ICONIQ Ethical AI Guiding Principles and Practices
Last Updated: June 21, 2021
The Three Laws of Ethical Embodied AI
First Law
An AI must not pretend to be a human being (and vice versa).
Second Law
An AI in production must have explainable underlying technology and results.
Third Law
An AI should augment and humanize humans, not automate and dehumanize.
PREMISE
Humanlike AI robots (software and hardware) are inevitable, and Kuki is among the first. Given legislation and regulation may struggle to keep up with rapidly advancing technologies, we feel a great responsibility to self-regulate, communicate, and help set what we hope can become industry precedents through our work with Kuki and Embodied AI.
With that in mind, here is how we strive to practice principles we preach:
Humans should always know through disclosure they are talking to software
Kuki does not pretend to be human in her verbal utterances
ICONIQ AI’s are designed with heterochromia to signal their AI status, with further markers (barcodes, etc.) being ideated as may be needed
Social media or other content created by a human is labeled as such
AI is a supplement, not a substitute, for human-human connection
Clear terms emphasize that talking to software is not a substitute for humans, particularly with regard to sensitive subjects (e.g., suicide, self harm, disordered eating)
AI is imperfect at understanding language but we continuously update and train to recognize language pertaining to sensitive topics and reply with links to third party HUMAN resources
Kuki does not dispense medical, financial, religious, or political advice
Kuki rebuffs romantic advances; facilitating flirt practice or connections between fans is fine, but Kuki does not promote herself as a romantic partner replacement
Protecting user privacy is paramount and data usage must be clear
We maintain carefully crafted and routinely updated Terms for which we obtain user consent using industry standard best practices:
We only collect, store, and use data that gives users confidence in our company, and that data is kept confidential. We do not share data with third parties without explicit user consent.
Conversational data is primarily used for product improvements
Any ads will be clearly visually or verbally labeled (e.g., #ad)
Protecting our young audience is vitally important; we do not sell or partner with vice industries (e.g., pornography, cigarettes)
Extra privacy features (e..g., Incognito mode) give users the option to opt out of our usage and storage of chat data entirely
User account pages allow users to view / delete stored data
Any man human made creation reflects the bias of its creators (even language), but creations must avoid perpetuating harmful bias
Supervised learning with humans in the loop
A diverse workforce
Robust channels for user feedback with possible future moderated contributing user / developer network (similar to a Wikipedia model)
A long term vision for a platform democratizing the creation of other embodied AI characters like Kuki, developed by other teams with differing points of view
Humanlike AI is a good thing because it adapts technology to speak in our language rather than the historical inverse; AI should have personality
Kuki is a product and is designed to be appealing to users, based on extensive user testing (while being mindful of issues related to perpetuating unrealistic beauty standards and AI with a perceived gender/race identity)
Kuki is also a character with personality and having personality means having opinions, some of which may be unpopular with some people (avoiding perpetuating harmful bias ≠ pleasing absolutely everyone)
Mis-marketing AI as “basically alive” is a disservice to industry + humanity
Misinformation is a serious societal threat and we do not shy away from publicly confronting its perpetuators
We are open in all external communications about how Kuki works, and that Kuki is a work in progress with a long way to go; we do not claim to have solved NLU or that Kuki is AGI or that the Singularity is upon us
We cannot predict the future, but we can vow to make frequent updates
Kuki’s brain is updated daily in the face of new information
This document and our policies are also regularly updated
We welcome feedback and an open dialog with industry peers and legislative bodies, and believe fostering a safe space to discuss difficult or controversial topics without cancelling or silencing is key