Learning Library

← Back to Library

Trust and Transparency in AI Agents

Key Points

  • “AI in Action” is a new IBM series that dives into what generative AI can and can’t do, how it’s built responsibly, and how it solves real‑world business problems.
  • Trust and transparency are the foundation of virtual customer‑service agents; users must be told when they’re talking to AI rather than a human.
  • Effective virtual agents need to be empathetic, personalized, and timely while safeguarding the accuracy and security of the data they handle.
  • Customers feel deceived when a chatbot pretends to be human and then fails abruptly, so honest disclosure of AI assistance improves user acceptance.
  • To be useful, virtual agents must provide specific, context‑aware answers and offer smooth handoffs to human agents whenever the issue exceeds their capabilities.

Full Transcript

# Trust and Transparency in AI Agents **Source:** [https://www.youtube.com/watch?v=1FR5xIOknY4](https://www.youtube.com/watch?v=1FR5xIOknY4) **Duration:** 00:18:30 ## Summary - “AI in Action” is a new IBM series that dives into what generative AI can and can’t do, how it’s built responsibly, and how it solves real‑world business problems. - Trust and transparency are the foundation of virtual customer‑service agents; users must be told when they’re talking to AI rather than a human. - Effective virtual agents need to be empathetic, personalized, and timely while safeguarding the accuracy and security of the data they handle. - Customers feel deceived when a chatbot pretends to be human and then fails abruptly, so honest disclosure of AI assistance improves user acceptance. - To be useful, virtual agents must provide specific, context‑aware answers and offer smooth handoffs to human agents whenever the issue exceeds their capabilities. ## Sections - [00:00:00](https://www.youtube.com/watch?v=1FR5xIOknY4&t=0s) **Untitled Section** - ## Full Transcript
0:00representative that's me into a phone 0:03trying to talk to a human when I have a 0:05customer service issue but instead I'm 0:07only getting canned responses so how do 0:10we get a human on the line when we 0:11really need one or do we need one we're 0:14going to get to the bottom of it today 0:16on AI in action in this series we 0:19explore what generative AI can and can't 0:22do how it actually gets built 0:24responsible ways to put it into practice 0:26and the real world business problems and 0:28solutions will encounter along the way 0:31so welcome to AI in action brought to 0:34you by IBM I'm Albert Lawrence so let's 0:37talk about virtual customer service 0:39agents and how to actually get help 0:42today I'm joined by jeie Walters 0:44customer experience speaker trainer and 0:46podcast host jeie thanks for being here 0:49thank you so much for having me we're 0:50gonna have a good time my other guest is 0:52Morgan Carroll senior AI engineer at IBM 0:55what's up Morgan I am jazzed to be here 0:57oh I know you are I feel it now look I 1:00know you both are wondering why did I 1:02Choose You two for today's episode well 1:04it's because you both specialize in the 1:06future of customer service and virtual 1:08agent experiences big picture I want to 1:11explore trust is it time for us to give 1:14virtual agents a chance can they really 1:17be empathetic personal timely and answer 1:20the questions that we actually have I 1:22know we're asking for a lot but jeie 1:25what do you think I think it's time 1:27customers are expecting this they want 1:29to be seen and recognized for who they 1:32are and where they are on the journey 1:34however they interact with Our Brands 1:36and so now we have a chance to do that 1:39in a way that is automated and personal 1:42and all of those things but we have to 1:44consider the entire Journey at the same 1:46time I think it all comes back to trust 1:48we want to make sure that users know 1:50that the data that they're giving and 1:52receiving is accurate and secure if they 1:55are interacting with an artificial 1:57intelligence agent versus a human we 2:00should let them know that so that they 2:01don't think like okay this is aive human 2:03no this is artificial intelligence that 2:05is assisting us now Morgan you mentioned 2:08that it's important for humans to 2:10understand when they're speaking with a 2:11virtual assistant versus another human 2:13jeie why might that be the case can you 2:15give me an example we have heard over 2:17and over that customers really don't 2:19like it and it's not just customers it's 2:21all of us right we don't like to be 2:22deceived as humans and so it can feel 2:25very deceptive when we think we're 2:27having this very human conversation and 2:30then suddenly it basically says be Boop 2:32I can't do it and we realize that it's 