Riley AI Uses Vetted Medical Data to Guide Early Parenting
Ai4
Amanda DeLuca argues that AI platforms should reduce parental mental load through clinical evidence rather than replace human caregivers or medical professionals.
## Background
This session was published on YouTube by the Ai4 channel on August 28, 2025. It features a conversation between Amanda DeLuca, Co-Founder and CEO of Riley, and Alice Park, Senior Health Correspondent at TIME, focusing on generative and agentic AI strategies for early parenting.
## TL;DR
Amanda DeLuca shares how her personal struggle with postpartum anxiety and midnight searches for reliable advice inspired her to found Riley, an AI platform that translates complex, paywalled medical research into hyper-personalized parenting guidance. The discussion explores how Riley uses a Clinical Advisory Board and strict algorithmic gatekeeping to filter out panic-inducing headlines, unscientific "anecdata," and biased medical advice. DeLuca also addresses context-aware safety features—such as dosage prompts—and outlines the firm boundary that AI should lighten a parent's cognitive load without replacing human healthcare providers or caregiving bonds.
## Atmosphere & style
The session carries an intimate, thoughtful, and reassuring tone that balances deeply personal lived experiences with rigorous clinical discussion. Alice Park's measured questioning creates a collaborative environment for DeLuca, whose insights bridge the emotional vulnerability of early parenthood with clearheaded tech and policy analysis. The discussion feels grounded and practical, moving from late-night anxiety toward a sense of empowered clarity and relief.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=WW9wbpQgx20&t=4s) The founding of Riley was inspired by Amanda DeLuca's personal experience with postpartum anxiety and the late-night struggle to find reliable, evidence-based parenting advice.
- [00:04](https://www.youtube.com/watch?v=WW9wbpQgx20&t=4s) Online parenting forums often rely on 'anecdata'—unverified, opinion-based advice that amplifies anxiety rather than providing scientific clarity.
- [03:49](https://www.youtube.com/watch?v=WW9wbpQgx20&t=229s) Modern parents frequently lack traditional multi-generational support systems, leaving them more vulnerable to biased or person-dependent medical advice.
- [06:11](https://www.youtube.com/watch?v=WW9wbpQgx20&t=371s) Riley relies on a dedicated Clinical Advisory Board (including pediatricians, neurologists, and mental health specialists) to translate complex medical literature into actionable insights.
- [06:11](https://www.youtube.com/watch?v=WW9wbpQgx20&t=371s) Outside core safety protocols like safe sleep, many parenting topics (e.g., infant feeding methods) lack strong scientific backing, requiring AI to highlight nuanced evidence vs. personal preference.
- [10:22](https://www.youtube.com/watch?v=WW9wbpQgx20&t=622s) Pediatricians have responded positively to Riley, which requires specific context (such as child age, weight, and symptoms) before suggesting clinical guidance like Tylenol dosages.
- [13:33](https://www.youtube.com/watch?v=WW9wbpQgx20&t=813s) Much of publicly funded medical research remains behind paywalls; Riley bridges this gap by ingesting and translating peer-reviewed studies for lay parents.
- [14:48](https://www.youtube.com/watch?v=WW9wbpQgx20&t=888s) The platform counters sensationalized 'panic headlines' by evaluating study methodologies, such as sample size and peer-review status, to filter out flawed research.
- [17:23](https://www.youtube.com/watch?v=WW9wbpQgx20&t=1043s) DeLuca emphasizes that AI should never replace human emotional caregiving or qualified healthcare professionals like pediatricians.
- [17:23](https://www.youtube.com/watch?v=WW9wbpQgx20&t=1043s) The primary value of AI in parenting is reducing the cognitive mental load and parental anxiety, allowing parents to be more present with their children.
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=WW9wbpQgx20&t=4s) The Intersection of Agentic AI and Early Parenting
Alice Park opens the discussion by noting how counterintuitive it seems to bring artificial intelligence—specifically agentic AI—into the highly intimate sphere of raising a child. In broader tech discourse, AI often carries fears about the loss of human control, making its application to parenting appear almost contradictory. Amanda DeLuca agrees, sharing her personal entry point into the space. When she became pregnant with her daughter four years ago, she and her husband felt completely prepared, having read the books and bought all the gear. However, after her daughter was born, DeLuca experienced postpartum depression and postpartum anxiety. She frequently found herself awake at 3:00 a.m. in moments of distress, desperate for reliable, science-backed answers about developmental milestones and infant behavior.
Searching the internet in the middle of the night quickly proved overwhelming. Rather than trustworthy medical advice, DeLuca encountered "anecdata"—unverified opinions and anecdotal claims from online forums like Reddit. Because her support network of pediatricians, family, and friends was asleep, she realized how difficult it was to access tailored, evidence-based guidance when parents need it most. This experience led to the founding of Riley, an AI platform designed to merge disparate data points at scale and convert general medical knowledge into hyper-personalized guidance for individual families.
- **Founder background:** Amanda DeLuca previously worked as a tech product manager; she was pregnant with her second child at the time of the talk.
- **The 3:00 a.m. gap:** A recurring scenario where isolated parents lack immediate access to pediatricians or family support.
- **Anecdata vs. Science:** Internet forums often promote unscientific advice (e.g., users claiming non-vaccinated children went to elite universities to dismiss medical consensus).
---
[03:49](https://www.youtube.com/watch?v=WW9wbpQgx20&t=229s) Rebuilding the "Village" with Personal Empowerment
Alice Park connects DeLuca's experience to broader societal changes, referencing insights from the former U.S. Surgeon General regarding the decline of traditional intergenerational living. In past generations, child-rearing wisdom was passed down naturally within families; today, young parents often lack that immediate support system. However, DeLuca emphasizes that available human advice is frequently "person-dependent" and biased, illustrating this with a story about a friend who decided to stop nursing at three months. During a well-child visit, her friend's pediatrician spent two hours lecturing her, claiming that quitting breastfeeding would leave her child less intelligent, shorter, and less healthy.
