... we develop low-cost, fast-charging, quiet and low-polluting portable power sources with enough capacity to power a human-sized android to walk around untethered for a few days.
That's it. That's the major remaining bottleneck for the development of general machine intelligence that can start to compete with humans and other animals. I expect this WILL happen in the next 3 decades.
Once those power sources are developed, we will see a proliferation of robots walking around all over the place. Production and deployment of large numbers of robots that can walk around and interact with the outside world is the *key* remaining step for developing powerful machine intelligence. These are the reasons why:
1) Sensors and Perception
Being free to interact in the external environment will require a much richer array of sensors and actuators on the robot armature. Sensors will measure things like temperature, moisture and pressure in addition to current inputs like position, acceleration, sound and light. Along with the richer sensor networks will come the computational subsystems for processing and integrating them. These subsystems will be the perceptual circuits of the robot mind.
When robots are free to roam around in the real world (not just driving along streets, but almost everywhere that humans and other animals can go), they will acquire capabilities for monitoring damage and preventing harm to themselves. In order for a robot to protect itself and survive outside, it must be able to identify, prioritize, categorize and compare all sorts of unexpected stimuli, threats and opportunities. As with animals, these low-level circuits will be the foundation for emotional behavior, which is a requirement for true intelligence. Until a robot has a well-developed capacity to sense and react to sudden and unexpected stimuli, it feels *nothing*. Current robot 'emotions' are simulations, nothing more.
3) Large Numbers
Once you have perceptual and emotional networks in machines that can move almost everywhere in the outside world, only then do you have the playing field to develop true, general-purpose intelligence. Intelligence that includes the ability to model the outside world, to make predictions and solve unusual and difficult problems. Once you have the playing field, all it takes are large numbers - large-scale and widespread deployment of millions of mobile robots to produce the rapid cycles of technological innovation and evolution, such as we have already seen in many other areas.
Only this time, the end result will be able to compete with human beings and that's *not* good for us, despite what you may hear from techno-optimists promising a future of global human leisure and luxury. Sorry, but it's not going to work out that way. If you need convincing, start considering what happened in the past when superior biological and technological groups encountered and competed with inferior ones for resources and space in the environment.
There's plenty more to this story, including the inherent dynamics of our current economic systems, energy issues and the trajectory of autonomous industrial manufacturing systems, but that will require quite a bit more explanation.