Real Fidelity, personal connections and tutoring
What part of "real" matters?
Gwern has a very interesting post on “open questions” he is personally interested in (h/t MarginalRevolution). I recommend reading the entire thing. One question in particular jumped out at me:
Face-to-face meetings, even brief ones, appear to cement personal connections of trust and liking to an extent not achieved by even years of more mediated contact like phone calls or Internet text discussions / emails / chat; this appears to be true in almost every context, even ones like British inventors meeting their heroes (in a different field) just once, with large step functions in connections despite the apparent near-zero marginal information conveyed by a brief physical visit after long-term interactions & track records. (This might be related to “Bloom’s 2 sigma problem”11.)
Is there something qualitatively different about personal meetings, and if so, where is it? Is it eye contact? Body language? (It’s probably not pheromones.)12 Is it mere physical proximity and a certain “inability to suspend disbelief” about a technologically mediated person? Can large wall-sized TV screens for teleconferencing achieve the same effects as regular conferencing? Or do they need to be 3D? What about VR headsets, are they adequate already with avatars and hand-tracking gestural control, or do they require eye tracking, or facial expression mapping? How much is enough?
Bloom’s two-sigma problem refers to the discovery that 1:1 tutoring seems to provide a 2-sigma improvement in student performance. The “problem” is how to obtain that level of improvement in a less expensive way. Asked a year ago the answer was “there is no way to have a personal tutor for every single individual without bankrupting the planet”, but all that has changed with AI. Khan Academy’s personal AI tutor is in beta (sign-up for the waitlist here - cost is $20/month). Even without Khan Academy bells and whistles anyone can use ChatGPT to get a tutor in any given subject. But it does not seem likely that a responsive generative text tool, no matter how sophisticated, is going to result in 2-sigma improvements in student performance.
What’s missing?
Gwern suggests that maybe something like a big screen TV, or VR might be able to re-create that impact that happens from in-person experiences. I think that is the right track. AI is getting all the attention now, but the latest VR/AR is really outstanding, and the sense of presence when you use a Meta QuestPro is real (or at least it feels real). Maybe if something like th Khan Academy tutor is mixed with the fidelity of a VR simulation (where you can see and talk to the tutor rather than type to it), we really might see a 2-sigma improvement in student achievement.
The Learning Bundle
When I was at General Assembly we talked about why students paid us $15,000 for a 3-month course of content that was, if they looked hard enough, basically available for free on the internet. The reason was that we were not selling access to content, we were selling a bundle:
Teaching the content
Curation of the content (WHAT to learn)
Forcing mechanism (“I can’t walk the dog today, I have class” is a lot easier than “I can’t walk the dog today, I am learning from the internet in my room on my own”)
Career Coaching (Mostly how to write resumes and how to interview)
Peer networking (other people going through the same thing you are at the same time)
The end result was that 97% of students that were looking for a job at the end of the course would get one (in the field). I think it is unlikely that 97% of people who tried to learn the material on the internet for free would succeed in a career change after 6-months. No one would pay for any individual part of the bundle alone. There are products out there that help you follow through on commitments. What would YOU pay for a product like that? $1000? Unlikely. The reason students paid far more for the bundle than the sum of the individual elements was because they were paying for the outcome: They knew if they paid $15,000 and 3-months of their life we were practically guaranteeing they would succeed in their career change. That is the power of the bundle.
A similar bundle exists for conferences (which is left as an exercise for the reader).
With some (still significant) improvements in AI + VR we may see the creation of a bundle that actually does achieve the benefits of an individual tutor for everyone (who is interested) on the planet. But it needs to be more than just a responsive text generator — it will need to be a curator, a forcing mechanism, AND a high fidelity teacher.
Keep it simple,
Edward
p.s., for more on my concept of “Real Fidelity”, and why it is important, see these essays:
Concerts, Zoom and Real Fidelity (where I coined the term)
Almost Real Fidelity (how real fidelity relates to VR)
Also: check out this Friday post from Ethan Mollick where he shows how to use the Bing AI to create Educational Games about any topic from simple prompts:
