A Letter in Scarlet*

Open the pod bay doors, HAL…

This is one those 1960s events you may actually remember: In Stanley Kubrick’s science fiction classic, “2001: A Space Odyssey,” a mainframe computer decides its space ship’s mission to Jupiter is too critical to remain in the hands of the ship’s human crew. The renegade computer, named HAL, takes things into its own hands, quietly squelching off life support for the sleeping humans. It’s a machine with a mind of its own. The remaining human (played by Keir Dullea) seizes control of the spaceship, violently, pulling out HAL’s computer memory banks.

 

Is there something familiar here? Picture this: You’re driving your car toward a busy intersection and instinctively take your foot off the gas pedal. Your electronically-controlled car, another machine with a mind of its own, decides it would rather not slow down. In fact, it decides to speed up…

 

A guest editorial in this morning’s (Feb 28) Los Angeles Times by muckraker Ralph Nader says that claims about “unintended acceleration” have been mounting since the 2002 the introduction of the Electronic Throttle-Control System (ETCS) on certain Toyota and Lexus models. He blames the problem on a lax and underfunded National Highway Traffic Safety Administration (NHTSA). “NHTSA's motor vehicle safety budget is a mere $140 million,” Nader says. “Taxpayers will pay more than four times as much — about $675 million — to guard the U.S. Embassy in Baghdad.” The NHTSA opened several investigations, Nader says, but closed them without action.

 

In interests of full disclosure, I’ve owned two Toyotas and I believe these cars — even the recall models — are safe. By the same token, I believe air travel is safe. It’s as statistically close to 100% as any piece of machinery can get. But that doesn’t mean a plane doesn’t fall out of the sky from time-to-time, for seemingly inexplicable reasons. Every time that happens, the airlines devote a ton resources to finding out how this occurred, and how it could be prevented from reoccurring in the future. It seems that Toyota is guilty of not acknowledging the acceleration problem sooner and is being pillared in public.

 

As an electronics industry supporter and enthusiast, I’d really hate to think that Toyota’s problems are electronic. It would be a black mark for all of us. David Strom had quoted me in the “New York Times” as saying the modern automobile has more computing power than the average small office. But that may not be a good thing as evidence is starting suggest that Toyota’s sticky gas pedal is not a metal shim problem or a floor mat problem. It looks like a runaway microcontroller — and it will not easily go away.

 

 

Proselytizing automotive electronics

 

Toyota’s Electronic Throttle-Control System adds motorized assist to the mechanical gas pedal assembly. A microcontroller (undoubtedly a pulse-width modulator) sends out complementary pulses to a pair of power transistors. The power pulses drive a multi-phase electric motor, what Toyota calls a “rotary solenoid.” Acceleration would be increased by stepping up the frequency and/or duty cycle of the pulses. If this is the cause of unintended acceleration, we’d want to know what is pumping up the pulse rate. A number of engineers have suggested this could be a consequence of electro-magnetic interference (EMI), which could artificially ramp up the microprocessor clock speed, or encourage the solenoid to register more pulses than were actually sent to it. But I have a hard time believing the ETCS wasn’t carefully engineered to minimize this kind of interference.

 

Audi had a problem of unintended acceleration with its Audi 5000 models back in the 1980s, which many believe was a sensor interface problem. But Audi refused to acknowledge it may be a problem with their machinery, citing “driver error” instead. The bad press, including a sensationalist “60 Minutes” story on CBS, nearly bankrupted the German car maker, and certainly kept them out of the American market for several decades.

 

I’d like to think the automotive industry has come a long way in its acceptance of semiconductor electronics. I remember the charismatic Alex Lidow describe the electronic motor assist system his company was developing to create more efficient autos — before an accounting scandal at International Rectifier forced him to “fall on his sword.” I remember Infineon engineers, after they took over the Sears Point Raceway, taking about the issue of fail-safe software. “You can’t have an ‘hour glass’ suddenly appear on your screen in the middle of a high-speed turn,” they told me. Like the electronic voting machines introduced and then pulled by Diebold, there are inevitably certain embedded processes that should be looked at again and again and again.

