Neural networks in AI and Neuroscience: Where are the overlaps, and what can we learn?
Emma Bannister
? Microsoft MVP | AI & Microsoft 365 Copilot expert driving innovation ? Delivering powerful, painless presentations designed by Asia's creative experts ? Helping businesses streamline growth & boost productivity with AI
When reading about artificial intelligence, you may encounter the term 'neural networks'.
This article explores 'neural networks' in our brains and AI systems and how this overlapping concept teaches us where AI development is headed.
In the brain, 'neural networks' are simply neurons connected with synapses, which facilitate the flow of electrical signals and message transmission across the brain.
Unlike AI networks trained on datasets, our brains are not pre-programmed: Our minds learn by strengthening the connections between neurons that fire together frequently.
Some AI systems are based on these organic human systems, with their own 'neural networks' that mimic the structure and behaviour of our brains. They have artificial neurons that form new connections with each other when exposed to new information, strengthening AI’s pattern recognition and memory.
AI 'neural networks' are used for image recognition, speech recognition, and machine translation, among other things.?
What do these neural networks in AI and our brains have in common?
All 'neural networks' can learn, adapt, and remember: AI can strengthen existing connections and form new ones based on the data it is trained on, and human brains can form new memories and process new information.
Both AI and the brain can also identify patterns in data. However, since information in the brain travels through interconnected pathways that are more complex than the often linear information flows in AI models, humans can reach more sophisticated and considered conclusions that don’t over-rely on existing patterns.
A key downfall of AI systems is an inability to think critically beyond pattern recognition.
Differences between AI and human 'neural networks'.
In a January 2024 TIME article by Andy Clark, he describes a key difference between AI and human cognition: humans take in new information in the wider context of sensory inputs from the world around them. This enables us to “select actions that help us survive and thrive in our worlds.” In contrast, he writes, “AIs have no practical abilities to intervene… so no way to test, evaluate, and improve their own world-model.”
An article in 'Nature by Matthew Hutson' asserts that AI's heavy reliance on pattern recognition through machine learning is precisely what makes it 'inscrutable to humans': it's hard for us to see what artificial neural connections are being strengthened or created.
This lack of accountability and transparency in AI’s processes has led to the rise in Explainable AI (XAI), a field of study which attempts to 'reverse engineer AI systems” to build a “decision tree that approximates an AI’s behaviour', according to Hutson. While these attempts to explain the outputs of AI are currently imperfect at best, they may help inform what guardrails need to be put in place to make future AI developments safer and more accountable.
How advances in AI and neuroscience can be mutually beneficial
Huston’s article also quotes Martin Wattenberg, a computer scientist at Harvard University in Cambridge, Massachusetts, who says that 'understanding the behaviour of LLMs could even help us to grasp what goes on inside our own heads', demonstrating the link between how we understand the workings of our minds and those of AI systems.
It’s also possible that 'neural networks' are similar to our brains. With technology tracking how humans communicate, encode and interpret information, we can hopefully move towards more efficacious AI prompting and maximise the utility of AI systems.
领英推荐
The realm of brain-computer interface (BCI)?tree
Technology is advancing at an unprecedented pace, and with it, the scope of animal research is expanding into exciting new territories. Recent studies have showcased groundbreaking developments in the implantation of AI-driven BCI devices in animals, paving the way for revolutionary applications that could transform the treatment of neurological disorders in humans.
One of the most notable advancements comes from a study that successfully implanted a deep-brain computer chip into a living animal without causing significant neurological effects. This study represents a monumental step in neurosurgery, addressing the challenge of accessing the brain's subcortical regions safely. These implications are vast, potentially leading to new treatments for neurological conditions.
Using AI in BCI implants for animals is a scientific and compassionate effort. It aims to improve human health by exploring the brain's capabilities.
Animal research contributes significantly to the development of AI BCI implants, blurring the lines between biology and technology.
The journey of BCI implants from animal to human applications involves ethical and technical challenges.
Despite this, progress offers hope to many awaiting medical breakthroughs. As we delve into the brain's complexities and leverage AI, we move closer to overcoming neurological limitations.
Platforms like PubMed and IEEE Xplore are ideal for peer-reviewed papers, while ScienceDaily and New Scientist provide news article summaries. Evaluate sources for credibility and check for the latest information.
Advancements in BCIs, notably the Wyss Center's ABILITY BCI, show promise in animal models. This fully implantable wireless device aims to enhance communication and independence for individuals with severe paralysis.
*All featured images are AI creations I created using Leonardo.
For more information,
I hope you found this article informative. My goal was to explore the similarities and differences between neural networks in the brain and AI systems, emphasising how these insights can enhance future developments in AI and neuroscience research.
Be sure to connect with me and follow along for more AI and productivity insights: Linkedin.com/in/emmabpresents or email [email protected]
--
5 个月Useful tips