Learn with Phoebe · Applied AI

Learn AI Digital Human with Phoebe

Six sessions, 45 minutes each, and one decision you can defend in a room. Most courses on this subject start with how to make a synthetic presenter. This one starts with whether you should, because that is what you will be answerable for. Along the way: what each vendor genuinely requires, why the on-screen face is the weakest lever in multimedia learning, and what happens to every video already made when the person in them leaves.

6sessions
3vendors, three answers
24→ 84% believability
6% trust once found out
45min per session
A studio monitor showing a composed, well-lit presenter mid-sentence. The chair in front of the camera rig facing the monitor is empty, and the key light is still on.
The chair is empty and the light is still on. Everything the audience needs to believe is already on the monitor. Nothing on it tells them which of the two to trust.

The six sessions 🎭 · one artifact, built across all of them

Bring a real decision. A presenter you are considering, a colleague who has been asked, or a policy you have been told to write. By session 6 you have a written position you can defend.

⚖️ Session 1 · easy

Should you at all

What consent each vendor actually requires, what enterprise money exempts you from, and why a meta-analysis of 86,000 people says nobody can tell.

▶ Start here ✍️ Session 2 · easy

Consent and likeness

What a consent form has to contain to be worth anything, why live capture beats an upload, and the clause about leaving that no vendor will write for you.

the contract 🎭 Session 3 · medium

Making one that works

Footage requirements, prosody, and the four craft levers that take believability from 24 to 84. Then the point where craft stops helping.

hands-on 📊 Session 4 · medium

What the evidence says

The on-screen face scores d = 0.22 while the script register scores 1.11. Engagement falls with an AI instructor and measured learning does not.

uncomfortable 🚨 Session 5 · hands-on

When it is used against you

Arup lost 25 million dollars. LastPass lost nothing and published why. Three seconds of audio is enough to clone a voice.

defence 📋 Session 6 · hands-on

The policy you have to write

Article 50 puts a duty on you as deployer. Four questions no vendor answers, which means your organisation has to.

🔒 unlocks the ceiling
easy - everyone medium hands-on hardest

Your passport 🎫

Clear all three quiz questions in a session and it gets stamped. Stamps live in this browser only.

Choose your path 🗺️

Three doors into the same conviction: the craft question and the consent question are different questions.

🔥 Someone has asked me to approve this 1→ 2→ 6 the decision, the contract, then the policy
📊 I need to know whether it even works 3→ 4 build one, then read what the research says about whether it helped
🎭 The whole journey 1→ 2→ 3→ 4→ 5→ 6 from a tool you were shown to a position you can hold

What is real here 🧾

A course about synthetic media owes you an unusually precise account of its own claims.

ClaimStatus
The viseme track and the mouth that follows your typingReal. Computed from whatever you type, grapheme by grapheme, in your browser
The believability and trust scoresA teaching model with two deliberately separate axes. Labelled as a model on the widget
Every consent requirement, price and policy quoteQuoted from the vendor's own policy, terms or help page, each with the URL, dated August 2026
The on-screen face is the weakest multimedia leverMayer's image principle, d = 0.22, against personalization at 1.11. Corroborated by an N = 108 trial finding lower engagement and identical performance
Whether disclosure reduces credibilityContested, and taught as contested. Single experiments find a penalty; a 47-study systematic review finds no consistent one. Nothing finds a benefit
Any FBI or FinCEN dollar figure for deepfake fraudDeliberately not stated. Both agencies published alerts and neither publishes a loss total

The knowledge map 🧠

The whole course at a glance - hover a session to spotlight its concepts, click any node to jump in.