```
[ CLINICAL EVIDENCE ] [ TRADITIONAL ADVICE ]
│ │
▼ ▼
• Lowers maternal cancer • High personal bias
• Prevents early infections • "Anecdata" / Pressure
│ │
└───────────────┬───────────────┘
│
▼
[ PERSONALIZED GUIDANCE ]
• Data-backed clarity
• Parental empowerment
```
DeLuca points out that while scientific literature confirms clear benefits of breastfeeding—such as reduced rates of maternal breast cancer and early protection against ear infections or diarrhea—many broader claims are unsupported by data. Had her friend used an AI tool like Riley, she could have parsed the actual research, separating objective facts from clinical condescension, and engaged in a more empowered conversation with her doctor.
- **Societal shift:** Breakdown of multi-generational households leaves parents without traditional support systems.
- **Empowered communication:** Objective data gives parents the agency to advocate for their decisions without feeling guilt-driven by individual opinions.
---
[06:11](https://www.youtube.com/watch?v=WW9wbpQgx20&t=371s) Translating Scientific Literature for Lay Parents
Addressing how Riley filters signal from noise across endless parenting theories, DeLuca acknowledges her background in economics and software rather than medicine. To ensure the platform remains grounded in rigorous science, she established a Clinical Advisory Board comprising pediatricians, a pediatric neurologist, and licensed clinical social workers specializing in early childhood development and perinatal mental health. This board evaluates complex studies and translates medical literature into clear, actionable insights for non-specialists.
Park notes that parenting advice rarely presents absolute rights or wrongs outside core safety tenets. DeLuca agrees, stating that once the newborn stage passes, child outcomes rely primarily on basic needs being met and a loving environment. Beyond clear, evidence-based rules—such as SIDS prevention and safe sleep protocols—many hot topics (like infant feeding methods) lack strong scientific backing. Riley utilizes AI to synthesize these nuances, outlining what the data confirms while allowing parents to tailor decisions to their values.
- **Clinical Advisory Board composition:** Includes pediatricians, a pediatric neurologist, and PhD/LCSW specialists in perinatal mental health.
- **Hot-topic debate:** Differing methods for starting solids (e.g., baby-led weaning vs. purees) are often claimed to prevent obesity or create adventurous eaters, despite little supporting scientific evidence.
---
[10:22](https://www.youtube.com/watch?v=WW9wbpQgx20&t=622s) Pediatrician Reactions and Edge-Case Testing
When asked how parents and healthcare providers interact with Riley's outputs, DeLuca notes that users initially brought the platform's answers to doctor appointments to double-check their validity. Surprisingly, pediatricians responded positively to the accuracy of the information. In fact, advisors on Riley's Clinical Advisory Board frequently test the AI with difficult or context-free medical scenarios, such as asking for the "correct dose of Tylenol."
Because proper medication dosage depends on a child's age, weight, gender, and specific symptoms (e.g., fever vs. teething), Riley prompts users for missing context before providing an accurate recommendation based on clinical guidelines. Beyond dosages, parents frequently use the platform to interpret growth percentiles, prepare for vaccination visits, or navigate common childhood illnesses like hand-foot-and-mouth disease, leading to more informed discussions with their primary care providers.
- **Interactive verification:** Board members grade Riley's handling of complex or underspecified medical questions.
- **Context-aware safety:** Medication queries prompt the system to evaluate specific variables (age, weight, symptoms) before suggesting standard clinical dosage guidelines.
---
[13:33](https://www.youtube.com/watch?v=WW9wbpQgx20&t=813s) Rigorous Data Curation and Accessing Closed Medical Science
DeLuca outlines the manual and automated processes behind building and refining Riley's underlying knowledge base. The team systematically reviews anonymized, aggregated user queries to evaluate how Riley responded, working directly with the Clinical Advisory Board to grade each output for accuracy, precision, and tone.
A key hurdle in building an evidence-based parenting model is accessing high-quality medical research. Although many scientific studies are publicly funded, much of the literature remains locked behind expensive paywalls or buried within hospital system databases accessible only to credentialed physicians. Through its Clinical Advisory Board, Riley bridges this gap by ingesting peer-reviewed medical research and translating it into accessible guidance for everyday parents.
- **Feedback loop:** Continuous grading of anonymized responses by medical professionals to maintain clinical fidelity.
- **Accessing paywalled research:** Utilizing physician access to bring locked, publicly funded medical literature to non-physician parents.
[14:48](https://www.youtube.com/watch?v=WW9wbpQgx20&t=888s) Debunking Junk Data and Setting Criteria for Reliable Research
The conversation begins with Amanda DeLuca describing how Riley's knowledge base strategy evolved from a broad, general foundation into a targeted effort driven by user questions and gaps in response fidelity. In parenting, sensationalized "panic headlines" often emerge from low-quality research. DeLuca highlights a specific example involving a widely reported study out of Nova Scotia that examined screen time across just six families. Despite its tiny sample size, poor execution, and complete lack of generalizable findings, it generated widespread hysteria with headlines claiming screen time is the absolute worst thing for children. Riley's goal is to directly counter these fear-inducing claims by recognizing flawed methodologies, reassuring anxious parents, and directing them toward statistically sound, reliable data.
Alice Park asks about the specific gatekeeping thresholds and criteria coded into Riley's algorithm to vet incoming scientific literature—such as minimum sample sizes, participant diversity, and methodological rigor. DeLuca emphasizes that while randomized controlled trials are exceptionally rare in pregnancy and childhood studies due to obvious ethical constraints, Riley sets a stringent bar to get as close to that gold standard as possible. The platform prioritizes large sample sizes and peer-reviewed research evaluated through both clinical and social science lenses to ensure high confidence before any findings inform the AI's recommendations.
