MW #1 | We Could Learn to Think More Clearly, You and AI
Plus Embodied AI, Brains in VR, and The Imitation Game w/ GPT-3
Welcome to Mind Whoosh! A newsletter series on thoughts, learnings, and speculations from my spelunking journey through minds, metaverse, and this messy meatspace called reality.
In this issue:
- Tackling Fear & Discomfort — Ways to find calm in chaos (because Science!)
- Embodied AI — a pathway to self-conscious machines? (and Ai-Imposter Syndrome?)
- The Imitation Game — AiCore built a Turing Test with OpenAI’s GPT-3 model for YOU
- BCIs + Virtual Reality — Neurofeedback in XR seems just on the horizon
Tackling Fear & Discomfort with Neuroscience
As you make the arduous commute from your bed to the computer in your favorite onesie pajamas( my latest favorite daily dress ), you might want something to help find a sound headspace to start your week. This is an awesome conversation with Dr. Andrew Huberman on the Rich Roll Podcast, focused on insights from his neuroscience lab’s extensive studies of fear.
Change Your Brain: Neuroscientist Dr. Andrew Huberman on the Rich Roll Podcast
Reflecting on this conversation conjured personal memories of wrestling with fear while growing up near the ocean. Learning to surf as a kid, I’d often find myself unprepared to face the mercy of an aggressive set of waves bearing down upon me. The surfboard leash would get tangled around my legs as I tumbled and rolled beneath the surface for what felt like ages between breaths of air, stuck in a washing machine.
In those moments, you quickly learned the only thing you could do was stop struggling, embrace the surge as it passed, and finally follow your bubbles sideways back up to the surface. Dr. Huberman’s insights helped me to appreciate these kinds of early experiences with finding calm in the chaos.
Embodied AI & Reinforcement Learning
Creating artificial general intelligence is often cited as one of the over-arching goals of the field of AI. As someone is deeply drawn to the exploration of minds and their mechanisms, this always seems to be one of the most intriguing prospects to consider. Might we eventually learn to replicate and even surpass what evolution gave us, but in synthetic form? Can minds create other minds?
While fantastic progress has been made in machine learning over the past decade, there remain domains where AI systems struggle to match humans. Open-ended creativity and problem-solving in complex, dynamic environments have become the focus of many teams in recent years. In the most successful examples, a combination of methods is often used to achieve better results.
Overview of Embodied Artificial Intelligence
I recently had a conversation on using reinforcement learning within complex, unstructured environments. Through searching for better ways to describe the challenges therein, I came across a great overview of what is being coined as ‘Embodied AI’. Self-driving cars and competitive video games like Dota are among the most well-known examples. The term “Embodied” feels like a good fit, as it implies a central ‘self-concept’ in comparison to others in an environment. It would almost seem to describe an ego, which we use constantly to navigate our squishy meat-space reality while we learn to thrive( and drive ourselves crazy ).
The article does a great job of breaking down the terminology being used among teams at Facebook AI Research and Intel Labs to describe embodied systems and how they differ from other approaches. They also link to more info on Reinforcement learning specifically, which feels like one of the most promising areas of research in the field right now.
Overview of Reinforcement Learning
The Imitation Game with GPT-3
The team at theAiCore, an online machine learning school based out of the UK, will be running an online Turing test using OpenAI’s GPT3 natural language prediction model API. It could be an interesting demo of the newer model’s capabilities. For participants, it looks like you’ll have 50/50 odds of talking to a GPT3-based Bot or another human. There has been a lot of press around this particular algorithm and its ability to produce full paragraphs of coherent text.
Brain-Computer Interfaces + Virtual Reality
It’s exciting to see that major players in the VR industry are taking a serious interest in combining Brain-Computer Interfaces (BCI) with virtual reality. People like Gabe Newell of Valve putting serious investment into partnerships with OpenBCI and Tobii is a good sign that across the industry, more is just over the horizon. Currently, reports seem to anticipate a 2022 release, coincidentally the same time Apple is purportedly launching an Augmented Reality headset of their own.
Valve, OpenBCI, & Tobii to Launch VR Brain-Computer Interface 2022
One of the applications I’m most excited about here is using neuro-feedback to create more personalized interactions with virtual personas. That may be with a character in a game, or conversing with a virtual AI assistant to help you throughout your day. It’s also terrifying to consider the data privacy implications this kind of tech opens up as it becomes more widespread.
That’s a fascinating topic in its own right, so perhaps more on that next time spontaneous inspiration strikes.
If you appreciate this post and want to catch any future ones coming down the pipe, please consider subscribing to my newsletter!