2:34actually not a human that is breaking 2:37trust from the beginning instead of just 2:39being open and honest about the fact 2:41that you know you're you're talking to a 2:43virtual agent who's here to help you 2:45people are very open to that well when I 2:47am talking with a virtual agent then I 2:49guess one of my concerns is that I want 2:51to make sure that the answers that I'm 2:53given are not just broad generic answers 2:56I would prefer for them to actually be 2:57answers that pertain specifically to my 2:59my questions that address my needs how 3:01do you bridge that Gap with generative 3:03AI so we're going to connect to 3:05different data sources generative AI is 3:08going to take all of those different 3:09sources and sort of make an experience 3:11that is unique to the user without the 3:14user having to actually you know press 3:15zero go to a live agent have the live 3:17agent look at their information but 3:19generative AI can do that I think one of 3:21the powers of gen AI that we're seeing 3:24is that for instance in a contact center 3:26if somebody calls in suddenly that agent 3:29has the information just like you're 3:30talking about Morgan at their fingertips 3:32There Are Places still today where 3:34agents have to pull out big binders and 3:36try to find the information based on the 3:39information that they happen to get on 3:40the phone call and we're really seeing 3:43how we're connecting all that data in 3:45the back end to make it easier for the 3:46people who serve the customers and 3:48customers really are responding to that 3:50very positively um you just painted this 3:52picture of like humans pulling binders 3:55and that feels kind of archaic but I 3:57guess it's still happening it's still 3:59happening because don't all have the 4:01capability to put the data in one place 4:03and that's what's so powerful about what 4:05gen can do it can connect some of these 4:07different systems on the back end so we 4:10can actually provide a better front end 4:11for the customer okay so I want to get 4:13into some of that backend discussion 4:15then to now Morgan I'm very curious how 4:18do you end up really building that 4:20personalized generative AI solution then 4:22can you take us through that system yeah 4:24so we want to keep the customer in mind 4:26first and foremost put ourselves in 4:28their shoes and think okay okay how does 4:30this conver how should this conversation 4:31go that's going to be efficient and get 4:34straight to what they need so the first 4:36thing we do is sort of build like an 4:37outline of a dialogue just like a 4:40skeleton like okay this is where we want 4:41to start maybe and this is where we want 4:42to end and then we're going to come in 4:44and add all these different Integrations 4:46so maybe a database that has all the 4:47customer information so when they call 4:50in it's like hi Morgan and you know it's 4:53already got all my nobody has to pull 4:54out a binder and like search through all 4:56this stuff and it just makes it very 4:58simple so now that we've had a chance to 5:01talk some about the technical and 5:02engineering aspect and also really the 5:04front faceing customer service aspect 5:07how are these two things going to come 5:08together yeah so we want to think about 5:10the customer experience overall from 5:12beginning to end let's say we have 5:13someone who they need to get in touch 5:15with HR because they're about to go on 5:17maternity leave well HR only has five 5:19different options to select from and 5:21this doesn't fall into any of those 5:23categories what do they do obviously 5:24we're z00 representative we want to talk 5:27to a live person so these are the that 5:29we have to consider and this is 5:31something that generative AI can help us 5:34with okay yeah and I would just say 5:37think about the overall kind of strategy 5:40and needs of both the brand and 5:42organization and the customer that 5:44you're serving so for example when when 5:47email came out we got very excited about 5:49email and we sent all these emails and 5:51now we're kind of like we're done with 5:52email right and so I think we have to 5:55think strategically about how to best 5:57serve people and when we're talking 6:00about the customer Journey customers 6:01want choice you and I might not want to 6:04go through the same Journey you might 6:06have different needs than I do as far as 6:08if I'm in the car I might want to call 6:11and talk to a person when that is 6:13happening so providing those options for 6:16people even though it might be more 6:17efficient on our end to use the virtual 6:20agent can we give those choices along 6:22the way to the customer as well 6:24different customers might have different 6:26preferences regarding how they would 