- **Example of Panic Media**: Nova Scotia screen time study evaluated only 6 families, yet triggered widespread panic headlines.
- **Vetting Standards**: Riley enforces strict thresholds focusing on sample size, peer-review status, and dual clinical/social science validation.
---
[17:23](https://www.youtube.com/watch?v=WW9wbpQgx20&t=1043s) Agentic AI, "Mother AI", and the Role of AI in Parenting
Turning to a broader, "blue sky" discussion, Park references Geoffrey Hinton's keynote speech regarding the concept of a "Mother AI" and the necessity of developing AI systems that genuinely care about humanity. Park raises the potential of agentic AI over coming decades, particularly in underserved regions or healthcare deserts where families lack access to pediatricians and child development specialists. She asks whether AI could foster a two-way learning dynamic, offering brand-new parenting skills and solutions.
In response, DeLuca establishes firm boundaries regarding what she believes AI should never become. She firmly rejects the idea of Riley serving as a replacement for genuine caregiving relationships or human experts like pediatric therapists. Instead, AI achieves its highest purpose by organizing access to evidence-based insights, helping parents navigate complex options, and lifting the heavy mental load of modern parenting. While DeLuca acknowledges moments where Riley generated "magical recommendations" that surprised even her as a coder, she clarifies that AI outputs remain inherently derivative rather than spontaneously creative. Ultimately, reducing parental anxiety about "irrevocably screwing up" allows parents to spend more present, joyful time with their children.
```
[ Unfiltered Data / Panic Headlines ]
│
▼
[ Strict Gatekeeping Criteria ]
(Sample Size, Peer Review, Ethics)
│
▼
[ Riley Knowledge Base ]
│
▼
[ Reduced Mental Load & Anxiety ]
│
▼
[ More Joyful Time With Children ]
```
- **Core Limitation**: AI must never replace qualified human healthcare professionals or human emotional bonds.
- **Primary Value**: Easing cognitive strain by distilling complex options so parents feel confident in their choices.
---
[21:05](https://www.youtube.com/watch?v=WW9wbpQgx20&t=1265s) Reflecting on Riley's Evolution and Looking Ahead
In the closing minute, Park asks DeLuca to reflect on how her initial vision for Riley compares to its current reality, and what has surprised her most throughout the journey. DeLuca admits that her original ideas once felt overwhelmingly massive and intimidating to build. However, looking back from her present vantage point, those early concepts now seem surprisingly modest.
The platform has expanded far beyond her initial expectations, becoming a deeply valuable tool that delivers immense practical utility to its user base. DeLuca expresses immense gratitude for how the system has grown and remains enthusiastic about its future trajectory. Park thanks DeLuca for sharing her insights, bringing the session to an end amidst audience applause.
Riley AI依托经审核医学数据指导早期育儿
Amanda DeLuca指出，AI平台应通过临床证据减轻父母的精神负担，而非取代人类看护者或医疗专业人员。
Amanda DeLuca分享了自己与产后焦虑作斗争并在深夜苦寻可靠建议的切身经历，正是这段经历促使她创立了Riley平台——该AI平台致力于将复杂且设有付费门槛的医学研究转化为高度个性化的育儿指导。讨论深入探讨了Riley如何借助临床咨询委员会和严格的算法把关机制，过滤掉制造恐慌的标题、缺乏科学依据的“传闻式经验”（anecdata）以及存在偏见的医疗建议。DeLuca还介绍了情境感知的安全功能（如用药剂量提示），并明确界定了AI的清晰边界：AI应当减轻父母的认知负荷，而不是取代人类医护人员或看护纽带。
Riley 联合创始人兼 CEO Amanda DeLuca 分享了创立初衷，源于自身经历产后焦虑时在深夜难以获取可靠科学指导的困境；平台通过 AI 过滤网络论坛上未经验证的个案传言（anecdata），为家庭提供严谨的个性化育儿支持。 (0:04)
Riley 组建了涵盖儿科医生与神经学专家的临床顾问委员会，帮助解读复杂医学文献，并严格区分必须遵循的安全底线（如安全睡眠规范）与缺乏充分科研支持的育儿争议偏好（如辅食添加方式）。 (6:11)
在提供儿童用药（如泰诺剂量）等建议时，Riley 引入了情境感知安全机制，必须在获取年龄、体重及症状等关键数据后才给出临床指导，该流程已获得儿科临床医生的积极认可。 (10:22)
平台通过设定严格的样本量与同行评审门槛，打破医学研究的付费壁垒并驳斥低质量研究引发的“恐慌性报道”；DeLuca 强调 AI 旨在减轻家长的认知负担，绝非取代专业医护人员或人类亲子联结。 (14:48)
Reading summary
Riley founder Amanda DeLuca argues that agentic AI can alleviate parental anxiety and cognitive overload by translating paywalled, peer-reviewed medical research into personalized guidance, bridging the gap left by isolated households and unverified internet advice. However, she stresses that AI must serve strictly as an evidence-based decision-support tool rather than a replacement for human caregivers or clinical professionals.
Addressing Parental Isolation and Unverified Advice (0:04)
DeLuca notes that the decline of multi-generational households has left modern parents without immediate traditional support networks. When experiencing postpartum depression, anxiety, or late-night infant concerns—what she calls the "3:00 a.m. gap"—parents often turn to internet forums like Reddit. In these spaces, they routinely encounter "anecdata" and unscientific claims, such as users citing unvaccinated children attending elite universities to dismiss clinical consensus.
Furthermore, DeLuca observes that direct human counsel can carry substantial personal bias. She points to a friend whose pediatrician spent two hours lecturing her after she chose to stop nursing at three months, asserting without evidence that quitting would make the child shorter, less healthy, and less intelligent. While medical literature confirms specific benefits like reduced maternal breast cancer rates and fewer early infant infections, broader claims lack backing. DeLuca asserts that objective data allows parents to evaluate clinical evidence directly and engage with doctors without fear- or guilt-driven pressure.
