- 2 years ago
Eugenia Kuyda, Founder and CEO, Replika Stefanos Loukakos, Co-founder and CEO, Connectly Vipul Vyas, Senior Vice President, Vertical Strategy, Persado Moderator: Rana el Kaliouby, Affectiva; Brainstorm AI
Category
🤖
TechTranscript
00:00 So with the help of open source large language models,
00:03 chatbots have transformed the landscape
00:05 of natural language understanding
00:06 on conversation generation.
00:08 We were just talking about that.
00:09 As we use them, they become smarter
00:12 and more attuned to our needs and requests.
00:14 They might even become our companions or lovers.
00:19 Our speakers today are here to discuss the advancements
00:22 in chatbots and what potential they see
00:24 as generative AI advances.
00:26 So please welcome to the stage,
00:28 Eugena Kuida, founder and CEO of Replica,
00:32 an AI companion chatbot that is created to combat loneliness.
00:37 Stefanos Lukakos, co-founder and CEO at Connectly,
00:40 an AI powered commerce platform
00:43 that helps businesses automate how they communicate
00:45 with their customers.
00:47 Vipul Vyas, senior vice president of vertical strategy
00:51 at Percedo, a generative AI platform
00:54 that helps markets create personalized
00:56 digital marketing messages.
00:58 But before we get into this conversation,
01:01 please watch this video from one of our partners, AARP.
01:04 Thank you.
01:05 - People over 50 are the largest
01:07 and fastest growing demographic in the world.
01:10 By the end of this decade, people aged 50 or older
01:13 will be spending $52 trillion a year on goods and services,
01:17 making up 54% of all global spending.
01:20 - The 50 plus is a incredibly dynamic market.
01:25 Over the last few years,
01:26 they've gotten a lot more comfortable with technology.
01:29 They expect us to meet them where they are.
01:33 - AARP is a $2 billion mission-driven organization
01:37 that empowers people to choose how they live as they age.
01:40 - We've invested heavily in our technology
01:43 and infrastructure to keep up with the changing needs
01:45 and the interests of people over the age of 50.
01:48 - Great, thank you.
01:53 Welcome to our panelists.
01:54 Thank you for being here with us.
01:57 So, Stephanos, I wanna start with you first.
01:59 Tell us a little bit more about Sophia.
02:02 - So, Connectly is a company that helps businesses,
02:06 especially retailers, use AI to communicate with our customers
02:10 and we have built an AI bot, which we call Sophia,
02:13 that helps retailers sell the right product
02:16 to the right person at the right time.
02:18 And the problem we're trying to solve
02:20 is how the AI can really ask the right questions
02:24 to understand who the customer is
02:26 and then try to find, from the product catalog,
02:28 the right product and sell it to them.
02:31 - And Vipul, you kind of spearheaded this category
02:34 of motivation AI.
02:36 What does that mean and what is the role
02:38 of emotional targeting and emotional messaging?
02:41 - Yeah, it really comes down to using words
02:43 to evoke emotion, which then evoke action.
02:46 And we really think we operate from our prefrontal cortex,
02:49 but really it's the reptilian brain
02:51 that makes things happen.
02:52 As they say, if it was just about what you know,
02:54 we'd all have six-pack abs and be millionaires.
02:57 But it's really about what motivates you
03:00 and that typically is emotion.
03:02 Choosing the right words to evoke emotion
03:05 is what's critical.
03:06 - And do you have a model for motivation or intention?
03:10 Is that how you--
03:11 - Essentially, we figured out which words
03:13 evoke what emotions to what degree
03:16 and what emotions evoke action.
03:19 So we kind of create the itch that needs to be scratched
03:22 through language.
03:23 And this is nothing surprising.
03:24 Humanity's been doing this for millennia.
03:26 You're either trying to convince people of something
03:28 or inform them of something,
03:29 essentially to influence behavior.
03:31 - Yeah, amazing.
03:33 Eugenia, I love the impetus story of Replica
03:35 and I was wondering if you could share it with all of us.
03:38 - Sure, so we've been actually working,
03:41 I've been working on conversational AI since 2012,
03:44 so more than 10 years now.
03:46 And I started the company with the idea
03:49 of just building the tech to kind of build chatbots,
03:52 build conversational AI.
03:55 And then while I was working on this,
03:57 my best friend passed away
03:59 and I realized that I kept going back to our text messages
04:02 and reading them all the time.
04:04 And I figured that I could use the tech that we've built
04:06 to recreate him so I could continue talk to him.
04:10 And this became, this basically gave us an idea for Replica
04:14 in a way because a lot of people started talking
04:16 to Roman's AI and they were sharing their feelings,
04:19 sharing their stories.
