55] Where do I sit in the organization? We're part of digital strategy and innovation, as part of our COO team. Essentially what that means is that all of the HR functions, all of the cabin [?] functions, the SEM [?] functions, group strategy, our property, if they want to develop AI, build AI, roll out AI, or have ideas about AI, I look after it from an end-to-end perspective, from idea inception to after it goes live, and all the ongoing monitoring with it. I do a lot of strategic advisory into the C-suite as well. I write people's talking points when they go out to WEF Davos, just a couple of things today.

[00:03:38] My journey into human-centered AI: I started out in biochemistry, so in the STEM fields, and then eventually ventured into natural language and human-computer interaction. I did my studies in academia and then went on to a startup before diving into user research. I got a little bit bored of that and went into corporate venturing, building startups for Fortune 1000 companies, then went back into research, and this time around it was human-centered AI. Some of the papers I've written in the past were on perceived emotional intelligence in virtual agents, back in those days, still relevant today in the age of AI agents and agentic AI, which we heard about earlier today.

[00:04:23] And I do understand the power of UX research. In 2023 Standard Chartered rolled out parental leave globally for both moms and dads. It started from a piece of research that I initiated way back in 2019, 2020, around what the experience of maternity and paternity leave looks like for our colleagues within. That rolled upwards, and it definitely made me understand the power of user experience research and the impact it can have.

[00:04:51] And here we are today when it comes to human-centered AI. I mentioned SAP, and I've also mentioned our AI learning hub within. These are some of the works that we have done, the fruits of the labor, quote unquote, of the human-centered AI department.

Amara's law

[00:05:07] A couple of things to share. I do represent myself; some of these perspectives come from a human-centered AI point of view, not necessarily Standard Chartered's. I'm also very cognizant about what I share today because of Amara's law. How many of you have heard of Amara's law? Okay, a quick recap. It means we often overestimate the impact of a vanguard technology in the short term but underestimate its impact in the longer term. So take everything that I say today with a grain of salt, think about how you might apply it to what you're doing today, and keep an eye out for how the future will change based on all the technological advancement, the laws and regulations around the world, and changing human behavioral dynamics.

What human-centered AI is

[00:05:55] Human-centered AI. These are some of the different spellings and abbreviations. How many of you think that human-centered AI equates to human-centered design? The lights are pretty blinding. Okay. How many of you think that human-centered AI equates to responsible AI? The lights are still pretty blinding. Human-centered AI equates to human resourcing? Okay, the lights are better, but I don't see hands anymore. What about human-centered AI equating to just AI? Yes, okay, I see a couple of hands.

[00:06:36] If you raised your hand for any of those, you are correct. It's just incomplete. From my perspective, as one of the first professionals, if not the first professional, trying to land this for the industry, I see it as a rising field where research and practice are coming together. We're looking at the intention of human identity and experience, humans flourishing in the age of AI, as well as sustainability, as very important when we look at the development, the creation and the adoption of AI products and solutions today. It operates at the intersection of AI ethics and AI tech, but most importantly it brings in all of the human perspectives.

[00:07:18] Typically there are two hats we need to wear, and coming from an HR function I definitely echo that. The first is how can we empower the rest of the organization to adopt AI at speed, at scale and safely? The other aspect we need to consider is how can we adopt AI in our own daily work and roles? This is very similar to how we, in our roles as user experience researchers and designers, adopt some of the GenAI or AI-powered tooling to help us become more productive and more creative today. But at the same time, how can we in our discipline, with all of our tools, help the rest of the organization roll out AI products and solutions that are safe, creative and adaptive for people's use?

[00:08:03] Here are the ten human factors that I believe are extremely critical. There are varying sizes of circles, because I believe there are varying degrees of maturity in terms of conversation, industry acceptance and putting it into documented practice today. Obviously responsible AI as a field, AI ethics and humanities, is one of the rising fields since last year, primarily driven by, or Tesla [?], instigated by the EU AI Act rolled out last year. I want to go over all of these ten factors very quickly, to give everyone a taste of what they mean, through a case study.

The case: therapy and companionship

[00:08:48] Let's define the case. What is the top GenAI use case in 2025 according to Harvard Business Review? Is it A, generating ideas; B, therapy and companionship; C, finding purpose; D, enhancing learning; or E, fun and nonsense? I'll give everyone a moment to think, because I think I see some movement in the crowd. Okay, any A's in the crowd? Oh, a couple of A's. Lovely. B's? Okay, a lot more B's. C? Okay, a couple. D? Ooh, B and D are starting to become popular choices. E, fun and nonsense? Oh, I definitely use that. I just used that yesterday.