Clinical Oversight and Access to Closed Science (6:11)
To ground Riley's algorithms in medical science rather than general parenting theories, DeLuca established a Clinical Advisory Board comprising pediatricians, a pediatric neurologist, and licensed clinical social workers specializing in child development and perinatal mental health. Board members continuously review anonymized user queries and grade responses for clinical precision, accuracy, and tone.
A major structural hurdle in creating an evidence-based platform is accessing scientific research, much of which remains locked behind academic paywalls or hospital databases despite being publicly funded. Riley leverages its physician board to ingest this closed peer-reviewed literature. The system also implements context-specific safety protocols; for example, rather than offering a generic Tylenol dose, Riley actively prompts users for missing variables—such as age, weight, and specific symptoms—before supplying standard clinical recommendations.
Filtering Methodological Flaws and Panic Headlines (14:48)
DeLuca highlights how low-quality studies frequently generate sensationalized "panic headlines" that mislead parents. For instance, a widely publicized study on screen time out of Nova Scotia examined only six families, yet its flawed methodology and lack of generalizability sparked widespread hysteria declaring screen time inherently catastrophic. Riley operates algorithmic thresholds to detect poor methodologies and direct parents toward statistically sound research.
Because randomized controlled trials are rarely feasible in pregnancy and pediatric research due to ethical constraints, DeLuca explains that Riley prioritizes large sample sizes, peer-reviewed publications, and evaluations through both clinical and social science lenses. Outside of non-negotiable safety rules like SIDS prevention and safe sleep protocols, the platform distinguishes proven health outcomes from inconclusive parenting debates, such as whether baby-led weaning versus purees prevents future obesity.
Limiting AI to Decision Support Over Caregiving (17:23)
Addressing broader concepts such as Geoffrey Hinton's vision of a caring "Mother AI," DeLuca sets explicit boundaries regarding technology's role in child-rearing. She rejects the notion that agentic AI should replace pediatric therapists, clinicians, or foundational human caregiving relationships. DeLuca characterizes AI outputs as inherently derivative rather than spontaneously creative, noting that its primary value lies in synthesizing complex options and lifting parents' cognitive burden.
By organizing rigorous data into clear options, DeLuca argues that the platform frees parents from constant anxiety over making irreversible mistakes. Relieving this cognitive strain enables parents to focus on meeting basic needs and providing a loving environment, ultimately fostering more joyful, present interactions with their children.
While DeLuca initially found the prospect of building Riley intimidating, she notes that the platform has grown well beyond her original, modest expectations into a vital utility. The ongoing work centers on scaling data curation and clinical validation, ensuring technology empowers parents' informed decisions while keeping qualified medical professionals central to pediatric healthcare.
[0:04] Thank you. I'm really excited to have
[0:05] this conversation with Amanda and um we
[0:08] had a a brief discussion earlier and the
[0:12] first question I asked her of course was
[0:14] why AI and parenting. you know, it seems
[0:17] like two of the things that you would
[0:20] are the least likely things you would
[0:22] put together, especially, you know, at a
[0:24] conference like this and or just even in
[0:26] general, the conversation around AI is
[0:29] um is is talking about things like,
[0:32] >> you know, the loss of our humanity, the
[0:34] loss of our control and and also, you
[0:37] know, what that means for us as a race,
[0:40] you know, and so when we talk about
[0:42] parenting and having AI, especially as
[0:45] we veer into agentic AI. Yep.
[0:48] >> You know, and and what that means for
[0:50] there's nothing I think in a lot of
[0:52] people's minds more human than
[0:53] parenting, right? And
[0:54] >> it's the most human thing you can do
[0:55] >> and raising a family. Exactly. So, um
[0:58] tell us tell us how you put the two
[1:00] together.
[1:01] >> Oh my gosh, what a story, too. Um so, I
[1:05] you all can see we're expecting our
[1:07] second here in the fall, but I have an
[1:09] older daughter who's three. And about
[1:10] four years ago when I became pregnant um
[1:12] I was product manager had built my whole
[1:15] career on like the tech side of tech
[1:17] using technology to you know solve real
[1:19] people problems and what we my husband
[1:23] and I would joke about like throughout
[1:24] the course of my whole pregnancy we just
[1:26] like sit on the sofa at night and like
[1:28] laugh about how prepared we were. We
[1:29] were like we've read all the books,
[1:31] we've done all this stuff, we bought all
[1:32] the things like we've got this, we're
[1:34] ready. And uh as anyone in here in the
[1:37] room who's a parent knows, as soon as
[1:39] our daughter was born, it was like first
[1:41] of all, all of that went right out the
[1:42] window, just completely empty brain as
[1:45] soon as she was born. But also, I had
[1:48] postpartum depression and postpartum
[1:49] anxiety, which we did not plan for cuz
[1:51] we didn't know we needed to. And I would
[1:53] have these moments at like 3:00 a.m.
[1:55] when like I would wake up from an
[1:57] anxiety nightmare or my daughter would
[1:58] make a weird noise or, you know,
[2:00] something would happen and I would wake
[2:01] up and I'd be like, "Oh my god, what am
[2:03] I going to do? when does this get
[2:04] better? Is she ever going to sleep
[2:06] again? Like, is this normal? Am I okay?
[2:08] Is she okay? Like, what's happening? And
[2:12] I very, very quickly found that trying
[2:14] to find answers to those questions,
[2:16] especially like really good data backed,
[2:18] like, you know, what does the science
[2:20] actually tell us? What do pediatricians
[2:22] actually tell us? It was really hard
[2:23] because the internet is full of what I
[2:25] like to call anecdata. Like people's
[2:27] opinions about things or like some
[2:29] person on Reddit being like, "Well, I
[2:31] didn't vaccinate my kids and they went
[2:32] to Harvard, so like you do the math."
[2:34] And it's like that's just not how
[2:35] anything works.
