Robot Predictions Evolution

Hans Moravec
April 2004

Here's how my anticipations about the time of arrival of human level robotic intelligence evolved:

In the early 1970s, doing simple computer stereoscopic vision, it became rapidly obvious that the computer power in our mainframe PDP-10 was hugely insufficient to do even that basic function in real time, implying that doing the job of the whole nervous system was even further out of reach. Besides enormously more speed, we needed enormously more memory.

This was contrary to the orthodoxy in AI at the time. My advisor John McCarthy wrote in many essays that existing computers were sufficient for human-intelligent AI, but we needed some theoretical breakthroughs (humorously 2 Newtons and 3 Einsteins) to achieve it.

I tried to quantify the power needed, first by estimating the number of switching operations in the brain and comparing it to switching in computer circuits, and got a rough number of about one trillion (10^12) operations per second (ops). At the time the computers available to AI workers had not budged from 1 MIPS (10^6 ops) for 20 years (and wouldn't for another 15 years), so this estimate was seen as horribly pessimistic. When I first wrote it up 1n 1974, I compiled a chart showing the cost of transistors had been dropping 100x every decade ($100 in 1950, $1 in 1960, 1 cent in 1970 ICs, by 1974 you could get a calculator for $5). Intel was now making a computer on a chip, and it might be possible to interconnect a million of them to get a trillion ops. I figured with an Apollo scale effort, that could be accomplished in about ten years, for a few billion dollars. The growing computer would be a platform to evolve the algorithms.

By the end of the 1970s, it was pretty clear there wouldn't be an AI Apollo project, but I thought that if people were willing to put as much effort as was expended in the weapons labs, we could have the requisite power in a supercomputer-class machine in about 20 years. That's the glimpse you got in that TV show.

By the mid 1980s, even an AI supercomputer seemed unlikely, and I had decided to present the thinking more rigorously and more dispassionately in a book. I redid the power calculations more carefully, starting with the actual known function of the retina compared to roughly equivalent computer vision steps. That indicated that it would take a billion (10^9) ops just perform the retina's edge and motion detecting functions. I extrapolated to the whole brain by neuron count and got 10 trillion for the whole brain. Even though our AI and Robotics computers were still stuck at 1 MIPS (but much cheaper - in the 1970s we used million dollar computers, by the mid 1980s equivalent power could be had in workstations for tens of thousands of dollars.), I was able to plot the historical decrease in computing cost to predict that we would have 10 trillion ops in a $10,000 computer by about 2020 or 2030.

Starting about 1990 the computer power available to robotics began to grow above 1 MIPS, and by the middle of the decade it reached 100 MIPS in expensive workstations. Robots began to succeed at things (like driving long distances fast) that had been impossible dreams before. I updated the "Mind Children" calculations for "Robot", taking into account the actual progress. I changed the retina to brain extrapolation from neuron count to volume (brain neurons tend to be much larger than retinal ones), and got that emulating the brain's function would require 100 trillion ops. Fortunately data from the 1990s updating the computer power chart showed that the decline in computing costs had accelerated, and the new goal would be reached by a $1,000 personal computer-class machine by about 2030.

By the early 2000s also there were several supercomputers in existence that could do more than 10 trillion ops, though not available for robotics work. In 2004 VA Tech connected 1100 dual-processor Macintosh G5 machines for the record low cost of about six million dollars, and benchmarked it at over 10 trillion ops. SEEGRID is building visual self-navigating vehicles using onboard computing of a billion ops or so, about the brainpower of a guppy by my numbers.