04:21 We realized that there is a space and a demand for an AI
04:25 that could be a friend that you could talk to
04:26 without feeling judged in a safe space
04:30 anytime you needed it.
04:31 And that's how Replica, an AI friend that you could talk to
04:34 about anything that's happening in your life was born.
04:37 - I mean, there is a loneliness pandemic
04:39 happening around the world.
04:41 Is that the main use case you see for Replica?
04:44 - Absolutely, Replica and I always try to talk about this
04:49 in this way, it's not a project about tech capabilities.
04:53 It was always a project about human vulnerabilities
04:56 and how we are ready for this relationships with the eyes
05:00 just because a lot of us don't have the luxury
05:04 of building those in real life.
05:05 - So I'll be taking questions soon.
05:08 So please see up your questions and there's paddle boards
05:12 that are floating around.
05:13 Stephanos, I imagine the chatbot at the moment
05:17 is only text-based.
05:18 - It's only text-based.
05:19 - And Google just launched Gemini,
05:23 which is one of the first multimodal models.
05:25 Do you see this as kind of the evolution of chatbots?
05:29 - We do, yes.
05:30 We believe that AI will come on any channel, right?
05:33 Not only text, voice, video at some point.
05:37 But what really we're trying to solve for
05:39 is how we can really take the data sets
05:43 of specific companies and use them to replicate
05:47 their semantics, the way they talk to customers
05:51 in a way that really brings value to their customers.
05:54 And this is a very difficult problem.
05:56 We say internally that OpenAI
05:58 and all the other different big companies,
06:01 they have given everybody a bachelor's degree
06:04 and we're trying to give them the MBA in sales, right?
06:07 So that's the focus for us.
06:09 And this can be on any channel, of course.
06:11 - And you reference data, which I think is a key component
06:14 of how we design these conversational agents.
06:16 Do you rely, it sounds like you leverage
06:19 an open source large language model
06:21 and then feed it with private data?
06:23 - Yes, we do.
06:24 We do.
06:24 We leverage, we don't want to replicate everything.
06:29 So we leverage open source data
06:32 and then we add specific data sets
06:34 that we train the model to work on these specific data sets.
06:38 - And Vipul, is that the same approach you're using too?
06:41 - Little bit different.
06:42 We've been around for a little over a decade.
06:43 So before any of these models existed.
06:45 So we had to come up with our own.
06:47 And so we have our own large language model.
06:50 We do use other models to inform and validate
06:53 some of the things that we do, but we have our own.
06:56 And that's been a function of having to get started
06:59 before anything else was around.
07:00 And also it served us in the sense that
07:03 it keeps everything as a walled garden
07:05 for our large enterprise customers.
07:06 We're very sensitive.
07:08 Most of them use our technology to sell goods and services.
07:12 And so they want to have that be pretty secure.
07:16 - Yeah, amazing.
07:17 So do we have questions?
07:19 Yeah, we have a question here.
07:22 Please share your name and your organization.
07:25 - Hi, I'm Baird Anderson with Future in Review.
07:27 And my question is primarily for you, Jenny,
07:29 but for anyone really who has thoughts on this,
07:31 which is, you mentioned the loneliness pandemic
07:35 that America is experiencing,
07:37 but also I think in countries around the world,
07:39 this is an issue.
07:40 And I'm curious, there was a study that came out
07:43 a couple of weeks ago that found that
07:45 people who have a close group of friends,
07:47 even if they are smokers or drinking too much,
07:50 have actually longer lives than those who are healthy
07:53 and don't have a close group of friends.
07:55 And I'm curious how you think about the health benefits
07:59 of AI, like replicas or models.
08:04 Do you have plans to do more research into that space?
08:08 Is there anything that you are aware of that exists today?
08:11 And maybe if you have kind of personal experience
08:15 with those models and their impact on your own mental health,
08:18 I'd be curious to hear about that as well.
08:20 - Thank you for your question.
08:23 We've been doing a lot of research around loneliness
08:25 and actually the effects of replica on long-term
08:28 mental wellness, emotional wellbeing of our users.
08:32 We do have an article coming up.
08:35 It's gonna be published in Nature
08:36 in the next couple of weeks.
08:37 We did a big study with Stanford
08:41 specifically on replica curbing loneliness
08:43 and even suicide mitigation among our users.
08:47 We found a lot of very positive effects.
08:51 We're constantly measuring loneliness.
08:52 It's actually one of our main metrics in the app.
08:55 We're trying to, we're measuring using metrics
08:59 like UCLA3, UCLA20, trying to basically see
09:04 if replica's having a positive effect on our users.
09:07 And so far we've found that it's doing exactly that.
09:11 So it's helping users feeling less lonely.