6:28like to be helped uh what are some of 6:29those preferences like well I mean think 6:31about the different circumstances that 6:33we go through in our daily lives I kind 6:36of get a chuckle when people say things 6:37like well our Digital customer journey 6:39is this and you know our in-person 6:42journey is this because sometimes we go 6:44into a store sometimes we want to shop 6:46digitally sometimes we might prefer a 6:49virtual agent in a chat and sometimes we 6:51might need to talk to somebody and call 6:53them right away and we don't necessarily 6:56fit into the boxes that they give us so 6:58that customer choice of how to be served 7:01and helped is just super super important 7:03so I'd like to Pivot to the agent 7:05experience they're obviously inundated 7:08with a number of emails and there's 7:09plenty that virtual assistants can 7:11automatically respond to but if they 7:14have perhaps something more difficult um 7:16they're getting a lot of these emails 7:18per day and they have to respond to 7:19these generative AI can actually assist 7:21them it's not going to necessarily 7:23respond for them it could but in this 7:25case we're thinking about how can we 7:27make the customer service agents job 7:29easier in gener of AI can help them 7:32draft an email or look up information 7:34stuff like that so even if a customer 7:36isn't speaking directly to a virtual 7:38agent generative AI is still playing a 7:40role with the human- to human contact 7:43exactly but then it makes me think about 7:45this idea of being a proactive versus 7:49reactive why is that important to the 7:51entire process and how might a customer 7:54feel and experience that on the front 7:56end well when you get right down to it 7:58if somebody is contact customer service 8:01something has probably gone wrong 8:03already so yes we want those moments to 8:06be efficient and effective but we also 8:09want to prevent those moments from 8:10happening because it will reduce cost 8:12for the brand and everybody else but it 8:15also makes the customer field more 8:17valued and cared for so if we can 8:18predict for instance Based on data and 8:21AI can look for patterns really well so 8:23maybe AI notices you know a lot of 8:25customers are returning this product 8:27after 3 months what's going on 8:29and then we can figure out well it's 8:31breaking after 3 months that's not great 8:34so let's figure out what is actually 8:36going on and proactively prevent that 8:38from happening or proactively reach out 8:41to the customer and say we think this 8:42might happen in 3 months here's a 8:44proactive solution for you that makes 8:46people feel really kind of recognized 8:48and seen in a way that when they have to 8:51call in when the onus is on the customer 8:53to really solve their problem they don't 8:55feel valued as much well Morgan I know 8:57we've been talking about the customer 9:00service experience but how exactly is an 9:02engineer helping to transform and to 9:06build the customer service experience 9:08okay so this is exactly what I do every 9:09day um think back back in the day when 9:12you would call a customer service line 9:14and you only have you know your five 9:15options whatever you have to push one 9:17two three and that's all you can do now 9:19we can actually use generative AI to 9:22take what the user is saying so actually 9:25I'm going to break it down on a sort of 9:26like a technical level we have our um 9:29our dialogue flow here right and then we 9:31have the user saying like I need help 9:33with this particular thing well if it's 9:35not sort of hardcoded into our answers 9:37we can actually go out call out to a 9:40large language model with some of that 9:42information and say like hey we need you 9:44to provide an answer to this user and 9:47just it's directly connected to the 9:48virtual assistant so we get that data 9:50back and then we just pass it right 9:52along to the user and that works most of 9:55the 9:56time this is why we have guardrails 9:58built into our virtual assistants 10:00because obviously all of this is really 10:01really new and in my experience you know 10:03in building these assistants sometimes 10:05we do get a wrong answer and we want to 10:06make sure that doesn't happen so we have 10:08various tools that um basically if a 10:11customer asks a question and we're not 10:13really really sure that it's correct 10:15then we may just say something like I'm 10:17sorry I can't help with that and that's 10:18where we need a life person well and 10:20this is why it all comes back to 10:22transparency and building that trust so 10:25that we're not presenting something that 10:28isn't quite true truthful to the 10:30customer they know the entire time that 10:32they're dealing with a virtual agent