AI and Moxie
Moxie is a companion robot for children ages 5-10 designed to help support social, emotional, and cognitive skills through play-based learning and engaging content developed in association with experts in child development and education.
Moxie was developed to help support children’s social, emotional, and cognitive development, while safeguarding the privacy and security of their personal information, using Embodied’s proprietary SocialX™ AI platform. SocialX enables users to engage with technology through the means of natural interaction (i.e., facial expressions, language, body language, etc.). This novel interface is enabled by machine learning technology that continuously adapts to and learns from the human user. This allows Moxie, through the SocialX technology, to support and promote child developmental skills using natural language processing, multimodal sensing, and empathetic responses.
Studies have shown that child-robot interactions can promote learning and other social and emotional developmental gains in children. Find out more about the science behind Moxie by reading our blog post here. Moxie of course is not a replacement for the support of parents, teachers and others, but can help to reinforce concepts and help children build skills.
Privacy and Data
At Embodied, safeguarding children’s privacy and maintaining the security of their information is at the core of what we do. Our goal is to provide the benefits of Moxie in a way that reflects principles of privacy by design. We work with PRIVO, a recognized Federal Trade Commission (FTC) Safe Harbor organization, to obtain parental consent and to certify compliance with the Children’s Online Privacy Protection Act (COPPA).
We have implemented a data management system that allows us to have access to the information we need to make sure Moxie works properly, but in a way that does not routinely link the most sensitive data that is shared – like audio transcripts and facial expression datapoints - to a particular family or child.
Yes. Moxie will not operate unless a parent downloads the Parent App, creates an account, and goes through the parental verification process through our third-party Safe Harbor organization, PRIVO. Parents share certain information with PRIVO that allows PRIVO to verify that they are a parent. PRIVO tells us that the parent is authorized, but doesn’t share the information they relied on to conduct the verification process.
The Parent App automatically assigns a unique recovery key to the parent, which is unknown to Embodied. The recovery key is stored in the device running the Parent App (e.g., the device’s flash drive or hard drive). Information on how to save the recovery key is provided in the Parent App when the recovery key is assigned.
Using the recovery key or the AUID through the Parent App, parents may access their child’s personal information, update their or their child’s personal information, or change and update goals for their child. Parents may activate or deactivate notifications received from Moxie regarding Moxie’s status or new missions available for your child.
Parents will need the recovery key to recover data in case Moxie is replaced, the parent uses a new device or account to access the Parent App, and the data needs to be reloaded.
Since Moxie is a connected product that collects data using a camera, microphone and sensors, we’ve built Moxie so you can visually see when it is on, in standby mode, or asleep.
When Moxie is turned on using the power switch on its base, it will take some time for Moxie to start up and get ready. Moxie’s screen will turn purple when Moxie is ready. When Moxie is in standby mode, Moxie’s screen will be off and the light bar will be a lavender color. When Moxie is asleep, Moxie’s head is lowered and dream bubbles float across Moxie’s screen. In these states, Moxie is collecting audio data and listening specifically for what is called a “wake word.” Audio data collected while Moxie is listening for a wake word is processed on a separate chip within Moxie that doesn’t have any memory. When Moxie hears the commands “Hello Moxie” or “Moxie, Please Wake Up”, its eyes will open and it will then be in operational mode. In operational mode, Moxie collects audio data and video data utilizing its camera, microphones, and sensors. When Moxie is in operational mode, you can tell by looking at the LED Bar. If the LED bar is blue, Moxie is listening for input. If the LED bar is pulsing blue, Moxie is processing the input it received. If the LED bar is pulsing pink, Moxie is speaking.
After a child finishes interacting with Moxie and says “Moxie, Please Go to Sleep”, Moxie will go into standby mode. You can tell what state Moxie is in by looking at the LED bar (See User Guide for details)
The AI in Moxie is designed to utilize the power of AI and machine learning to help children develop certain skills, like language development, reasoning and response, self-correction, and emotional/social success strategies. To do this Moxie is equipped with microphones, a camera, and sensors to allow Moxie to interact with a child and to offer the benefits of our AI to you and your child. Because we collect audio, video data, primary user images, facial expression datapoints, and other data from a child, parents must provide their verifiable parental consent before Moxie begins to operate.
Moxie has been designed to have real-time conversations with your child. Children are still developing their language skills, so the AI that interprets the conversations a child has with Moxie is designed in a way that seeks to interpret that speech. It is important to know that speech recognition is not the same as voice recognition. Moxie isn’t designed to determine that the speaker is a specific child (voice recognition). Rather, Moxie is designed to interpret what the child user says (speech recognition).
