#71: 🤖 Why AI is Getting Stuck
AI was built on cognitive science, now it’s time to incorporate neuroscience
Hi Friends,
Each week I address topics on health, longevity, psychedelics suggested by you.
If you’re not a subscriber, here’s what you missed so far this month:
Want a topic covered?
Drop it in the comments.
🗣️ Looking for the read-aloud version of Nina’s Notes?
💬 In this note:
🤖 Why AI is Getting Stuck
📚 Incognito: the Secret Lives of the Brain
⚡️ An Octopus Traveling Across the Seafloor
🤖 Why AI is Getting Stuck
If you’ve been playing with ChatGPT and other AI tools, you understand how quickly the field is moving.
AI is extremely good at quickly processing pre-existing information to identify patterns and present us with that information in an answer to our question.
One problem is that it is based on past information.
How many times has ChatGPT started a prompt with “I’m a large language model trained on data until 2021…”
Of course, human intelligence is also shaped by our past experiences.
What I find lacking from ChatGPT interactions, which I prefer in human interactions, is the back-and-forth and the generation of new ideas.
The sharing of past experiences between people in a conversation is what keeps conversations interesting.
When having a conversation with ChatGPT4, it rambles off a lot of information, sometimes hallucinating, but it’s not then prompting the user to ideate or strengthen the text it has generated.
At least, not yet.
So what’s missing from the AI and how is it getting stuck?
I recently read Incognito: The Secret Lives of the Brain and author David Eagleman discusses where he thinks AI is getting stuck.
Keep in mind this book was published in 2012, so it’s basically ancient when it comes to AI.
Nonetheless, Eagleman discusses several neuroscience concepts that have not yet been fully integrated into AI and a potential for AI advancement if leveraged.
Current AI algorithms are advanced neural networks.
The idea of an artificial neural network is to mimic the structure and function of the brain’s neurons and neural network.
By mimicking the structure and function of the brain’s neurons and synapses, these neural networks can facilitate complex pattern recognition and make very fast decisions.
By adjusting the nodes in a neural network, the network can perform a wide range of tasks, from image recognition to speech recognition and predictive analytics.
There is a limitation on the nodes in a neural network and that’s where AI starts to get stuck in trying to mimic the brain.
Our brain is much more complex, there are many mechanisms in our brain still being discovered in neuroscience that are not yet used in AI.
The more we learn about our own brains and how the brain organizes and processes information, the more sophisticated and efficient AI systems we can build.
So what principles of neuroscience have yet to be leveraged in AI?
Neuroplasticity
In neuroscience, neuroplasticity is the brain’s ability to change and adapt as a result of experience.
Forming new connections in the brain is fundamental to learning and memory.
In AI, algorithms that mimic this adaptability can improve over time, adjusting to new information and environments without being explicitly reprogrammed, thus improving generalization across tasks and requiring less retraining.
If we are able to develop AI that can simulate the dynamic nature of the synaptic connections and plasticity of the brain, it could be able to learn from experiences, adapt to new tasks and improve its performance offering a way to develop flexible and resilient AI systems.
Models like this could further improve areas like autonomous systems where adapting to dynamic environments is essential.
Neuroplasticity not only involves forming new neural connections, but it also means deleting old connections.
We refer to this as “rewiring the brain.”
This type of neuroplasticity allows the brain to adapt to new experiences and recovery from injuries.
If we can enable an AI to have neuroplasticity, it could discard irrelevant information, and optimize its own architecture for new types of problems, which could lead to better energy efficiency and processing speed.
Learning
At the moment, when you want to solve a problem using a neural network, the neural network first must be trained on huge amounts of information, and then a model can use brute force to iterate through all the possible solutions.
The model may appear to learn, but it had to iterate through all possible solutions, requiring a lot of time and energy to reach the right solution.
The human brain is much more energy efficient.
The human brain doesn’t have to perform all of those iterations to find the best solution everytime we solve a problem.
In his book, Incognito, David Eagleman introduces the “team of rivals framework” where he suggests that there are multiple overlapping ways to solve a problem. Our brain chooses the best solution through the competition of the different ways to solve a problem.
Eagleman features a quote by Michele de Montaigne, in reference to the team of rivals framework:
“there is as much difference between us and ourselves as there is between us and others”
This is so true. We grow, change our opinions, and learn through experiences.
For example, if someone gives us a piece of cake, we have an internal conflict as our brain has evolved to crave sugar, but another part of our brain wants us to skip the cake and eat healthy. The brain then makes a decision based on the conflict of those two inner dialogues.
Leveraging that internal rivalry in ourselves can help us to problem solve by competitive elimination.
Eagleman suggests that if AI could leverage that internal conflict, and make decisions through a democratic architecture, it would unlock the next age of biologically inspired machinery.
Emotion and Motivation
A core element of being human is our capacity for emotion.
Emotions influence attention, memory, decision making and problem solving.
There is nothing more powerful than a human with a strong emotion that’s motivating them to do something.
If we are able to replicate that motivation in an AI, by stimulating emotional states and motivational drivers, AI could perhaps achieve more humanlike interactions, and make decisions that consider emotional contexts.
AI definitely isn’t there yet, but if you tell ChatGPT to “take a deep breath” before giving you an answer, the answer is better than if you didn’t give it that instruction.
What’s Next?
The field of AI is evolving rapidly, with tools like ChatGPT showcasing the remarkable capabilities of these large language models and in processing huge amounts of pre-existing information to answer our questions in our natural language.
However, as we play with the tools and recognize their current limitations, we see that they still fall short of capturing human intelligence.
Through leveraging the principles of neuroscience and embracing the complexities of the human brain’s adaptability, competition, and influence of emotions, we can get closer to developing AI systems that can truly learn and interact.
The next wave of AI advancements will interact in ways not just reflective of human intelligence but also be capable of generating new ideas and fostering genuine, dynamic interactions.
📚 Book of the Week
Incognito: The Secret Lives of the Brain by David Eagleman
★★★★★
A neuroscience book for the non-scientist.
It’s full of interesting ideas, and Eagleman does a great job explaining complicated neuroscience concepts.
Throughout the book, Eagleman touches on using neuroscience as a way to assess and handle criminals appropriately which is not currently done by the U.S. Justice system.
His arguments are compelling, backing them up with explanations of how genetics, hormones or even cancer can influence our behavior. Making us act out of character and completely out of our control.
I listened to it in audiobook form, which I would not recommend as sometimes it’s easy to get distracted and miss key concepts.
I’ll be adding two of his 51 published works to my to-read list:
⚡️ Check This Out
Image Source: Reddit: Cirroteuthid octopus billows like a circus tent
Watch this video of a Cirroteuthidae octopus moving across the ocean floor.
The ocean really is an alien planet.
..or the landscape of my nightmares.
Edited by Wright Time Publishing