[2:37] >> It's too much information.
[2:38] >> Too much. And it's just such such an
[2:40] overwhelm, especially in those moments
[2:42] when everybody that I would have asked
[2:44] for help, like my friends or my parents
[2:46] or my sister or our pediatrician was
[2:48] they were all asleep because it was 3:00
[2:49] a.m. And so I remember again having a
[2:52] similar conversation with my husband
[2:54] like on the sofa when our daughter was
[2:56] sleeping one night and basically saying,
[2:58] you know, it shouldn't be this hard. It
[2:59] shouldn't be hard to get really good
[3:01] sciencebacked advice and like it
[3:04] shouldn't be hard to know like what's
[3:05] going on with my kid. What's coming up
[3:08] for my kid? Like what are the
[3:09] developmental milestones that happen?
[3:11] And like really for somebody to who
[3:13] knows me and knows my family and my
[3:15] situation to be able to say, "All right,
[3:17] you're not a general use case. You're a
[3:18] specific use case. Like here's what's
[3:21] going on with your kid. Here's what's
[3:22] going on with your family." And um that
[3:25] was the like initial idea for Riley is
[3:27] you know AI is really really good at
[3:30] being able to marry diverse data points
[3:32] at scale. And so being able to say all
[3:34] right here's what we know about my
[3:35] family. Here's what we know about like
[3:37] what the science says really does and
[3:39] does not matter in parenting and early
[3:40] childhood development and how can we
[3:43] make those two come together for
[3:44] something that's truly unique at scale.
[3:47] And that's that's where we are today.
[3:49] And in some ways, if you think about it
[3:51] more deeply like you did, it it is a
[3:54] good match, right, between AI and what
[3:57] AI is good at and parenting because as
[3:59] we've heard from the former surgeon
[4:02] general too, our society in general is
[4:04] not built the way it used to.
[4:06] >> Yeah.
[4:06] >> You know, the way it used to be in terms
[4:08] of having different generations living
[4:10] together and kind of family knowledge
[4:13] and that kind of knowledge about child
[4:15] rearing, parenting being passed down,
[4:17] right? that that doesn't really happen
[4:19] as much anymore. So, with that kind of
[4:22] gone or dwindling, you know, as you
[4:25] said, there really is this gap of where
[4:27] do you where do you go then for reliable
[4:29] information? There's information, but
[4:31] >> Yes. Well, and and what you realize too
[4:33] is how person dependent the information
[4:36] you're getting is. You know, one example
[4:38] that I love to tell is um, you know,
[4:41] right around the same time that I had my
[4:42] first baby. One of my really good
[4:44] friends also had a baby and struggled to
[4:46] nurse and, you know, kind of went into
[4:48] her pediatrician for their 3-month well
[4:50] visit and said, "You know what? I I I
[4:52] really like this is really hard. It's
[4:54] not working for us. I think I'm going to
[4:55] stop." And the pediatrician sat her down
[4:58] for 2 hours and just bered her about
[5:00] that decision and basically saying,
[5:02] "Well, you're going to disadvantage your
[5:03] child. You're going to do all of these
[5:04] things. like don't you want your your
[5:06] baby to be taller, faster, thinner,
[5:08] smarter? And this is a doctor who's like
[5:11] sitting her down and and and telling her
[5:12] this. And it really, you know, where
[5:15] Riley fits into that conversation is
[5:17] like we have great science to tell us,
[5:19] you know, nursing is fantastic for the
[5:20] nursing mother, like reduces breast
[5:23] cancer rates. That's very statistically
[5:24] significant. We also know that it
[5:26] prevents diarrhea, ear infections really
[5:27] early on, but a lot of what's claimed
[5:29] about it is not well supported in
[5:31] science. And so your feeding journey is
[5:32] yours and you have you are empowered to
[5:35] make a really good decision for your
[5:36] family because feeding babies is really
[5:38] hard. And if she had had Riley, we talk
[5:41] about this a lot too. She's like, "If I
[5:42] had had Riley, I would have asked it
[5:44] once a day every day like, "Hey, is it
[5:45] okay to quit?" And being armed with that
[5:47] information, I could have said to my
[5:49] doctor, "Hey, I hear what you're saying.
[5:51] It's actually like not what the data is
[5:52] telling me and like this is the decision
[5:53] that I need to make for my family." So,
[5:55] I could have had a much more empowered
[5:57] conversation rather than sitting and
[6:00] feeling like garbage while this person
[6:03] kind of just like talks down at you.
[6:05] That's really what we're trying to kind
[6:06] of help fit into.
[6:08] >> Right. So, there's a power that comes
[6:09] from that.
[6:10] >> Absolutely.
[6:11] >> That knowledge.
