Dr. AI

Thank you both for your comments! These are super helpful. I’ve updated the description of this breakthrough per your feedback, @MachineGenes.

To @ymedan’s point: I think we’re talking about two breakthroughs, one for people and one for caregivers.

  1. AI-powered (predictive) model that collects data from biomarkers, body scans, sensors, wearables, and user input. Contextualized to culture, place, and individual needs and wants. Integrates deep learning to predict preventable acute care. Enables users to understand their health at any time, make a plan for their health, receive actionable insights, and avoid sick-care.
  2. Doctor AI. Clinical tool that synthesizes relevant medical research in collaboration with clinicians, analyzes electronic medical records, and makes hypotheses and/or explainable and interactive recommendations to caregivers, empowering them to decide a consequent collaborative course of treatment.

Thanks for clarifying. I am OK with #1 except with the term “acute”. Most health cost burden is due to chronic conditions, which should be cured or proactively managed to avoid becoming acute. So preventing and curing chronic conditions should be #1 goal.

@MachineGenes
You make a valid point. I would argue that the cost of care will become prohibitive so that the cheapest and most cost effective option will be to stay healthy, even if there is a privacy penalty.
So far I have not seen hostility to wearables like mobile phones, smart watches, Fitbit, Oura ring and alike. If NEST or Aura Air monitor air pollution condition at home without having a personalized profile, I don’t see why people will object.

Of course, the wealthy can afford to get sick and keep their privacy intact, but most people won’t.

Time will tell…

Good point! Thank you.

Here’s an emerging answer
Cash for Pounds: Prospr Health Is Using Cash Incentives to Encourage Healthy Habits

Rather than saying ‘integrates deep learning’, say it ‘integrates continual learning architectures such as deep learning’. Point is that there are other, emerging forms of ML completely different from deep learning artificial neural networks (ANN) that haven’t been publicized yet (one of which we demonstrated in the AI XPRIZE), and which have equivalent or better capabilities to deep learning ANN.

1 Like

Will do! Thanks, @MachineGenes.

@akastner, @eakinyi, @yanchen, you may have thoughts on this discussion as well! Please let us know what you think.

To recap, here are the two AI-related breakthroughs we’re looking into for health:

  1. AI-powered (predictive) model that collects data from biomarkers, body scans, sensors, wearables, and user input. Contextualized to culture, place, and individual needs and wants. Integrates continual learning architectures (like deep learning) to predict preventable care. Enables users to understand their health at any time, make a plan for their health, receive actionable insights, and avoid sick-care.
  2. Doctor AI. Clinical tool that synthesizes relevant medical research in collaboration with clinicians, analyzes electronic medical records, and makes hypotheses and/or explainable and interactive recommendations to caregivers, empowering them to decide a consequent collaborative course of treatment.

So the next question is… what if you have technology that does all of the above, for a specific medical condition? Which breakthrough competition would one apply for? Perhaps these two breakthroughs should have a single umbrella competition over the top, given that they are going to have an overlap.

(This is not a hypothetical question. The technology we were demonstrating in the AI XPRIZE is expected to be able to do all of the above, but we were unable to secure funding in time for a clinical demonstration for the semi-finals.)

1 Like

I didn’t think that would be possible! But then perhaps they should be combined into a single prize competition.

Let’s see how others feel about that.

Yes, it’s possible. We were trying to demonstrate all of the above could be done for type-1 diabetes using new forms of ML and adversarial AI, but ran out of funding at the very end of the AI XPRIZE.

@MachineGenes
While it may be possible, it may not yield the optimal results. By separating diagnosis from intervention recommendations we may pair best-of-breed solutions, as well as offer a personalization layer of patient preferences. Even for the same diagnosis, different people may have different ideas about treatment options that they like or dislike.

More over, for different diseases we may have different intervention modules, each optimized for that indication.

1 Like

@ymedan We’re not doing diagnosis (note that diagnosis isn’t mentioned in the final text above). Provided the disease is sufficiently well understood such that its drug-disease pharmacokinetics (PK) and pharmacodynamics (PD) could be expressed as differential equations, we are able to generate personalized medication therapies. The medical conditions I’ve worked with tend to have onerous treatment (hence the imperative for medical AI), so there’s a limit on patient preferences, but we can certainly do interactive human-AI medicated therapy design for therapies that have PK/PD profiles that can be reconstructed using our evolutionary ML.

Important to emphasize that (at least using our technology) this is all disease-specific, so distinct modules would definitely need to be built for each disease.

Sorry, my bad.

Not all interventions are phrmacological… Some behavioural, some digital therapeutics and some involve a device intervention (watch my TED talk)

That’s all good @ymedan. I’ll look up your TED talk!

Hello, I think # 1 is better. Combining with 2 would be too ambitious. I would explicitly mention genetic data in this list (very important of course). Didier C.

1 Like

Engineers have developed the DxtER device as part of a race to create a modern version of “Tricorder,” in which Dr. Leonard “Bones” McCoy waved to patients on Spacecraft Enterprise to diagnose the disease.

DxtER combines a variety of sensor arrays and intelligent diagnostic software in a package of less than £ 5, said Philip Sharon, a technical design expert who is a member of the team that created the device. Read more.

Has this device seen the light of day?

This discussion is relevant to 2 of our 4 prize ideas for health:

  1. Digital Twin
  2. Last-Mile Medical Care

I hope you’ll continue to share your feedback there!

Surely all of these make for good assessment criteria when judging various candidates for the prize?

1 Like

#1 and #2 are both possible. We were racing to demonstrate both for the AI XPRIZE but ran out of runway.

Explicit genetic data is not necessarily essential, if you’re reconstructing PK/PD at an organ scale, in which case the genetic information is essentially implicit in the organ-scale parameters.