[00:09:49] According to Harvard Business Review, the top use case is in fact therapy and companionship. Woohoo, great work. But we can also see finding purpose coming up as a close second, at least according to my short list. These are the data between 2024 and 2025. We're going to bear this in mind as we walk through the ten factors.

[00:10:12] Here are some examples of different types of virtual therapists, virtual coaches, virtual psychologists and GenAI companions. Some of these are probably very well known, like Character.AI, which made the news last year. There's also Replika. If you're in the enterprise you've probably heard of Azro [?], which I think is marketing quite heavily in London these days. And then you have Skillsoft with virtual coaching, a difficult conversation simulator and so on. There's a lot out there.

AI ethics: fairness, transparency and accountability

[00:10:47] Let's take a moment to think about virtual therapists, psychologists and so on from the perspective of AI ethics and humanities. How many of you have worked with or know of responsible AI and AI ethics? Yes, okay, a couple of folks in the crowd. Fantastic. To simplify this for today's talk, let's consider three perspectives: fairness, transparency and accountability.

[00:11:27] In terms of fairness, what are some of the questions we want to ask ourselves about whether this virtual therapist or virtual psychologist is demonstrating, in so much as possible, minimization of bias from a design perspective? Oh, I love my microphone, it's coming in and out. If you're at the back and you can't hear me, please give me an indicator by waving your hand. Okay. One of the key aspects is looking into the gender of the virtual agent.

[00:12:02] Typically, if we look back to 2015, 2016, the virtual agents that we started interacting with were always cast as a female character, a female voice, a female avatar and so on. When we are designing these systems and interacting with these systems, do we see a pattern of that happening? Because according to academic literature, when we look at gender and ethnicity and cross that with occupation, there are inherent biases in terms of stereotypes coming through, because AI a lot of the time is a mirror of what's happening in our society. So thinking about the gender of this virtual agent: are we unconsciously, subconsciously perpetuating biases that we want to be cognizant of? That's one aspect. And by the way, when we're thinking about these perspectives, there are a multitude of questions we can ask. These are just sample questions.

[00:12:58] Transparency. With the power of GenAI, images are a lot of the time very authentic looking. Do our users know whether they're interacting with a generative AI-driven virtual agent or with a real person? And if they do, do we have their consent? Do people know whether they're interacting with a real person or a virtual agent? This is quite an important question to think about, because this year, across industries, we are experimenting with AI agents and agentic AI systems.

[00:13:42] And then my favorite question: who's accountable when something goes wrong? Ah, will things go wrong? Yes, there's a probability that things could go wrong if we don't have the right guardrails in place, if we don't think about user centricity, if we don't think about human centricity. That's one of many examples.

Education and AI literacy

[00:14:06] Next, education. Why do we need education when it comes to rolling out some of these agentic AI systems? Who will be using them? Do they know how to use them? Do they know how to use them responsibly? Are they aware of some of the considerations that we have put in as designers or creators? If they notice something that isn't quite aligned, how do they share feedback? What is critical here is the responsible use of AI.

[00:14:38] Next, once we create these AI products, they are going to be operated on a day-to-day basis. Someone needs to look after them, and these are likely not going to be developers. So who will be governing them and who will be operating them? We really need to think about education, or in this case AI literacy, because the technical understanding of AI systems will vary for the end users of our AI solutions versus those who are governing them and those who are operating them.

[00:15:16] Those who are using them will need to understand, for instance, what responsible use of this particular system is. Those who are governing them may need to think about what explainability metrics we need to look into, what fairness metrics we need to look into, and how we monitor them on an ongoing basis. For those who are operating them: what are some of the signals from our end users indicating that we might need to go back to the product or the technical team and share the feedback we're noticing on the ground, so that they can retrain their algorithm, or so we can redesign the interface? Things to think about.

Workforce and philosophy

[00:15:58] Workforce. Why is workforce an important consideration? What roles would be impacted by AI? In this case, AI therapists, AI psychologists, AI coaches. Next question: would the roles be displaced or would they be augmented? And if they are displaced or replaced, what does it mean for an organization's workforce transformation? You can substitute AI therapists and psychologists with any other type of AI solution that you're thinking about.

[00:16:34] Then there's the part at an individual level, in terms of philosophy. Why is that? For those who are impacted, in this case human coaches, therapists and psychologists: can they keep up? Will they have a job in the age of AI? What's going to become of their role? Can they stay relevant? And if they stay relevant, can they be more competitive? And if they can become competitive, how do they stay ahead? It is this circle of anxiety.

[00:17:08] And on an individual level: what is the meaning of my work life? How do I attain self-actualization at work? Will I find meaning at work? Will I find joy at work? You can substitute coaches, therapists and psychologists with any other roles impacted by a GenAI system. Here's a quote from our research within indicating exactly that: people are actively thinking about what the meaning of work is in the age of GenAI.