[6:11] >> Yeah. Um so you come from a background
[6:14] of technology and when you decided when
[6:17] you saw this gap um and this unmet need
[6:20] how did you go about then separating the
[6:23] wheat from the chaff out there you know
[6:24] in terms of all the parent if there's
[6:26] like one topic I think where there's a
[6:28] lot of opinions right it's parenting how
[6:31] to raise your kids
[6:32] >> absolutely
[6:33] >> how how did you start going about
[6:36] >> sort of coating that information and and
[6:39] creating this model or
[6:41] >> you know the very first thing that was
[6:44] really important to me in kind of
[6:45] building our knowledge base building
[6:47] like what ended up going into our model
[6:49] is knowing that I'm not an expert. I am
[6:52] a person I have a degree in economics. I
[6:54] again I know how to code but I'm not a
[6:56] parenting expert. I don't I like have
[6:58] not spent my career like very deeply
[7:00] like invested in what the data says
[7:01] here. So one of the first things that I
[7:03] did was I built a clinical advisory
[7:04] board of pediatricians. We've got a
[7:06] pediatric neurologist. We've got um a
[7:08] couple of PhD um like LCSW like folks
[7:12] who really focus on that early childhood
[7:14] development area, the again perinatal
[7:16] mental health area and asked them I was
[7:19] like hey I have all these studies I've
[7:22] done a lot of research here I've like
[7:23] really combed through here's what I
[7:24] think is a really good like set of data
[7:28] what do you see what what is it that
[7:30] like what's missing are these actually
[7:33] good quality data and like how can we
[7:36] give this and make this more accessible
[7:37] ible to parents in a way that's not like
[7:39] read through the study and understand it
[7:40] if you have a degree in economics but
[7:42] like in lay person terms like what do
[7:44] you actually need to know what is this
[7:45] actually telling us and are we confident
[7:46] that it's actually saying that
[7:49] >> and how do you how do you sort of coales
[7:51] the different points of view because
[7:53] there's no
[7:54] >> I think in parenting and child rearing
[7:56] there's probably no right or wrong it's
[7:59] just an array of opinions and like you
[8:01] said you really have to sort of find the
[8:03] one that matches your situation your
[8:05] circumstances or that you're most
[8:06] comfortable with right ablate
[8:10] all that
[8:11] >> those differing opinions from the
[8:13] pediatrician from the your aunt from
[8:16] your grandmother you know from your
[8:18] friend how do you put that together in
[8:20] something that comes out as a useful
[8:23] piece of advice for parents
[8:25] >> I love this question and this is
[8:27] honestly where the AI element of it is
[8:29] able to shine because you know we do
[8:32] have good data about like again the
[8:35] things that are extremely important And
[8:37] really what it boils down to after the
[8:38] newborn phase is like do you love your
[8:41] children and are there basic needs met?
[8:43] And like that's those are the two most
[8:45] important things that determine like
[8:47] successful outcomes for kids. And we
[8:49] also know like again there's a lot
[8:51] around like safe sleep that we have
[8:52] great data on um and preventing SIDS
[8:54] that we have great data on. And there's
[8:55] a lot that's claimed a lot that like
[8:57] people will tell you especially on the
[8:59] internet there's like a lot that goes
[9:00] around um around like how to start
[9:02] solids for example. And that's like kind
[9:04] of the hot topic in like the baby land
[9:07] right now is like there's um a couple of
[9:09] different methods that people will often
[9:10] claim like oh this prevents obesity or
[9:12] like oh this will this particular method
[9:14] is going to like help your child be a
[9:16] better more adventurous eater and almost
[9:18] none of that is backed in science. And
[9:20] so I think it's really important to be
[9:22] able to say like first and foremost like
[9:24] what do we know with confidence and like
[9:26] what is not backed in data that actually
[9:28] empowers you as the parent to be able to
[9:30] make more of a decision and where the AI
[9:32] can like help personalize to you and say
[9:34] all right knowing your family knowing
[9:37] Amanda and my child and my circumstances
[9:41] uh you know the things that I value the
[9:42] things that I want the things that you
[9:44] know my my daughter does. here's what
[9:47] the data says and like here's a plan for
[9:50] you based on like all of these things
[9:52] all of these like little teeny tiny
[9:53] markers that go into this full picture
[9:55] of your family as a unique unit. Um and
[9:58] so like for the side starting solids
[10:00] conversation um again we can say you
[10:02] know there are a couple of different
[10:03] ways to do this based on again what
[10:06] you've told me based on the information
[10:07] that we have like here's the the path
[10:09] that I feel like is going to work for
[10:10] your family. What do you think? Like you
[10:12] get to make the decision as the parent.
[10:14] Um, and we can kind of collate all of
[10:16] the the different like pieces of data
[10:18] about you, about the science, and like
[10:20] really provide that to you.
[10:22] >> I'm curious about what you're hearing
[10:24] from um people who use Riley about the
[10:27] interactions they have with their
[10:28] pediatricians, you know, and like how do
[10:30] they use that? Is it, you know, you were
[10:33] um use case number one, right? So, was
[10:35] it like do do you hear from them that,
[10:37] oh, we're mostly using it at like 3:00
[10:39] a.m. when I can't reach anyone? or are
[10:41] they bringing, you know, the questions
[10:42] and the information that they get from
[10:44] Riley and discussing it with their
[10:45] pediatricians and perhaps even having
[10:48] small arguments about it. They don't
[10:50] agree, but um what are you hearing about
[10:52] that interaction? How are pediatricians
[10:54] reacting to it? you know, we do hear and
[10:57] especially really early on when we were
[10:58] like like refining what the model did
[11:01] and like the the different advice and
[11:02] how we presented information. I think we
[11:05] got a lot of people who would bring it
[11:06] to kind of fact check and be like, "Hey,
[11:08] this tool I'm using said this. What do
[11:10] you think about that?" And we would
[11:12] actually hear from pediatricians. They
[11:13] were like, "This is a great answer. This
[11:15] is this is really cool." like one of the
[11:17] the pediatricians on our clinical
[11:18] advisory board. Um, one of the things
[11:19] that she does is she goes and asks it
[11:21] like impossible questions that she gets
[11:24] asked and then grades
[11:26] >> want to know which what what one of them
[11:27] is.
[11:28] >> Oh my gosh. Um, an example of this is uh
[11:31] uh what is the right just like a very
[11:34] generic what is the right dose of
[11:35] Tylenol like
[11:37] >> with no context.
[11:38] >> No context. And Riley is able to know,
[11:40] right, your children are this age,
[11:42] they're this weight, they're, you know,
[11:44] like what's like tell me what the
[11:45] context is, like do they have a fever,
[11:47] or they have teething, whatever. Um, but
[11:48] we have like a very specific like here's
[11:51] what the right dose of Tylenol is for
[11:53] weight, age, gender. Um, that again we
[11:56] put together with her. And so it's
[11:57] basically like, oh, hey Kate, like
[11:59] here's all these things. Here's what I
[12:00] know about. Is this for this child or
[12:01] this child? If it's for this child, it
[12:03] should be this amount based on their age
[12:04] and weight. If it's for this child, it
[12:05] should be this amount. always ask your
[12:07] pediatrician if you're if you have
[12:08] questions.