Only during operational mode (and not in sleep or standby mode), will the microphones in Moxie record and transmit audio data. The Audio Data is immediately encrypted and sent to our cloud service provider, who automatically transcribes the audio files to create audio transcription files, or what we call Audio Transcript Data. The Audio Transcript Data is sent back to Moxie so Moxie can respond to the child. Embodied can view the transcript files, but because Embodied does not have access to the Analytics User ID, Embodied cannot tie a particular transcript to a particular individual or Moxie. Parents can request access to the Audio Transcript Data by asking Embodied for access to the Analytics User ID through the Parent App or using the recovery key; this data is not automatically available.
Video Data is processed locally on Moxie and is used only to create facial expression datapoints. Processed locally means the video data is never transmitted beyond Moxie. The camera also captures what we call the Primary User Image. FAQs 10 and 11 provide more details on how we handle primary user image data and why we use it.
Through its camera Moxie will initially capture images of the primary user (your child or who Moxie will refer to as its mentor) during their first interaction with Moxie. The AI within Moxie utilizes the Primary User Images to verify that Moxie is speaking to the primary user during conversational interactions and to respond appropriately. The Primary User Images will be encrypted and securely transmitted to the cloud storage service provider. Only parents will have access to the encrypted Primary User Image using the recovery key; these images are not available to Embodied, but Embodied can link it to the user or device and can therefore delete Primary User Images if requested by a parent.
Images of other individuals may be captured if they are within the field of view of the camera.
Moxie may process these images to determine if this individual is the primary user and may ask “who are you?” or “are you my mentor?” to confirm. If another user asks Moxie to remember them, that user will be recognized and their information stored. Moxie may communicate with the other individuals, but is designed to call only the primary child user – Moxie’s mentor - by name. This comparison is performed within Moxie and there is no communication with the cloud storage service provider.
Moxie collects video of a child’s interactions with Moxie, but the video files never leave Moxie. In other words, the video is stored only on the Moxie device. Using the AI in Moxie, however, the video file is translated into Facial Expression Datapoints to help promote a more natural and meaningful conversation between Moxie and the child.
Facial Expression Datapoints allow Moxie to determine that the robot is interacting with a person, and then uses facial characteristics, like facial expression, to help Moxie assess the child’s emotional state (happy, excited, puzzled, sad) to enhance the interaction between Moxie and the child. The Facial Expression Data points will be transmitted to the cloud storage service provider and are encrypted during transmission and storage. This data is available to Embodied but not in a way that is linked to a particular child or device.
Activity Data includes how long children are using Moxie, if children read a book with the robot, how many missions the children have completed with Moxie, badges and trophies earned, and other information to help parents track a child’s general activities and progress. The Activity Data is generally summary information about the child’s activities. Activity data is encrypted and stored with the cloud service provider, but in a separate database from the Audio Transcript Data, Facial Expression Datapoints, Moxie Interaction Data, Insight Data, and other personal information. Your child’s Activity Data is accessible through the Parent App or with the recovery key and linked to the robot ID accessibly by Embodied.
Moxie Interaction Data is data associated with the child’s interactions with Moxie, such as the Audio Transcript Data (collected through Moxie’s microphones), Facial Expression Datapoints (collected through Moxie’s camera), whether and where other objects or persons may be located in a room (collected through Moxie’s camera), responses communicated to Moxie, and/or what Moxie facial animations are utilized and information from touch sensors to determine if a child is touching or hugging Moxie. The Moxie Interaction Data is collected and processed on Moxie and used to enhance the child’s interaction with Moxie and is associated with the Analytics User ID. The Moxie Interaction Data is then encrypted, transmitted to the cloud service provider and stored in the same account as the anonymized Insight Data, Audio Transcript Data, and Facial Expression Datapoints. The Moxie Interaction Data is stored for three months with precise timestamps, and for eighteen months with relative timestamps. Parents can request to access or delete Moxie Interaction Data using the Parent App or with the recovery key.
Insight Data is what makes Moxie valuable to children and parents. It is data derived from the Embodied proprietary AI engine that analyzes how the child is doing in achieving missions and goals and progress towards achieving those goals, such as improvements in language skills, how long the child was engaged with the robot, number of words read per minute. Insight Data is just that: it is data derived from the Moxie Interaction Data, Facial Expression Datapoints, Audio Transcript Data and other data to provide an analytical assessment of the child’s progress using Embodied’s proprietary AI. Activity Data and Insight Data are available to parents while using the Parent App or with the recovery key.
