12I/ Identity in the Age of A.I. - Identity Crisis? “An open conversation”
Identity Crisis? - Identity in the Age of AI
Session Convener: Wenjing Chu
Notes-taker(s): Wenjing Chu, Shannon Wells
Tags / links to resources / technology discussed, related to this session:
Discussion notes, key understandings, outstanding questions, observations, and, if appropriate to this discussion: action items, next steps:
- Is Digital Identity - as defined by structured trusted data - the Wrong Question?
- What is Identity in social science and humanistic sense?
- Digital Identity is trying to capture a partial representation of that - it’s always a partial representation. All models are wrong.
- AI can build a similar partial representation through deep learning for example, unstructured model but a model that can be very accurate/effective - it is therefore also an Identity
- Which kinds of identity is “better”
- Which kinds of identity is more effective in addressing the problems we face (e.g. principles of SSI etc.)
- Case studies
- Biometric authentication
- AI bots
- Social media intermediated by AI
- Bots on social media
- Who is learning? You or the Machine.
- A language between human and AI
- If you are interested in continuing this conversation, join me in a new ToIP Task Force for AI and Trust that is coming soon. Ping me on ToIP (Trust over IP Foundation) slack channel. Or Email: chu.wenjing AT gmail D O T com.
Thanks for all of your questions and interests.
Notes from Shannon Wells:
In one of the first slides, was shown that there are two ways to conceive of an identity. One is to build a descriptive structure, for example, a list of observable features, test results, facts about someone, etc. However these things can all change.
The second is to take everything observable about the person and put it into an AI that will create a “learned structure”.
Keeping a descriptive structure up to date is very hard. People age, they gain and lose weight, they change their hair, maybe they are injured, they change jobs and have children, etc. But we humans continue to recognize someone who has been through these things and an AI may learn even better.
“Rather than digitizing the human world for machines, machines should learn to live in the human world”
In the cases of determining whether someone’s face matches their ID card, or whether someone is over 21, AIs using facial recognition and being trained to comprehend these things outperform people. In both cases there is no need to store anything in a database, so someone’s privacy can be preserved better, rather than storing a bunch of “descriptive” data in a database, even encrypted is less secure and less private than a well-trained robot who performs a single task and then forgets about it afterward. Next there is no “disclosure” to agree to, the person simply walks in the door, or presents an ID card.
These and similar tasks are suited to AIs and can be implemented in ways that not only preserve a level of privacy, and accomplish goals of verifying attributes with high accuracy, but when there are errors they can be addressed immediately instead of being stuck in a database and the person is unable to correct the errors or has tremendous difficulty getting the errors corrected.
In the realm of identity verification, one attendee remarked that smartphone sensors have been shown to be able to produce a profile that can determine whether a phone’s owner is holding the phone based on how it’s oriented, how their hands move, how tall they are, etc.