Are we on the threshold of Asimov’s robot dreams? We can’t make humanoids useful enough. We can’t make them cheap enough. Why not?

(1) we can only make the things we call robots very useful in a narrow range of applications (e.g., large-scale manufacturing operations)

(2) most kinds of these very useful robotic devices are very expensive (many tens of thousands of dollars each for industrial robots) and are mostly becoming even more expensive to make

(3) we can make them quite cheap when they’re more ‘entertaining’ than useful (e.g., toys) and these are getting cheaper

(4) we can’t make them very useful in a much wider range of applications because we don’t yet know how to make humanoid interfaces which are adequate for safe, reliable and productive interaction with humans in a wide range of contexts

We can’t make very useful humanoids cheap because:

How many Roombas does it take to change a lightbulb?

The Roomba is small, widely available, useful and much cheaper than an industrial robot.

So, does the success of the Roomba, as it autonomously navigates floors in order to vacuum them, imply that the problems associated with making robots have already been solved?

Well, some of the humanoid characteristics it lacks, such as the need for articulated limbs and ‘above floor level profile’, as well as its narrow application, put it squarely in the category of non-humanoid robot.

(a) they tend to need to be big (often larger than most people) for most of the applications where they are currently most useful (big devices tend to have much more expensive components)

(b) they need very high precision construction in order to be durable enough (high precision introduces negligible cost for small equipment like phones, but high precision is very expensive for big equipment like industrial robots)

(c) they tend to be highly specialised (mass production ‘economy of scale’ is minimised)

To make a machine which combines the agility, stamina and dexterity of most humans, we need further advances in mechanical engineering, although we are getting closer every day, and can regularly demonstrate major advances in each area, occasionally exceeding human capability. Making these new physical capabilities inexpensive is yet another challenge.

Similarly, for a robot servant as ‘mentally’ responsive and resourceful as a human, we need humanoid interfaces way beyond those we have right now. We don’t really know exactly what to do to get to what we might need. In fact, there’s no real consensus on what we might need. As with the mechanical side, we’re making very significant progress, but we have even less of an idea of what this progress might mean in terms of ‘reaching a tipping point’.

The humanoid innovator’s dilemma is this: what compromises and  interim solutions will represent opportunities for a much wider proliferation of useful humanoids?

Some of those developments may not be technological, they may be legal. Here’s an excellent panel discussion at Stanford Law School which tackles these issues.