I need help building AI that helps
Authors: Jaan Altosaar Li, PhD
Edit this page or propose changes
Last year was hard:
I wasn’t sure whether I’d be able to work.
I declared disability. I was experiencing PTSD symptoms. My workplace at the time didn’t help, and I was in debt to pay for therapy to recover. My mom lent me money to cover rent.
I had failed and felt cornered. My goal had been to build AI to benefit mental health. But now I was suffering from mental health issues myself, after I helped my friend indict a sex abuser in a Manhattan courtroom.
The irony wasn’t lost on me—I was studying this shit!
Information asymmetry: institutional access to medical knowledge
I persisted using what I knew best: research. Instead of my big data PhD thesis using AI trained on hundreds of thousands' of people's data, this was the
N=1 kind with me as the subject. I learned about treatment plans on UpToDate, which I could only access thanks to being employed in a hospital campus. (Doctors at top hospitals frequently use UpToDate learn what the latest research says about health and disease to make clinical decisions.)
UpToDate taught me about the symptoms of post-traumatic stress disorder I was experiencing. I also learned about antidepressants, for what my plan might be should first-line treatment with traditional therapy fail.
In the middle of a particularly hard time when I was crying every day and not feeling like cognitive behavioral therapy was working, a friend said I should try acceptance and commitment therapy – it had helped him.
I reached out to Dr. Michael Maher on my friend’s referral, and fought to get the care I needed.
I ended up needing to write 10,000 words and spend dozens of hours on the phone with United to get a “single case agreement exception” approved by United to prove that Dr. Maher's expertise was “medically necessary” and my suffering was sufficient to create a liability. Footnote
Here I was, a tall white dude working on health equity on the top floor of “the first [medical] school in the United States to award the MD degree”, Columbia University Irving Medical Center.
But at last, some of the irony lifted. I was no longer alone; it was clear that something was broken besides me.
Instead of completely giving up, I had wanted to verify and reality test certain facts about the world, in accord with the interaction principles my friend Toby and I had inscribed after our friend's ordeal with sex abuse.
Accept, commit, take action
Now that I was in recovery and had stopped crying every day, it was time to act.
Dr. Maher, my therapist, saw that I was thinking about starting a nonprofit to escape academia and recommended I talk to Steve Hayes, one of the most cited scholars in the world, and the creator of acceptance and commitment therapy. Steve replied to my sophomoric email and he is now on the board of the One Fact Foundation to help deploy our AI for mental health to countries in which it is most needed.
Throughout this journey I have stayed connected to how many things I do as a 6’6” white male that few cannot – how many people can pore through expensive medical resources, loan money from their mom, and appeal a denial three times just to stop crying every day?
The Magic Grant
Galvanized, I knew I needed a lot more help than I had access to, and started talking to hundreds of people. One of them suggested we might apply for the Brown Institute Magic Grant program. The program name quotes Helen Gurley Brown, the longtime editor of Cosmopolitan Magazine, when she was describing a particularly hard time in her life when “she just needed a chance, and someone did a little “magic” to help her on her journey”.
My friends Susan, Brad, Rohan, Maxim and I applied to the Magic Grant program with this video:
In it, we describe why we need to start a nonprofit to be able to use artificial intelligence models to collect, standardize, and distribute the hundreds of terabytes of the hospital prices that are public by law - to reduce increase the power of patients to shop for quality health care they can afford.
I had been building AI for a while and knew that building this technology could make millions while increasing global inequality and information asymmetry. (Today, it is even more pressing to ensure AI models that can influence decisions at scale are built equitably, backed by nonprofits, and imbued with values such as transparency and non-discrimination that can be publicly verified.)
We got the grant – $55,000 – and with the money were able to collect 4000+ hospitals’ price sheets (through a partnership with DoltHub; see the first blog post about this work here which got on Hacker News).
We built a rockstar team. We have now started making maps of all the hospitals in the country, such as on this page in support of a recent film:
Building our team: we need your help and donations!
With the funding we raise, I can’t wait to continue hiring as diverse team as possible on as many axes as possible: women, immigrants, non-binary folks, neurodiverse individuals, and the many more categories that the people with whom I am lucky to work choose to associate with.
For example, three people we work with would’ve had to leave the United States after their degrees at Duke and Columbia. We helped them find employment through our nonprofit and they are staying in the country.
Similarly, our Community Advisory Board and open source community of contributors are also necessary (yet, not by mere existence sufficient) requirements to build technology that serves the diverse unmet needs of disadvantaged populations worldwide.
It’s now a year later after making the above video, and I’m able to work again. I’m off meds.
I got my green card.
I got engaged.
I get to work with the creator of acceptance and commitment therapy that helped me recover, and Power to the Patients which is supported by Fat Joe, Busta Rhymes, Rick Ross, and French Montana (our team attended and documented the kick-off on April 27, 2023).
With all this bipartisan help from the government, institutions, musicians, scientists, community organizations and more, we are ready to scale our work globally and need your help to do so:
For the TD Five Boro Bike Tour this year, I ask for money that will enable us to reduce the cost of health care as quickly as possible – first in the United States, then worldwide – using open source, equitable artificial intelligence.
Might you be willing to contribute?
- Cash app: give.onefact.org/cash
- Stripe: give.onefact.org/stripe
- Paypal: give.onefact.org/paypal
- Venmo: give.onefact.org/venmo
- Ether: give.onefact.org/ether
- Checks to:
One Fact Foundation, 2093 Philadelphia Pike #1764, Claymont, DE 19703
- Via Schwab Charitable or Vanguard Charitable: look up EIN number 88-2145154 in the dashboard.
Your contribution is tax-deductible to the fullest extent under law; the non-profit 501(c)(3) designation our corporation achieved can be verified here:
- At the Internal Revenue Service list of nonprofits incorporated in Delaware (search for employer identification number 88-2145154)
- At Guidestar, where we earned a Gold seal of transparency
- Through the notification of our 501(c)(3) designation by the Internal Revenue Service
Please email me if you have any questions, need further information, or might be interested in contributing with pro-bono consulting, software engineering and AI research, design, or communications: firstname.lastname@example.org.
P.S. I also teach anyone how to use GPT, and understand the principles that govern its behavior and implementations; this is a 165+ learner pilot at the University of Tartu. The course materials are publicly available, for free, at https://www.datathinking.org/university-of-tartu.
P.P.S. In the spirit of informed consent, acceptance and commitment therapy, and the mental health work I do, I think it is important to mention that I use these principles in fundraising as well. You can even learn about these principles in our budding employee handbook.
I followed this guide and this one to write this appeal.
The appeal reads pitiful: I beg for help, I have insight into what I am doing.
My writing is also clinically inaccurate because of the way I felt forced by insurance to write about OCD when I actually had symptoms resembling PTSD and depression. OCD was ICD code F42.2 in the theater of medicine's word salad, and I didn't want insurance to have one more reason to deny my care.
I paid $7875 out-of-pocket with a salary of $70k/year, I got reimbursed months later for most of it, and this is what it took. Dr. Maher said I was the second person in ten years to do this, and Dr. Maher has treated 10,000+ people and is a leading expert at The Reeds Center which he founded.
If you need help paying for therapy, please reach out to email@example.com and we will do our best.