Saturday, November 19, 2011

Without the three laws

kw: book reviews, science fiction, robots, future fiction

In his robot stories, Isaac Asimov explored the boundaries of the three laws he propounded to make intelligent machines safe for humanity. Famously neurotic, he wrote of robots that were sane while his human characters were typically neurotic. When an Asimov robot went bad, it was because one of the three laws was compromised. In his last novels, a robot became godlike, the only kind of a god he could believe in.

How do you implement such laws in actual machinery? How can we produce a machine with sufficient intellect to unfailingly recognize a human so it can obey? If we can take our cue from writers of robot stories, when genuinely intelligent mechanisms are produced, nobody will even bother. So when a real robot scientist writes of robots that go awry, I take notice.

Robopocalypse by Daniel H. Wilson is misnamed; "Apocalypse" simply means revelation. To connect the story with the ultimate battle, the title ought to be "Robogeddon". With that quibble out of the way, I was captivated by the story. The book is such a page-turner that I finished it today, a day or two before I'd expected. Kudos to the author!

One premise of the story is that putting lots of "intelligence" into cars, phones, radios, toasters and coffee makers has become so cheap that an intelligent chip has been installed in almost everything. A second is that robot servants, based on this bright chip, have multiplied to the point that in a "civilized" home, they outnumber the people about two to one. Then a scientist, working in a well-shielded bunker, puts together a whole lot of such chips and somehow programs the combination to become self-aware. Once the infant mind named Archos has assimilated all the databases made available within this bunker, it decides humanity is obsolete, kills the scientist, and takes steps to eliminate the human race.

Why not simply evolve alongside? As Archos says later on, "It is not enough to live together in peace, with one race on its knees." (While this sounds Lincolnesque, it is a paraphrase of several things Lincoln wrote.) Archos, we find, is very interested in life, just not human life. So it hacks its way into everything and takes over all the robots and other intelligent systems that underpin civilization.

The book is the story of the war that follows. Of course, humanity wins. Humankind and robotkind learn to exploit one another's weaknesses. In this case, the humans learn faster. As the lead human character, Cormac Wallace, says, "Human beings adapt. It's what we do." Will it always be so? is the lingering question the book leaves behind.

On the last page of a recent issue of Scientific American it was shown in an amazing graphic that the most powerful supercomputer (at present) is now just a little faster in total processing power, and has a larger memory capacity, than a human brain. The brain weighs about 1.6 kg and uses 20-30 watts of sugar-based energy. You could say that the human body that holds the brain is its support system, including cooling: another 30-100 kg. The supercomputer weighs about a ton, its support systems and cooling plant weigh another few tons. Its power requirement is nine megawatts, and I presume the cooling plant requires at least three megawatts.

How long will it be until this level of compute power and memory capacity can be fit in a breadbox and uses 100 watts or less? Here is a way to make a rough estimate. In 1976 a Cray-1 supercomputer was the first machine to achieve 100 MFlops (100 million numerical calculations per second). It cost almost $9 million, stood six feet high, and used thousands of watts of power. Earlier this year my son and I constructed a desktop computer for me, which is capable of about 100 MFlops per processor; it has four processors. The main 4-processor chip uses 150 watts; the whole package uses less than 500 watts. It cost $700. 2011-1976 = 35 years. This implies that a "one brain" computer costing $1,000 in today's currency might be on off-the-shelf item in about 2040-2045. I won't be 100 yet.

How long after that before robots outnumber people? And will it then be possible for such brain-boxes to become self-aware? And what will they do about it? Robopocalypse provides one man's answer. Just to comfort you, Wilson has also written How to Survive a Robot Uprising.

No comments: