New AI Possibilities

I came across this youtube video in my AI research. https://www.youtube.com/watch?v=tPCJDkHh3lw

Fair warning, I would surmise this is more of a fringe application of AI than you’ve encountered prior.

In summary, PropheticAI has positioned their company’s vision “as being dedicated to building gateways to Human Consciousness, pursuing answers to life’s biggest questions”,  with their Morpheus-I product being capable of inducing lucid dreaming within its subjects – essentially triggering an extraordinary state of consciousness.

Their AI-driven system utilizes a wearable headband prototype device (called halo), developed by a company called Card79, which is capable of reading EEG data of subjects, and performing low-frequency TFUS (Transcranial Focus Ultrasound Signals). The headband used dual transducers, and a unique AI transformer control capability, allowing it to project acoustic holograms, in the form of ultrasound pulse patterns in three dimensions, within the brain.
The AI model’s dataset used is unique as it is organized temporarily, tying together both fNIRS  (Functional Near Infrared Spectroscopy), which captures hemodynamic activity in the brain, cross-indexed with EEG (Electroencephalography)  data, which captures electrical activity in brain. This was created via a partnership with Donders Institute in Netherlands (of Radboud University) to generate the largest dataset (so far) of EEG and fMRI data of lucid dreamers
I am sure there are more than a few questions that come to mind with this product: Is it safe? Are there long term effects? Will it affect my quality of sleep? What’s the true value proposition on being able to lucid dream? Do I really need such a device?
When I put my “product manager” hat on, yes I do see some definite market challenges here, but hold on, let’s go a bit deeper, as this is a product that utilizes a signal-then-measure-response approach, accumulating bio-feedback (EEG) data to further refine their AI model. Could we ante-up on this approach?  Perhaps one could incorporate an alternative bio-feedback capability such as fNIR (near infrared spectroscopy) which is also now a mobile technology, and much cheaper alternative to fMIR. Now we’ve pulled together an incredibly powerful biofeedback capability.  Ah, my apologies, went off on a tangent… let’s get back to the value proposition of this AI driven bio-modulation creation.

By utilizing an adaptable, learning AI controller which manages the bio-modulation, this tool can become very capable at adapting to every unique wearer (as I’m sure, not every person’s brain is the same). We are on the cusp of developing devices that are capable of performing low-invasive modulation functions to our brain (using low frequency TFUS), effectively “nudging” us to a preferred brain activity, (and some of these applications have size extra-large value-to-market),  with the possibility and potential to:

• Manage or possibly cure insomnia
  
• Manage or possibly eradicate anxiety and/or depression
  
• Enhance cognitive function
• Stimulate neurogenesis and stem-cells
• Stimulate anti-inflammatory responses and trigger the innate immune system  (to counter degenerative conditions from diseases such as Alzheimer’s)
  
• Manage ADHD
  
• Manage OCD
  
• Manage Autism
  
• Manage or eradicate addiction
  
• Create a non-invasive communication interface (to control technology, or possibly to enable communications with patients in coma or vegetative states)
  

This technology could also provide the capability of incisionless neurosurgery to address issues, some of which are already employing direct electrical stimulation methods such as DBS, tDCS, tES, and ECoG. TFUS can either stimulate or selectively kill specific (problem) areas in the brain (although it utilizes a higher-frequency ultrasound signal to accomplish thermal-ablation). We could have alternative treatments for such conditions as:

• Parkinson’s
  
• Dystonia
• Epilepsy
• Tourette’s
  
• Chronic Pain and/or Neuropathy

Just as a disclaimer here, I’m not a neurologist – but certainly would love to hear from the specialists on this.
As in all innovative change, we tend to learn in one area, and apply that knowledge to another, sometimes allowing us to literally take leaps from our initial capabilities and understanding. I do believe there is a minor issue that does need consideration, and that’s our current ability to have AI’s cross-learn from one another. What’s interesting here is that we can train and refine (in the field) an AI, to a point where it exceeds expectations, however, when we look under the covers, we really do not understand why it works so well.
We certainly have some approaches to cross-train AI models, but these are somewhat manual techniques: knowledge-distillation, model ensembling, data-augmentation, transfer learning. Forgive me but I’ll not dig into those methods further, suffice it to say I’m far from a specialist in this area, but wouldn’t it be nice for AIs to be able to train other AIs on everything they know about a specific knowledge area?  Maybe someone is already working on this problem, but I suppose, that would require some standardization.

Hopefully this provided you some food for thought, and I'm certainly interested in what you think about the new possibilities just over the horizon.

Exciting stuff!

Some references info if you are interested:
A great background in Ultrasound Modulation: https://www.youtube.com/watch?v=8cSDYx7UoBY
https://www.dailymail.co.uk/sciencetech/article-13048401/ai-headband-dream-control-long-term-effects.html
https://www.vice.com/en/article/m7bxdx/scientists-are-researching-a-device-that-can-induce-lucid-dreams-on-demand
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9124976/
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5398988/