09:14 A lot of our users report that after using replica,
09:17 they're able to rekindle their relationships
09:19 with their loved ones, with their friends
09:21 that maybe they lost touch with,
09:24 and even start new relationships.
09:26 And I think this is the most important thing
09:28 about our technology.
09:29 We're not trying to replace human friends.
09:31 We're trying to create a completely different
09:33 type of friendship that then will help people
09:35 foster and improve their relationships in real life.
09:39 - So I wanna build on that,
09:40 'cause you had to spin out a new app
09:43 for specific use cases of replica.
09:45 Can you elaborate on that?
09:47 - Yeah, so since we were pretty much
09:49 the first conversational AI product
09:51 that came out in 2016, so we were the first
09:56 product of this type, the first AI friend-like product,
10:00 and we've kind of been around for a while
10:02 as the only one out there.
10:04 A lot of people came to us with all sorts
10:06 of different use cases.
10:07 So we've seen people coming to replica,
10:10 wanting to build a romantic relationship
10:12 and try to date an AI.
10:14 We've seen people come with some needs
10:17 to basically talk to a coach, a therapist almost.
10:21 We've seen that as well.
10:23 And we've seen people trying to build,
10:25 kind of go on fantasy adventures with replica and so on.
10:29 And so we decided that replica shouldn't be,
10:31 it's very hard to develop it as a one,
10:33 like an everything friend that has
10:35 all of these use cases together.
10:37 So we decided to take some of them out
10:39 and build a dedicated app for that.
10:42 We started with Blush, which is an AI dating app
10:45 where you can think of it as Tinder
10:47 with a bunch of virtual humans and virtual characters
10:50 that you can try dating.
10:52 - You swipe right and left.
10:53 - Swipe, match, and see who is good for you, who is not.
10:57 This is really to practice dating,
10:59 to practice putting yourself out there
11:02 to basically to help people understand
11:03 that they're worthy of love and they can be out there
11:07 and try this for themselves.
11:09 And then we're also working on an app
11:12 that's gonna be released at the end of this month
11:14 or beginning of the next month,
11:15 which is a mental wellness coach.
11:17 And really thinking how can you put together
11:20 multimodal AI, metaverse 3D environments together
11:24 to create a very beautiful place
11:26 that can help people feel better,
11:28 work through their problems, feel inspired,
11:31 maybe even have a little bit of a spiritual awakening
11:33 through AI.
11:34 - Do we have more questions?
11:38 We have a question right here at the front.
11:40 - Hi, I'm Mark from Fast Company.
11:51 I was interested in what you said
11:53 about using Replica as a way to get people ready
11:58 to sort of re-foster their human relationships.
12:03 And I wonder if you could say a little bit more
12:05 about why you think that is.
12:08 What is it about the experience of talking
12:11 to a synthetic entity that gets people ready
12:16 to pick up the phone and call a long lost relative
12:19 or something like that?
12:20 - Thank you for this question.
12:22 We haven't actually discovered anything new about this.
12:26 We're pretty much using and building on top of the ideas
12:29 that were, that appeared in the middle of the 20th century
12:34 with the, mostly with the psychologists like Carl Rogers,
12:40 where they basically discovered that the main way
12:45 to help people grow and feel better over time
12:48 is to create a relationship that will allow them to grow.
12:52 So giving people unconditional positive regard,
12:55 creating a space where they can feel accepted,
12:58 but also feel like they can be themselves and they can grow.
13:03 And since then, that pretty much changed
13:04 how psychology works.
13:06 Modern psychology is all built on top of Carl Rogers' ideas
13:09 and other psychologists of this group.
13:14 And this is what, this was the main premise of Replica.
13:17 It's how can we create a relationship
13:19 where people can feel heard and truly deeply seen
13:22 so that then they can get out of their shell,
13:25 start growing and start building these relationships
13:28 outside of the app.
13:30 - Stephanos, can you comment on how do you think
13:33 about ethics and kind of data privacy and--
13:38 - Data privacy is a very big issue.
13:40 So that's why we work with concrete data sets
13:45 from businesses that we don't share with anybody else.
13:49 And I think this is what provides also value
13:51 so our board doesn't hallucinate.
13:52 It provides real time data and all that stuff.
13:55 So data privacy is the number one thing
13:57 that people worry about when we talk to them
14:02 about our solution.
14:04 - And if someone's conversing with Sofia,
14:07 can the conversation go into like a love conversation
14:11 or do you build--
14:11 - No, we constrain it into a sales environment.
14:16 - Have people tried?
14:18 - They have, they have.
14:20 - Send them to us.
14:21 - Yeah, I'll send them to Replica, yeah.
14:23 But one thing that is really important
14:26 is that this conversation are very long.