Culture and psychology

[00:17:42] Culture. In the case of virtual psychologists, virtual therapists and virtual coaches, a lot of the time it's an interaction with a GenAI-driven virtual agent. Users type in something, and the GenAI system reinforces or tells the user, this is something you can improve on, and overall did you do great or not so great, and why. This is an example of what the end result, quote unquote, could look like after the interaction.

[00:18:15] But here's the real question: by whose values are we reinforcing our end users' responses with these GenAI systems? This is a question that is gaining traction in the age of agentic AI, primarily because we potentially have a values alignment, or misalignment, problem at hand. Is your culture represented? Is your diversity represented? Is your voice represented? And how well, in the context of this particular use case, will the advice or positive reinforcement translate cross-culturally? We know, for instance, that Japanese culture versus US culture could be different, and that's just the dimension of culture, not even taking into account generation, people's different belief systems and so on. So culture is extremely important.

[00:19:14] Cyberpsychology, or psychology in general. Understanding what would motivate people to use these agentic coaches, psychologists and therapists in the first place. Do we know them? Do we know why? Or is it purely driven by business? Things to think about. And if we do roll them out, what would be the desired impacts, as well as potential unintended consequences? And if people start using virtual psychologists, virtual coaches and virtual therapists in the longer term, what would be the longer-term impact on you and me as individuals? Big questions.

Trust and experience

[00:20:05] Then we move on to society: trust. Do our end users trust the advice that's provided by some of these agentic coaches, psychologists and therapists? And if you are creating something like this and trying to roll it out within, how do product teams work towards ensuring trust between end users and this particular product? In other words, will this product erode the baseline of trust that end users have today with a human therapist, psychologist or coach?

[00:20:50] This is something I shared last year with the team that was working on the WEF report. This quote is mine: the only way we can counter fear and uncertainty is with trust. That's super important, something to really think about as we're designing all of these AI solutions.

[00:21:12] Experience. When do end users interact with this product? How do they want to interact with it? Do they want to talk to it? Do they want to type to it? Do they want to see a virtual avatar rendered to mimic human interaction? Are they using it during their commute to work? Are they using it in a private, quiet room, which is where conversations with therapists or counselors usually happen in human interaction? Which of these is it? And if they want an image, how realistic? If it's realistic enough, will it cross over into the uncanny valley? And if it does, will it impact user adoption and user trust?

[00:22:07] And coming back to it, will this erode the human touch or the human experience, or in general lead to a loss of empathy? I'm not saying this without reason, because our internal research last year also showed that people within are worried that GenAI systems will erode the amount of human trust, or human touch, we have with one another today.

Sustainability: it costs a lot to be polite

[00:22:34] Sustainability, fun topic. How would rolling out or developing a product like this impact the organization's consumption of energy? Are end users cognizant about their usage? Because it costs a lot to be polite. How many of you use please and thank you when you're interacting with your GenAI systems, Gemini, ChatGPT, Claude, Perplexity? Please and thank yous? Oh, come on, people, there must be a lot more than this. Thank you, thank you for being self-aware here.

[00:23:20] I don't know if you can read this, but it says that a report from the Electric Power Research Institute indicates that if we ask ChatGPT a question, it's going to cost ten times the amount of electricity we use if we just query Google. So when we ask questions, it consumes energy. When ChatGPT and the like reply to us, it also consumes energy. Here's another way of looking at it. It takes about 40 to 50 milliliters for ChatGPT to come back with the phrase "you're welcome." How much is 40 ml? A small bottle of Dior perfume. Yes. That's costly, isn't it?

[00:24:06] So are people cognizant about their usage of GenAI systems? Because in conversations, especially with virtual therapists and coaches, there's a tendency to say please and thank you, and for it to come back with "you're welcome" and all those other amazing pleasantries. And if energy consumption is a little bit difficult for us to wrap our heads around, there was also a very interesting exchange on Twitter not too long ago about how much money OpenAI is spending dealing with all these please and thank yous, and Sam Altman mentioned tens of millions of dollars, probably. Fun times.

Economy and return on investment

[00:24:45] Coming back to economy. What is the return on investment of an AI solution like this? What are the meaningful attributes that we need to measure when it comes to an agentic coach, a therapist or a psychologist? What are we baselining against? Do we want to keep track of near-term or longer-term success metrics? What about indirect costs? AI literacy for those who are operating these systems, for those who are using them, for those who are governing them. And the gains: a lot of the time, if your well-being is great and you're happy, you're more productive at work, but you're also more creative, happier, more satisfied, more engaged. How do you measure some of those? It's very indirect. And tying everything back together, how do we consider the sustainability trade-offs?