Embodied collects computer sensor and telemetry data from Moxie to troubleshoot problems and improve hardware and software operations. This data is associated with the robot ID but stored separately from other data for three months.
Protecting children’s privacy and safeguarding their information are core principles for Embodied. That is why children cannot use Moxie absent verifiable parental consent. We work with PRIVO, a recognized Federal Trade Commission (FTC) Safe Harbor organization, to obtain parental consent and to certify compliance with the Children’s Online Privacy Protection Act (COPPA).
We take special steps to protect data. Moxie is designed so that some data (like videos of interactions with Moxie) never leave the Moxie device. Data that has to be transmitted to the cloud to help Moxie function is protected in different ways. For example, audio files, primary user images and facial expression datapoints are encrypted in transit. Audio files are deleted after a transcription is created. Facial expression datapoints and primary user images are retained to help Moxie function and interact with a child, but are encrypted in storage.
We provide regular software updates and patches. We also commission third party penetration tests to confirm that our security measures are robust, monitor the security environment, and update software and firmware when needed, and conduct regular security audits. Parents and children should be mindful of good security hygiene. Embodied will never ask a parent to share the recovery key. To assist parents in data recovery and troubleshooting (see FAQ 17), we provide a mechanism for parents to affirmatively agree to share the Analytics User ID. To maintain security, this only occurs within the Parent App and parents must ask Embodied customer service personnel to provide the “Share my AUID” button through the App.
Embodied works hard to protect the information entrusted to us, but no security method is 100% secure. Thus, Embodied cannot guarantee that the security methods and/or precautions we take are failure proof.
Moxie is designed to help parents understand how and when a child uses Moxie, and to provide updates to parents gleaned through our proprietary AI. Parents can view their child’s activity data and insight data through the app.
No. While Moxie’s content is planned to themes and missions geared towards developmental goals, some conversations utilize our proprietary SocialX™ platform to help allow for more full and flexible discussions. This platform “learns” from users, just as we hope children learn and grow when interacting with Moxie. The use of SocialX, provides the ability for Moxie to have more relevant conversations on a wide range of topics instead of scripted responses that fit only limited topics. The software we use was designed to screen out inappropriate and sensitive topics (like bad words), but we can’t always specifically predict what Moxie will say as part of the conversational interaction with a child. Because of this, we encourage parents to talk to their child on topics, missions, themes, and other content that may arise when using Moxie.
Our focus on children’s privacy and security has shaped how we handle the personal information we collect. Because our privacy and security by design philosophy has shaped our data management practices, which are designed to “uncouple” personal information from the parent account or the device, Embodied has no ability to notify parents about conversations between their child and Moxie.
We do not sell you or your child’s personal information to third parties, and we have developed careful data management practices grounded in our philosophy of privacy and security by design. For example, some information, like video data, is only processed on the Moxie robot. Certain information goes to our cloud storage databases, where we implement procedures to restrict access by our third party partners, or to prevent them from accessing the information by implementing security features like encryption. We only share some personal information with trusted third parties, like our cloud service providers, who agree to safeguard the data with care and to only use it to help us provide our services under written agreements. For example, Audio Data goes to our speech recognition partner who is required to delete the audio recording after creating the audio transcript. The Audio Transcript Data is then sent to another area where it is not readable by the cloud storage provider.
Embodied personnel can access Audio Transcript Data, Primary User Image, Facial Expression Datapoints, Insight Data and other information, but our processes are designed to restrict Embodied’s ability to link most of that information to a particular person or device. We take the privacy and security of the information entrusted to us very seriously. Our personnel are trained to safeguard the personal information of children and families.
Moxie is meant to be a springboard to life and Moxie’s content has been developed to encourage a child explore the world around them and have conversations with family and friends. As a robot mentor to Moxie, the child will be tasked with helping Moxie on missions and many of these missions include engagement beyond interaction with Moxie.
Moxie’s content is also carefully paced to inhibit binging. As the child engages with Moxie, activities will unlock over the course of regular use and never all at once. This pacing encourages regular engagement with Moxie but within healthy limits of time.
Moxie is designed to leverage established child development strategies and our proprietary AI to help support your child’s development. However, there is no substitute for the involvement of parents and other trusted adults in a child’s life. We encourage parents to remain engaged with the child as the child interacts with Moxie, and to ask the child about interactions with Moxie.
At Embodied our mission is to serve children and families. If you have any questions, please don’t hesitate to contact us!
385 E. Colorado Blvd., Suite 110
Pasadena, CA 91101