While it does not seem likely at current progress, human-level artificial intelligence (AI) by 2025 is not as far-fetched as some may believe and we are already on that journey.
By Lionel Bisschoff, CEO and founder of Learning Machines
Major enterprises like Google, Apple, Microsoft, Facebook and Amazon are pumping billions into AI research, including neuroscience research.
Good examples of major artificial intelligence labs conducting research include DeepMind, OpenAI, and Google Brain.
Neuroscience is an evolving discipline and AI is an evolving technology. The exploits of Ray Kurzwell can be used to explain this.
Kurzweil is known for stating that exponential progress is often underestimated, especially since it is difficult to recognise exponential progress in the beginning stages. An example he often uses is that of the Human Genome project.
It was a 15-year project, and by year seven, no more than 1% of the human genome had been sequenced. A sure-fire failure? Nope, it was done and dusted just three years later.
What is often overlooked is that 1% progress was gained after enormous investment in understanding and tools and techniques, with these forming the basis for faster future progress.
Brain matter
Might something similar be happening in our understanding of how the brain, especially the neocortex, works? A number of global projects are under way to map and understand specifically the human as well as the general mammalian brain.
There are strong examples that come to mind, including Europe’s Human Brain Project, US Brain Initiative, Japan’s Brain/MINDS, Korea’s Brain Research Initiative etc.
We are gathering more physical data on brain structure and functioning on a daily basis, and building better and better tools to understand and model in ever more detail how the brain works.
Using this expanding set of information on physical structure and workings of the brain, a small research firm Numenta has been making steady progress in formulating a general theory on the workings of the neocortex.
The neocortex is the part of the brain most responsible for endowing humanity with greater smarts than any other earthly creature.
The remarkable thing about the neocortex is that it is remarkably self-similar. It therefore seems likely that there is a single learning architecture and algorithm involved in how the neocortex observes, learns, predicts and takes action through our bodies.
Crack the neocortex and you’ve probably cracked AGI (Artificial General Intelligence), the holy grail of AI.
Numenta’s progress ranges from understanding cortical columns and how their regions interconnect to why we have thousands of synapses per neuron (because they place connected neurons into predictive states based on recent events) to recent progress in understanding how grid cell-like cells in the neocortex enable us to not only navigate, but also build object models in the brain – whether these models are for real-world environments and objects or for abstract thinking.
Deepmind recently published a paper where some of their recurrent and self-reinforcement learning algorithms spontaneously developed grid cell capability and related navigational abilities.
Artificial brains are starting to evolve approaches similar to those evolved by the human and mammalian brain. Interesting …
It seems to me that progress is speeding up.
Might current research be the key to unlocking the secret to how the neocortex works?
Just imagine that we uncover the fundamentals of how the neocortex works in the next seven years. Then we create this algorithm in chips and connect them to terabytes of perfect real time memory, petabytes of perfect storage memory, lighting fast calculation, billions of high definition sensors.
A fascinating and somehow scary though: human level AI by 2025?
Maybe …