Ten factors, ecosystem thinking and humans in the loop

[00:25:49] What I've just walked everyone through is how we consider the ten human factors in AI solutioning. Whether you're building an AI solution, rolling it out, thinking about adopting it, thinking about using it for yourself or your kids, teaching people how to use it, or thinking about design solutions, these are the ten core human factors to think about and consider. What we have really done is practice our critical thinking and our ecosystem thinking, as we move towards a conscientious, informed and realistic optimism when adopting vanguard technologies like GenAI.

[00:26:31] What I hope everyone can take away from this is that AI is an innovation that is happening for us instead of to us. In order to really amplify its power for us, we need to make sure that we are actively involved and engaging our stakeholders to think about some of these potentially difficult questions.

[00:26:53] I hope you have gathered from these ten factors that human-centered AI is not just human-centered design. It's not just responsible AI. It's not just HR, and it's not just technology. In fact, according to McKinsey, technology is only about 30% of the solution when it comes to AI. The other 70% is really about people. This is human-centered AI, where we focus on human identity and experience, making sure that humans flourish and that there is a sustainable future for us in the age of AI. I'm Emily, and I head up human-centered AI at Standard Chartered. It's always about humans in the lead [?] and machines in the loop, not the other way around. All right, thank you.

Q&A

[00:27:45] Host: I love that. Amazing, well done. I enjoyed the hesitancy you had at the end there to say thank you. Were you worried it was going to cost 40 to 50 milliliters? To be honest, I'm now very conscious of the environmental footprint of me trying to train ChatGPT, when I say "cheers mate" to it, to come back with "all good, fella." I think it's maybe worth it, but we'll see how it goes. Let's dive into the proper thing. So interesting. I would never have guessed that companionship and therapy was the biggest use case of GenAI. So here we go, a difficult question to start us off. Do you think it's dangerous for computers to pretend to have emotions?

[00:28:24] Emily: How many of you think that it is not dangerous for computers to pretend to have emotions? Lovely, a couple of people. I wonder if it's more dangerous for machines to have them, or for us to anthropomorphize that machines will have them. Is it perceived emotions that a silicon-based system intelligence, quote unquote, is eliciting, and how we're responding to it? Or is it that it really does have them? This is where I'll pull in Amara's law. The real answer is I don't know. I really don't.

[00:29:12] But what we can be cognizant of, in the user experience community, is how we frame some of these measures and experiments. Is it about the fact that AI has values, so values in AI systems, emotional intelligence in AI systems, empathy in AI systems? Or what we're really looking at is the perceived values alignment in AI systems, the perceived sense of empathy from agentic AI, the perceived degree of emotional intelligence elicited by these machines. I think that nuance is super critical for us to discern.

[00:29:48] Host: I agree. I think it's interesting from the therapeutic standpoint that a lot of therapy is based around the therapist acting as almost a blank slate of unconditional positive regard. Most therapists don't put too much emotion into how they describe your problem back to you. So maybe it's actually a good thing.

[00:30:07] Emily: That could be.

[00:30:08] Host: It's maybe not as dangerous as you think. A lot of great questions. Are we going to answer any of these, or should we just move on? We'll move on to the next question. How do you reconcile the concept of human-centered AI with, fair challenge, the reality that most AI systems are being built to aid business goals or oriented around profit?

[00:30:36] Emily: I love this question because this is the exact question I grapple with all the time. When I look at the people seated at the decision table making decisions about AI, whether it's AI strategy, adoption, or how it lands in people's hands, who's there? Our governance folks, who represent legal and compliance. Our business folks; they carry the money. Our technology folks, because they carry the art of the possible, or feasible. But the group that is really left out, quote unquote, at the moment is the user experience teams. We carry the perspectives of people. Our leadership needs to be informed of people's perspectives when they are making these decisions. They don't have it, so how do they come up with informed decision-making?

[00:31:28] Which goes back to something I mentioned before. I agree that user experience today might be too constrained to just pixel-perfect production, where potentially we have way more impact, and I believe we do, operating at the level of strategy, where insights can be channeled up to our C-suite and our board in so much as possible.

[00:31:48] Host: Good. I'll finish there; we're slightly over time. Apologies to the number of people saying why did I skip that question. It's my human, not AI, bias towards the practical questions rather than the deeper philosophical questions, with only a minute left on the clock. Although I'm sure, Emily, you would be happy to answer any of those questions in person on a deeper philosophical level, or via LinkedIn. Or am I throwing you under the bus there?

[00:32:09] Emily: Absolutely, feel free to reach out if you want to have a conversation about human-centered AI. This is a very new field. I hope it gives people a little bit more inspiration around the ecosystem thinking that we really need to bring to the table when we talk about AI.

[00:32:22] Host: Amazing. Emily Yang, thank you very much.

[00:32:25] Emily: Thank you so much for having me.