Tag Archives: gpt

An API for the Human Mind

Patent, , , , , , , , ,

by Dennis Crouch

The pace of technological advancement never ceases to amaze me, and it seems like even science fiction is struggling to keep up with reality. In recent months, we’ve witnessed some truly remarkable breakthroughs in the field of artificial intelligence (AI), and this latest development is right up there.

Researchers have used GPT-style machine learning architecture to decode human thoughts by analyzing their functional MRI (fMRI) brain scans. That’s right – we’re talking mind reading. The AI was initially trained trained by exposing participants to 16 hours of narrative stories while recording their brain activity.  And, once trained, the AI was able to accurately describe the content of their thoughts.  Of some interest, the AI was better at identifying concepts and meaning rather than the actual words at issue.

While this technology has the potential to revolutionize communication for individuals with disabilities or neurological conditions, it also raises some privacy concerns that could involve both reading thoughts and writing them.

I imagine that all this wouldn’t require a fMRI technology and instead could rely upon much smaller safe-for-home equipment such as Functional near-infrared spectroscopy (fNIRS) devices shown in the image above or made by Artinis.

A key for me here is that this is potentially moving toward an API for the human mind and body.  For those outside the software world, an API (Application Programming Interface) is a set of protocols, tools, and standards that allow different computer applications to communicate with each other. APIs provide a standardized way for software components to interact.  An API for the human mind and body would create an interface that allows software systems to access and interact with the brain or and other body parts. Over the past several years, folks have been making huge advances with brain/body connection, and I’m excited about where we are headed in this expanded mind approach.

Enjoy!

An AI Journey From Fractals to GPT

Patent, , , ,

By Dennis Crouch

I recently was thinking back to 1996 and the start of my senior year at Princeton University. Although I was a mechanical & aerospace engineering major, I had become fascinated with AI and so focused my senior thesis on developing a new AI model within the department of electrical and computer engineering. Instead of employing traditional layers, I utilized a fractal metaphor to design the neural networks. The main theoretical advantage of this approach was its potential to offer a deeper understanding of how the network operated, allowing us to peer into the brain and gain insights into its learning process based on the structure created. Furthermore, the model facilitated greater human control and direction.

This past weekend, computer law expert Van Lindberg reminded me of the Dartmouth Summer Research Project on Artificial Intelligence that had a game plan of solving AI during the summer of 1956. My senior thesis project took 8 months, and met with roughly the same (lack of) success. I’ll write more about this later, but the project was one of the first to use massive parallelism offered by human players across the internet as the learning model. That part was a big success — as well as the Applet front-end developed mostly by my partner Ryan Kotaro Akita. 

I feel like things are coming full circle for me, but this time the AI model has improved exponentially. The rapid progress in human-machine interaction and generative AI is astonishing. I find myself constantly exploring new, innovative technologies poised to disrupt bloated organizations. The venture capital landscape is necessarily shifting, as small teams rapidly develop and release disruptive products and services with higher speed but lower financial requirements.

Today’s exploration for me is autoGPT that allows users to stack various traditional and AI services to create the best autonomous assistant that I have seen so far.  After being provided with general instructions, autoGPT can generate a dynamic project plan that it executes through online interaction and GPT-4 training results. This technology has the potential to equalize expertise, much like what happened with chess several years ago. However, unlike chess, which relies on a fixed board and finite options, this new model addresses real-time, real-world problems. Indeed, it’s a fascinating time to be alive.

This setup is wonderful but also so scary. Awesome in all senses. The AI is ruthless and without emotion or wisdom. It empowers anarchists, terrorists, and reckless operaters to inflict significant harm. GPT layering enables more sophisticated attacks that exploit a combination of human-social and technological weaknesses in a massively parallel manner. Scams, both big and small, are becoming increasingly easier to execute. Of course, there will be those who use these tools to fight for good. As the battle of technology unfolds, life may begin to resemble a futuristic graphic novel more and more.

What are your thoughts on where we are headed?