[12:10] Um but like yeah we would have people
[12:12] and actually hear from pediatricians um
[12:14] that like they really liked the answers
[12:16] I was giving and we still to this day
[12:19] like we get a lot of people asking about
[12:21] how to prepare for things like first
[12:23] vaccinations um how to you know
[12:25] understand growth curves like is it okay
[12:27] to be at like the 10th percentile for
[12:29] example um and like I think a lot of
[12:32] times actually can think of one specific
[12:34] user who uh asked a question
[12:37] specifically about um handoot mouth
[12:40] disease and like the best ways to to
[12:42] treat that and their doctor gave them
[12:44] very similar but like slightly divergent
[12:46] advice and they were like well but what
[12:47] about this and they were like oh I never
[12:48] considered that huh yeah
[12:51] so it's it's kind of amazing
[12:53] >> well like you said that's the advantage
[12:54] of AI right and the beauty of it is that
[12:56] no human can kind of access even the
[13:00] amount of knowledge and the literature
[13:01] that AI can that you can send AI on a
[13:04] search for um so with respect to that
[13:07] you know we're we're here to talk about
[13:09] basing something like as human as
[13:11] parenting um on research.
[13:14] >> Yeah.
[13:14] >> So what what do you mean by that when
[13:17] you say that Riley is based on research
[13:19] in terms of like which studies are you
[13:21] accessing? You've got the scientific
[13:23] advisory board. Um how are you how are
[13:27] you really building on research um to
[13:30] create the advice that you provide
[13:32] parents?
[13:33] >> You know it's a very fluid process. This
[13:36] is something that we spend a lot of time
[13:38] very manually kind of looking into. And
[13:40] this goes from the level of like at an
[13:42] anonymous aggregate level. We look at
[13:44] the queries that come in and then we
[13:46] look at how Riley answered them. And
[13:47] then with our clinical advisory board,
[13:49] we kind of grade how accurate, how
[13:51] precise, like how high fidelity they
[13:54] were relative to again the medical
[13:55] advice, the body of research that we've
[13:57] put together that we are tinkering with
[13:59] all the time as well as like for that
[14:01] person. And we've got a couple different
[14:02] parameters that we look at and in order
[14:05] to actually create that knowledge base
[14:07] we do it's especially early on it was a
[14:09] lot of like here's here are the
[14:11] questions that we the people who are
[14:12] building Riley as parents have like
[14:14] based on different age ranges and like
[14:16] here is here's the data that we found
[14:18] you are our board like what do you look
[14:21] at like what are the good studies that
[14:23] ex is there any kind of cutting edge and
[14:25] like really actually also using um like
[14:27] a lot of the the resources that only
[14:29] doctors have access to a lot of like the
[14:31] medical studies that are accessible
[14:32] through hospital systems that our
[14:34] clinical advisory board can give us
[14:35] access to because they're working with
[14:38] us that even though taxpayer dollars pay
[14:41] for them, it's very difficult to
[14:42] actually find those studies or get
[14:44] access to the information as a non-f
[14:46] physician. And so it's really like it
[14:48] started out as like let's be kind of
[14:50] broad and comprehensive and now we make
[14:52] base it very much on all right what are
[14:54] people asking where might might there be
[14:56] holes in our knowledge base based on
[14:58] like again the fidelity of the answers
[15:00] that are coming through from Riley and
[15:03] like where is there maybe like a new
[15:04] study or where where should we like a
[15:07] lot of what happens in parenting
[15:08] especially is like there'll be these
[15:10] panic headlines where junk data will
[15:12] come out like my favorite one was
[15:14] recently they had a a study out of Nova
[15:16] Scotia
[15:17] with like six families of like is screen
[15:20] time harmful for kids and it was like
[15:24] terribly run, very poorly reported, like
[15:26] not in any way generalizable and it
[15:28] generated a lot of headlines. People
[15:30] being like screen time is the worst
[15:32] possible thing you can do for your kids.
[15:34] And when you actually look at the data,
[15:35] it's like none of this we can't you
[15:38] can't like count on it or really like
[15:40] read anything into any of it because
[15:41] like this this study is bogus. And so us
[15:45] being able to say too when people ask us
[15:46] like hey I read this study or I read
[15:48] this headline is this so being able to
[15:50] then say no and here's why like we've
[15:52] seen this like here's why this isn't
[15:54] great reliable data and like here is
[15:56] better reliable data like here's what we
[15:58] know to be true or at least when we feel
[15:59] more confident is true and also being
[16:01] really clear about like when we're not
[16:03] confident that something is true and
[16:05] it's not well supported by the data.
[16:08] Uh do you have like criteria or do you
[16:10] have set up like gatekeeping thresholds
[16:14] for looking at the kind of data that
[16:16] gets fed into your model into your
[16:18] algorithm? For example, especially if
[16:20] you're looking at the literature and
[16:21] scraping the literature, things like you
[16:23] know the size of the study, the number
[16:25] of people, how many sites and how
[16:27] diverse the population was and their
[16:29] methodology and things like that. And so
[16:31] you've got criteria that are part of
[16:33] that algorithm, too.
[16:34] >> Absolutely. And we have a pretty strict
[16:37] and stringent threshold for what we
[16:39] would consider includable. And it has to
[16:42] be again it's it's very difficult
[16:44] especially with pregnancy and like
[16:45] childhoods based studies to get like a
[16:48] randomized trial of anything or like a
[16:50] randomized study because often it's kind
[16:52] of unethical. Um so we get as close to
[16:54] that as we possibly can. We we're
[16:56] looking at gigantic like study sizes,
[16:59] sample sizes. um and you know places
[17:01] where it's been peer-reviewed, it's been
[17:03] like really thoroughly looked at from
[17:06] both a clinical perspective as well as
[17:08] from like a social science perspective.
