Designing single purpose machines is relatively easier than generalized AI.
IBM’s Watson machine, in its Jeopardy playing mode, was never designed to be a “fully rounded individual”.
Because Watson was designed for a particular goal, it is easy (in retrospect) for people to accept that it would excel at the ‘1-dimensional task’ that it was built for. After the initial euphoria of success, the audience readily placed it on the same spectrum of clever/smart machines on which computer chess Grandmasters had already been assigned a spot.
Humans are faced with a very nebulous task ahead when building a generalized AI. In contrast, the task of building bigger, more powerful or more efficient machines sounds like something that could be characterised as a ‘1-dimensional task’. Given the available precedents, it’s relatively easy to accept that creating a machine that could do this specific task better than the best available human is a practical goal. After all, humans already have to use computer assistance to build microprocessors.
Simple proposal : Create a Watson that is good at designing better Watsons.
When does the Singularity Arrive?
Humans may not be able to design AIs - but the singularity doesn’t have to wait until we can.
We can accelerate pace of development by building a better developer first.
The singularity may be closer than it first appears : It doesn’t happen when we can surpass an average human’s generalized intelligence. The singularity occurs when there’s a machine-designing Watson that can design its own successor (and financiers are willing to invest in its designs).
blog comments powered by Disqus