Over 20 years ago i sat in a lecture hall while a professor talked excitedly about artificial neural networks ANN and their potential to transform computing as we knew it. If we think of a neuron as a basic processing unit then creating an artificial network of them would be akin to multiplying processing power by several factors and replacing the traditional view of each computer having one central processing unit or CPU.
I found the concept of modeling the neurons of the human brain compelling but frustratingly inaccessible. For one thing how exactly do we translate the firing of neurons, a complex electrom chemical event that is not yet fully understood and apply it to computing.This is before we even get into neural communication the coordinated activity of neural networks and the even more complex questions surrounding consciousness.
How exactly do we get computers to think in the same way that we do. Back then,i had a laptop that was an inch thick and capable of a relatively limited number of tasks. It seemed almost unfathomable to think that these ideas could be put into practice alongside the current incarnation of silicon based chips.It would take 10 years before a research paper from Google changed everything. In 2017 a team of researchers at Google published a paper with an unassuming title.
Few could have predicted that this work would mark the beginning of a new epoch in artificial intelligence. The paper introduced the transformer architecture a design that allowed machines to learn patterns in language with unprecedented efficiency and scale. Within a few years this idea would evolve into large language models LLM the type of which was popularised by OpenAI ChatGPT. It was the foundation of systems capable of reasoning translating coding and conversing with nearhuman fluency. But this was not the first step.
A year earlier Google Exploring the Limits of Language modeling had shown that scaling artificial neural networks in data parameters and computation yielded a predictable steady rise in performance. Together these two insights scale and architecture set the stage for the era of generative AI. Today these models underpin nearly every frontier of AI research. Yet their emergence has also brought about a deeper question could systems like these or others inspired by the human brain lead us to artificial general intelligence AGI machines that learn and reason across domains as flexibly as we do.
There are now two distinct paths of research for AGI.The first are LLM trained on huge amounts of text through self supervised learning they display strikingly broad competence in several key areas reasoning, coding translation and creative writing. The suggestion here and a giant leap from lecture hall discussions about ANN is that generality can emerge from scale and architecture.
Yet the intelligence of LLMs remains disembodied they lack grounding in the physical world persistent memory and self directed goals. And this is one of the central philosophical arguments that hampers the legitimacy of this path for AGI. Our ability to learn is arguably grounded in experience our ability to perceive the world we live in and actively learn from it. If AGI ever arises from this lineage it may not be through language models alone but through systems that combine their linguistic fluency with perception, embodiment and continual learning.
If LLM represent the abstraction of intelligence whole brain emulation WBE is its reconstruction. The concept articulated most clearly by Anders Sandberg and Nick Bostrom in their 2008 paper Whole Brain Emulation A roadmap envisions creating a one to one computational model of the human brain. The paper describes WBE as the logical endpoint of computational neuroscience attempts to accurately model neurons and brain systems. The process, in theory, would involve three stages: scanning the brain at nanometre resolution converting its structure into a neural simulation and running that model on a powerful computer.
If successful the result would not merely be artificial intelligence it would be a continuation of a person perhaps with all their memories preferences and identity intact. WBE in this sense aims not to imitate the brain but to instantiate it.Running from 2013 to 2023 a large scale European research initiative called the Human Brain Project HBP aimed to further our understanding of the brain through computational neuroscience. AI was not part of the initial proposal for the project but early successes with neural net deep learning inarguably contributed to its inclusion.
Source link