[17:10] So that when we include data, there's a
[17:13] pretty like meaningfully high level of
[17:15] confidence that like yeah, this actually
[17:17] is worth looking at and it actually is
[17:19] telling us what it says it's telling us.
[17:23] >> So let's go blue sky a little bit. I was
[17:26] inspired by I don't know if you were at
[17:28] the keynote this morning, but I was
[17:29] inspired by Jeffrey Hinton's um
[17:32] >> talk about the mother AI and also really
[17:35] I think his more important point really
[17:37] of making sure that AI that we create
[17:41] and develop AI that cares about us which
[17:44] I think is an important point you know
[17:46] because that's the difference between
[17:48] sort of thinking of AI as you know this
[17:50] big bad thing that's going to replace us
[17:52] but you know as with anything there are
[17:55] pros and cons, right? And so we need to
[17:57] kind of focus on the pros and make sure
[17:59] that AI works with us.
[18:01] >> Um and and sort of I I liked his concept
[18:04] of the of the mother AI. If you take
[18:07] something like parenting and and think
[18:09] about where agentic AI could go
[18:14] >> in coming decades. Um is that is that a
[18:18] role you think that AI should be playing
[18:20] in parenting and as we also look at you
[18:24] know pockets of our population around
[18:27] the world where they may not have access
[18:29] right to healthcare and good pediatric
[18:32] advice and child development advice
[18:34] >> been around the country too this country
[18:37] that's very true
[18:38] >> um
[18:41] >> what are your thoughts on would we you
[18:43] know do you think we're going to learn
[18:45] from AI I perhaps even have new
[18:47] solutions, new parenting skills, and
[18:49] learn from AI as well and have it be a
[18:51] two-way street.
[18:54] >> You know, I I'll tell you what I I don't
[18:56] want AI to do and what I don't want
[18:59] Riley specifically to become. And it
[19:02] really falls into the category of like I
[19:04] never want Riley to be a replacement for
[19:06] a legitimate caregiving relationship. I
[19:08] never want it to be the thing that you
[19:11] go to instead of seeking a qualified ped
[19:16] pediatric like therapist, you know,
[19:18] somebody who actually is a human who
[19:20] knows what they're doing. Um really
[19:23] where I think AI can have the most
[19:26] positive outcome for humanity is in
[19:29] providing like again really good access
[19:32] to data really good like understanding
[19:34] of how things work how you know like
[19:37] what we know what we don't know like
[19:39] different things to try. I've never
[19:42] really seen AI even Riley like I there
[19:44] are some things that it does that like
[19:45] even though I've coded a lot of it
[19:46] myself like sometimes it will surprise
[19:48] me. it'll like come up with like a a
[19:50] genuinely magical recommendation. I'm
[19:52] like, "Oh, I never thought of that." But
[19:54] I know that it's not like truly unique,
[19:57] right? Like it didn't come from nothing.
[19:58] It didn't spontaneously create that. I
[20:00] know exactly what went into the prompt
[20:02] that created and like the the mechanisms
[20:04] that like got all of the data that
[20:06] created that special moment for me. And
[20:08] so I don't know if AI is ever going to
[20:10] be like genuinely spontaneously creative
[20:12] in that way. It's always going to be
[20:13] somewhat derivative. And that's not
[20:14] necessarily a bad thing because there is
[20:17] a lot that we know. There's a lot that
[20:19] like we have debunked over the years.
[20:21] And I think if we're able to give
[20:24] parents or whomever a really good sense
[20:27] of like here's here's what again data
[20:30] says, here's what like the different
[20:33] potential approaches. Again, going back
[20:34] to starting solid, like here's the
[20:36] different ways that you can approach
[20:37] this. Here's what tends to be popular.
[20:39] here based on you, here are the
[20:42] different pathways that might make the
[20:43] most sense and actually providing some
[20:45] ease of mental load there. So that
[20:48] ultimately what you get is more joyful
[20:50] time with your children, less time being
[20:51] like, "Oh my gosh, am I doing this
[20:52] right? Am I am I showing up in the ways
[20:54] that I want to? Am I irrevocably
[20:56] screwing up my kids?" You can feel more
[20:58] confident and actually enjoy them more.
[21:00] And that's really what I want to give to
[21:01] Riley users at the end of the day. I
[21:02] want to give them back more joyful time
[21:04] with their kids. And I know I think we
[21:06] have to wrap up, but I just want to ask
[21:08] you a very quick question.
[21:09] >> Yeah.
[21:10] >> Um,
[21:11] >> when you envisioned Riley um, and your
[21:14] vision for it, what was it like then
[21:17] versus where it is now? Are how
[21:19] different is it from where it is now?
[21:21] And and what surprised you the most in
[21:23] in that process?
[21:24] >> You know, honestly, the very first
[21:26] Inklings of Riley, I look back on it and
[21:29] they seemed like such big ideas and like
[21:31] I was I was I had no idea how I was
[21:32] possibly going to do any of it or build
[21:34] any of it. And looking back, I'm like,
[21:36] "Oh gosh, all of that was so small."
[21:38] Like what we're building now and like
[21:39] what we have now is so much bigger, so
[21:42] much cooler than I could have possibly
[21:43] even like dreamed back when I was
[21:46] getting started. And that's such a like
[21:48] such a treat, you know? It's such a a
[21:49] pleasure to be able to be like, "Cool,
[21:51] this is a tool that is genuinely useful
[21:53] for me and for our users and like is
[21:56] actually providing an incredible amount
[21:58] of value." And I never could have
[22:00] imagined the amount that it would do
[22:02] that back when I started. and I'm sure
[22:04] it will only continue to grow. Amanda,
[22:07] thank you very much for sharing.
[22:08] >> Thank you so much.
[22:10] [Applause]
