Digital Humans
I'm an Informatics student at Luddy and the student project lead for Digital Humans. The product question is not whether AI can seem human. It is whether AI can help people show up more honestly without replacing the human relationship.
My working rule: coach the person, protect the connection, and never let a persuasive interface hide what it is doing.
The work
Digital Humans are not just avatars. They are interfaces that borrow the social cues of a person. That can make a system easier to approach and easier to misunderstand.
One proposed future is an AI concierge that talks with other people's concierges, filters the noise, and brings you only a few human matches to meet. It sounds efficient. It is also intimate, consequential, and easy to get wrong.
Before asking whether people will adopt it, I ask what the system can infer, what people can see, what they can consent to, and who is accountable when it fails.
In HCI (human-computer interaction), disclosure cannot be a footnote. It has to match what the system actually does, at the moment a person needs to decide.
When the interface becomes a proxy
In a 2024 Bloomberg Technology Summit conversation, Whitney Wolfe Herd described a dating concierge that could learn what makes a person vulnerable, talk with other people's AI concierges, and return only a few humans worth meeting. The promise is less burnout. The uncomfortable part is that the system would handle the most personal context before two people ever meet.
- Promise: reduce the endless scrolling and small talk that make online dating feel like unpaid labor.
- Risk: turn insecurities into data, preferences into rankings, and a recommendation into a decision that feels inevitable.
- Design test: disclose what the system inferred, show why it made a recommendation, and let the person correct, refuse, or leave the process.
That is why the idea drew both excitement and a Black Mirror comparison. A useful concierge can reduce friction. A persuasive one can quietly become the relationship.
Her public AI thesis
This page uses publicly reported statements, not imagined private beliefs. Across interviews and product discussions, Wolfe Herd's position is consistent: AI should help people build healthier relationships while keeping human connection, authenticity, and agency at the center.
1. Start with self-improvement
Her concierge concept begins with a private conversation about relationships, breakups, insecurities, and patterns a person does not want to repeat. The purpose is coaching: help someone understand themselves and communicate better, not diagnose them or make the emotional decision for them.
2. Use AI to reduce dating burnout
She describes AI learning from a person's profile and preferences, then reducing a huge field of possible matches to a small number worth meeting. The promise is less scrolling, less repetitive small talk, and more time spent in actual human connection.
3. Keep the human relationship as the destination
She has drawn a boundary against virtual boyfriends, virtual girlfriends, and replacing human partners with bots. AI can be a coach, filter, or concierge. It should not become the relationship it was supposed to help someone find.
4. Protect authenticity
Her public guardrails also reject falsified photos, deepfakes, and AI-generated messages sent as if they came from the user. The useful version of AI helps a person present their real qualities more clearly; it does not manufacture a more convenient person.
Sources: CNBC, TechCrunch, Fortune, and NBC News.
The product boundaries
The most interesting version of this future is not a virtual partner. It is a coach that helps a real person become more legible to themselves, then gets out of the way when real human connection begins.
- Coach, do not replace: help someone understand their patterns and goals. Do not ask them to fall in love with the system.
- Improve the signal, do not fabricate the person: help someone choose a clearer photo or say what they mean. Do not deepfake their appearance or ghostwrite a false self.
- Keep the human message human: private coaching can support reflection, but the words sent to another person should remain the user's responsibility.
- Use ML for nuance, not isolation: understand specific compatibility needs inside one shared ecosystem instead of forcing people into separate boxes.
This is where the product earns its right to exist. It should make the real person more capable, not make the system more difficult to leave.
Accessibility is emotional equity
This is how I translate the empowerment logic into product work: accessibility is not a checklist after the launch. If the interface creates extra friction for disabled, neurodivergent, tired, grieving, or overwhelmed users, it makes them work harder to reach the same human connection.
1. Reduce burnout at the interface
A platform that requires fast swiping, perfect vision, precise motor control, or effortless social parsing is not neutral. It makes the people already carrying more friction work harder. Semantic regions, screen-reader announcements, captions, text alternatives, and calm pacing make the private space usable before it becomes intimate.
2. Make safety readable
You cannot ask someone to share a breakup or a relationship boundary while the interface clips text, hides state changes, or relies on sound. The user needs visible labels, scalable text, high contrast, clear "AI is thinking..." status text, and a simple way to edit, undo, or leave.
3. Let the interface adapt
Voice can be an option, never a requirement. Keyboard and switch access must work. Motion needs a reduced-motion path. Match explanations and private reflections need a clear focus state and a reversible next step. The system should adapt to the person's needs, not demand that the person become fluent in the system.
The standard is emotional as well as technical: if a system asks someone to show up authentically, it has to give them a perceivable, navigable, and reversible way to do it.
Prototype interaction
AI Sanctuary
Private reflection · no account connected
- ConfessShare only what you choose. The assistant helps you name a boundary or pattern.
- UnderstandTranslate your words into values and preferences you can inspect and correct.
- Hand offLet authorized assistants reduce the noise, then return the decision to you.
Tell me about a boundary you wish you had set in your last relationship.
Reflective insight
It sounds like clear communication is important to you when things start to change. We can look for that value without turning one difficult experience into your whole identity.
Progressive disclosure
The Commons
Three possibilities · human choice required
Sam · open communication
Why this surfaced: both assistants noted a preference for direct conversations and independent weekend routines.
Jordan · steady transitions
Why this surfaced: both people described wanting clarity when plans change. The recommendation is an invitation, not a verdict.
Riley · curious and grounded
Why this surfaced: both profiles value learning, honest pacing, and room to say no before saying yes.
Never: fabricate a person's identity, replace a human partner, or send automated messages to a real person without their knowledge.
PIT-UN
PIT-UN (Public Interest Technology University Network) gave me a go-to-market lens: a useful idea still has to reach people, explain itself, and survive contact with real use. Serve-AI was founded via PIT-UN, separate from Serve IT.
NSF research
Earlier NSF-related research work gave me the first version of this question: can people understand what a system is asking of them and still choose to use it? Digital Humans is where I keep pressure-testing that question.
Undergraduate research with IU Undergraduate Education
With IU Undergraduate Education, Aug to Dec 2025, I studied privacy and consumer technology through one practical test: can people tell what they are consenting to before the system moves them forward?