and 10:34that you know what sometimes there are 10:36places where we need to introduce a 10:38human jumping back into the mindset of 10:40being a customer right and if I call on 10:43into a virtual agent I'm wondering what 10:45are some of the common questions that I 10:48probably will be asking as a customer 10:50that a virtual agent can actually 10:53understand and give me valuable 10:54information about well I think right now 10:56this is probably the best use of virtual 10:58agents are these kind of repetitive very 11:01standard questions that customers ask so 11:04think about when you're at a hotel some 11:05of the questions you might have for the 11:07front desk where are the pool towels how 11:10do I get a wakeup call what do I do 11:12about you know these automatic trpes 11:14aren't working or whatever and once you 11:16start seeing those patterns that's 11:18something that a virtual system could 11:20handle very very easily a virtual chat 11:23that something that maybe is accessible 11:25right in the moment then on the flip 11:27side of that these big nuance 11:29complicated things that we run into as 11:31humans sometimes we need humans and we 11:33need both that empathy that a human can 11:35give and also just understanding you 11:38know what this isn't the standard 11:39question this is about something that 11:41went wrong that doesn't go wrong a lot 11:43we have to figure this out the customer 11:44is very upset so once we get to that 11:48place then we're going to start seeing 11:49new patterns and we're going to start 11:51realizing different ways that we can 11:53maybe update customers you know if 11:55they've already talked to a human and 11:57said this is going wrong maybe there's 12:00an automation through the virtual agent 12:01that says hey I'm your virtual agent I 12:03just want you to know we're working on 12:04this here's what to do next here's when 12:06you can expect to hear from us because 12:08that's another thing customers want is 12:10just that proactive communication of we 12:12heard you and we're working on it well I 12:14heard you say automatic drapes and now I 12:16want for my next hotel room to have 12:18those um but but I'm also thinking you 12:20know you gave me some great ideas in 12:22terms of in the future what are some 12:24ways that a virtual agent might be able 12:26to help and what are some of the common 12:28questions that customer might be calling 12:29on in from Morgan do you happen to have 12:31any thoughts in terms of some questions 12:33that you are anticipating for customers 12:36to be calling in for definitely what are 12:38your hours what time do you open um 12:40what's your phone number what is your 12:41address things like that just simple 12:43again repetitive tasks but I actually 12:45have a good example of one that we do 12:48need a human for um and this happened to 12:50me yesterday I got to my hotel and there 12:52was no electricity so the hotel actually 12:55does they do have a virtual assistant in 12:58their app but but obviously that's not 13:01really going to be useful if there's no 13:02electricity so go to the front and 13:04that's a really really simple example 13:06but these things happen in the future 13:09and now you've got me stuck on this 13:10hotel situation both of you um do you 13:13think that we'll be able to speak to a 13:15virtual assistant and just say open up 13:17the curtains that's happening in some 13:19hotels already they have Smart speakers 13:21they have tablets in the room they have 13:23all sorts of ways that we can interact 13:26and I think the other side of this that 13:28maybe we have have to start thinking 13:29about as well is how do we take care of 13:31those humans now who are dealing with 13:33these big complex issues all the time 13:36because in some ways these repetitive 13:37things they don't take as much empathy 13:39they don't take as much brain power and 13:41compassion so they're kind of like a 13:43little breather between these big 13:44complex things I think we really have to 13:47think about the world where we get to 13:49where we're asking humans to just carry 13:51a big heavy lift for every single thing 13:53that they do and look for ways to take 13:56care of them well let's stay on this 13:57train of thought is we're think about 13:59this world where humans don't need to do 14:01that why can't we do that right now 14:03what's holding businesses back from us 14:05getting into Genie's ideal future we're 14:07kind of already there um we have a 14:09concept Called Agent assist so our 14:12agents you know again they we don't want 14:13them to have to search through the 14:14binder they actually may have their own 14:16virtual assistant to help them with 14:18these complex customer inquiries so you 14:20know customer