Turns out, that time may be sooner than previously thought.
London’s Deep Mind recently published a paper in the scientific journal Nature stating that it has built a neural network that has dynamic external memory.
Using a new machine learning method schema called a differentiable neural computer (DNC), Deep Mind’s engineers have created a machine that can take the lessons it’s learned in one problem and apply them to another problem demonstrating the ability to generalize between solutions.
I understand that sounds convoluted, and it is, but think of it this way:
When a traditional neural network scheme is launched to solve a problem, it has to be trained how to behave using data sets. Once trained, the neural network can make predictions to solve similar problems very quickly. The catch is that for each problem, the neural network has to be retrained.
For this reason, a traditional neural network cannot make generalizations about a series of problems; it can’t create a memory from which it can recall a set of predictions or assumptions that would make solving problem easier thanks to previous experience.
According to Deep Mind, DNCs can do just that.
“We hope that DNCs provide both a new tool for computer science and a new metaphor for cognitive science and neuroscience” read Deep Mind’s blog. “[H]ere is a learning machine that, without prior programming, can organize information into connected facts and use those facts to solve problems.”
While the jury is still out on whether DNCs represent a giant leap forward for AI, I think that it’s safe to say that DNCs represents a completely new stage in understanding how to build effective artificial intelligence systems.
When or whether artificial sentience will ever emerge is still up for debate. Some thinkers even believe that creating consciousness from whole cloth is impossible. Maybe that’s the case.
What’s more interesting to me is the snap judgement that people have when thinking about AI. Sure, I know Musk and Hawking considered the issue deeply before coming to the conclusion that the risk is just too dire, but I wonder if there’s a path to benevolent AI that’s been overlooked.
In the end, humanity would be the collective parents of this new sentience, and most offspring don’t emerge ready to unseat their progenitors. What if AI was born with that innate grace?
I think that reality is equally as compelling as one that spells out either a quick and brutal end or a plodding doom driven forward by the hyper competitiveness of a superior intelligence.
Don’t you?
For more AI news, find out what artificial intelligence can do for additive construction.