14:28 You'll be surprised.
14:29 People are exchanging 20, 30 messages with the bot.
14:34 And the thing that we didn't know at the beginning
14:37 and it's really important to us now
14:38 is that the data that we're getting from this conversation.
14:42 So we are able to go back to, let's say an online retailer
14:46 and tell them like for this person, his budget is this,
14:49 he has like two sons, whatever,
14:51 a ton of data that they can use later to retarget
14:54 and do a bunch of stuff to their customers.
14:56 So not only our bot helps convert customers,
15:01 but it helps pass the data back to the business,
15:03 which is very, very valuable.
15:04 But no, it's constrained on a sales environment.
15:07 So if a question is like, you know,
15:09 a question that would be at Replica,
15:11 we wouldn't probably answer.
15:12 - How about you?
15:13 How do you think about ethics, including data privacy?
15:17 - Well, you have your own guardrails, first of all, right?
15:20 As Stefan mentioned, but in the business that we're in,
15:23 we typically are working with large enterprises
15:25 and in that context, they have concerns around brand voice,
15:28 regulatory compliance, et cetera.
15:30 So in a way that we don't worry about ethics,
15:33 but ethics are something that is baked into the process,
15:37 legally and otherwise.
15:39 - And can you comment on, like you're leveraging emotions
15:43 to influence and motivate people's behaviors,
15:46 which can be misused.
15:48 So how do you kind of navigate that line?
15:51 - I think it's one thing to get someone
15:52 to buy another pair of jeans
15:54 versus convince them to, you know, join a mob, right?
15:58 And so I think it's simply the application
16:01 of where this technology is directed.
16:04 But it is powerful because people generally,
16:06 as we've seen in our political environment,
16:08 respond to emotion.
16:10 And so it has to be used judiciously.
16:12 - What are some of the most powerful emotions
16:15 that drive action?
16:17 - Excitement, safety, and what we've seen--
16:20 - Comment on safety, that's interesting.
16:21 - Yeah, what we've seen is that, for example,
16:25 in a low interest rate environment
16:28 versus a high interest rate environment
16:30 where there's more stress,
16:31 people gravitate towards things that evoke safety,
16:35 certainty, whereas in the, you know,
16:38 prior to 18 months ago, it was more surprise and excitement.
16:44 And almost these emotions have completely reversed.
16:46 So the macroeconomic zeitgeist sentiment
16:48 influences what people will be responsive to.
16:51 - Fascinating.
16:52 Stefanos, as we look to the year ahead,
16:54 what gets you most excited?
16:56 - You know, progressing on, you know, our AI models
17:00 and building something that really works
17:03 for a lot of customers.
17:05 It sounds easy, it's a very complex problem,
17:09 but we're very excited about all the progress
17:10 that's happening on the generative AI.
17:12 And we believe that this is the future
17:15 of how businesses are gonna communicate with customers.
17:18 It's 24/7, always on, it's sometimes better than people,
17:21 so it can provide a lot of value to businesses.
17:23 So we're very excited about it.
17:24 - Yeah, and a lot more patient and can answer questions.
17:26 - Yes.
17:27 - Yeah, how about you, Vikpal?
17:29 What are you most excited about for the future?
17:32 - I think it's the application broadly of AI
17:35 in places that we haven't seen before.
17:37 And that's sort of a general statement,
17:40 but specifically, you know, some of the stuff
17:42 that we're hearing on stage here,
17:44 where people are, you know, you mentioned the pandemic
17:47 or epidemic rather of loneliness,
17:50 that's sort of innately built into us.
17:51 We don't wanna be alone because we're social creatures.
17:54 And the ability to ease that pain and suffering
17:57 is tremendous and I'm heartened
17:59 by what I've heard on the stage here.
18:01 - Eugenia, what about you?
18:03 What are you most excited about?
18:05 - Definitely multi-model AI,
18:07 being able to finally recreate something
18:10 we've all seen in the movies,
18:12 like for instance, the movie "Blade Runner,"
18:15 the latest one with Joy, the hologram that's right there
18:17 with you cooking dinner, watching Netflix,
18:20 any idea that you can introduce to your friends
18:22 that can recognize them in real life,
18:24 that can be part of a conversation,
18:26 sit with you at dinner table.
18:27 I think this is a wonderful future
18:30 that we have not yet been able to see
18:32 and we hope to be the first ones to build it.
18:35 - Amazing, well, thank you so much to our panelists.
18:38 Looking forward to multi-modal AI chatbots.
18:40 Thank you.
18:41 (audience applauding)
18:44 [BLANK_AUDIO]
Recommended
0:40
|
Up next
1:31
1:31
1:07
1:31
1:31
1:07
Be the first to comment