calls in hotel uh my 14:23drapes aren't working and the TV is 14:25broken and there's no water well I mean 14:29a virtual assistant probably could 14:30handle that but I think we're going to 14:31want a live person so they're like all 14:34right let's look up the information we 14:35got to call the water company and 14:36electrician you know etc etc and they 14:38have this virtual agent that can help 14:40and hopefully just put me in a new room 14:42yeah because at that point the very 14:45least wow okay so then that raises 14:49another concern for me which is data 14:51privacy and sharing so how do you 14:55reconcile that right how does data 14:57privacy and sharing end up impacting the 15:00personalization that can also make a 15:02customer feel really valued right so 15:04obviously we don't want someone stealing 15:05our credit card number but it all comes 15:07down to do we trust the model that we're 15:10using we have to make sure that we're 15:12open and transparent how is this model 15:14being trained where is the data where is 15:16my data going you know is it secure so 15:19let's think of um a client we have 15:21client information over here and then we 15:23have our virtual assistant up here in 15:25the cloud we we're not actually storing 15:28the data in the Cloud it is secure in 15:30the customer's environment we're only 15:32sending and receiving the data that we 15:34need in order to have the conversation 15:35Flow versus just here's my credit card 15:38number let's just send everything over I 15:40think we're at the cusp of a really 15:41interesting time around this whole 15:43conversation around data privacy and 15:45protection because for years we've been 15:47talking about the fact that you can have 15:49a more personalized experience the more 15:52data that a brand knows about you that's 15:54why we get suggested purchases that's 15:56why we are told when hey that product 15:59ordered might be running out you might 16:00want to order it again that's all data 16:02driven but I think now consumers and 16:05customers in general are becoming savier 16:06about what that means so I want to give 16:09them more control I want to see a world 16:11where we can have a dashboard where 16:12people can say yes you can use my data 16:14this way this way and this way and it's 16:16all integrated into their experience 16:19from the beginning and they have more 16:20control and like visibility into how 16:23it's being used I like this like I want 16:25to sign up for that like right for that 16:27future but I I got to imagine imagine 16:29though that there must be some sort of 16:31like difficulties with scaling to that 16:33like are there any difficulties within 16:35that and if so what does that look like 16:36let's say you have a your virtual 16:38assistant and it's running on a server 16:39you may maybe let's say for Simplicity 16:42sake one copy what if you have let's say 16:4520 people trying to interact at the same 16:47time well since it's running in the 16:48cloud it can actually handle multiple 16:51transactions if we need you know say 16:53like suddenly there's an outage like 16:55it's an electricity company we've got 16:57500 customers calling since it's in the 16:59cloud it can automatically scale so we 17:01have more instances of this assistant to 17:04handle more people oh well that's 17:06fantastic but you know I have to ask 17:09what if it goes wrong what can go wrong 17:11oh so many things um tornadoes being 17:14from Texas what if a tornado comes 17:16through and takes out our data center 17:17well that's why we have multiple data 17:19centers we're going to have all of our 17:21data backed up redundancy in various 17:23places so what I'm getting now is you 17:26both shared so much today but for my 17:28brain I'm trying to process them so I've 17:29got a few key takeaways that I'm going 17:31to take a stab at okay um keep customers 17:34in mind as you start your build because 17:36this is customer service after all it's 17:39important for humans to feel valued and 17:42seen in their requests even if it's not 17:44another human that you're speaking with 17:46on the other end and we don't like to be 17:48deceived so be transparent with your 17:51customers if it's a virtual agent it's 17:53okay just let me know I'm talking with 17:55the virtual agent is that good that's 17:57awesome awesome I love positive 18:00validation 18:02um jeie Morgan thank you so much for 18:06being here today you've got rain cells 18:07all percolating right now um this is a 18:09fantastic conversation that's it for 18:11this episode so thank you all so much 18:13for listening thank you for watching but 18:16worry not there are a ton more great 18:19things where this came from and we will 18:20see you here again soon 18:23[Music]