 

Infineon spent a lot money and energy, mid-decade, promoting its leadership position in automotive electronics. Gartner’s 2008 market share compilation lists Infineon as the world’s second largest supplier of automotive semiconductors, behind Freescale. STMicrolectronics was the third largest supplier in 2008; Renesas and NEC were fourth and fifth, respectively. I have no idea which (if any) of these manufacturers supports Toyota, who, in fact, has its own IC manufacturing. (I believe this is devoted to power transistors, and not microcontrollers.)

 

The automotive industry has actually embraced the use of semiconductors in a big way. Even in 2009, where a lack of credit and consumer confidence causes auto sales to implode, the automotive sector consumed some $15 billion in semiconductors  (down from the roughly $20 billion consumed in 2008). Despite bewildering price decreases for all types of semiconductor devices, the semiconductor content of all the typical “light vehicle” has remained steady — about $500 — during the past 10 years. The largest content growth in recent years has come from console electronics —GPS-based navigation aids, and infotainment electronics — but future growth will come from safety systems, such as automotive radar and collision avoidance systems. Semiconductor consumption by the automotive industry will return to 2008 levels in 2011 and reach $22 billion by 2014.

My relationship with GM

 

I squirmed at the thought of the Federal government bailing out General Motors, though injecting stimulus money struck me as analogous to the process of inoculating a cat — you have put the needle somewhere. I similarly squirmed when George Bush senior, who claimed Republican “lassie faire” economics wouldn’t distinguish between potato chips and computers chips, nonetheless accompanied American auto industry execs on a crusade to Japan.

 

I have not been fan of General Motors, although their Buick model appears very popular in China. GM’s finance arm (full disclosure once again) GMAC, holds the mortgage on my house in New Jersey. (Shortly after I refinanced, a GMAC telemarketer called me to say, since my credit was so good, they were prepared to offer me a GMAC-endorsed Master Charge card. Build enough points on the card, the marketer told me, and I would qualify for a 5% discount on a General Motors car…) Ask me sometime where I am with the Chevy Volt.

 

I should give credit where credit is due: GM has been pretty aggressive in deploying passenger sensing systems which modulate the force with which airbags are deployed, in an effort to protect young children. It’s taken the company awhile to recognize the virtue of electronic systems in cars. Early in my career as an electronics trade journalist, I visited engineers at Delco Electronics in Kokomo Indiana. Delco (now Delphi Systems), then under the General Motors umbrella, was developing their own console radios, engine controls — and they fabricated many of their own parts (including power transistors) in-house. They showed me some of the advanced systems they were working on, and I frankly was impressed with their brilliance. I asked about the apparent time lag between the design of these systems and their actual incorporation in cars. Why wasn’t GM more aggressive about incorporating the technology their own subsidiary was developing? I wondered.

 

 

My hosts were very humble and accommodating: “We’re just the tail on a Big Dog,” they told me. “The Dog understands internal combustion engines and gear trains — but not much else.”

I’m wondering now, with Toyota having displaced GM as the world’s largest car maker, whether it hasn’t turned into what those engineers would have called ‘The Big Dog.’ q

What goes around, comes around…

Let this be the final word (please!) on “unintended acceleration:”

 

I had written last month that Toyota’s acceleration problems may be symptomatic of a runaway microcontroller… I had written earlier about electronically-controlled exercise equipment, also big consumers of microcontrollers. This morning, when I showed up that gym, a good number of the Star Trac running machines were down. A sign by the injured machines read: “The treadmill speedup is due to a software problem. We apologize for the

inconvenience.”q

If computers could talk

I’ve been with this for quite some time (most of my career, actually): Our notions of what computers can do for us (or TO us) are implicit in the jokes we tell about them. Try this one, circa late 50s, early 60s:

 

Scientists assemble the most powerful computer known to man, ostensibly with the idea of getting it to help them resolve cosmic questions about the origins of the universe. They utilize what we might call “massively parallel” resources, linking together processing resources from around the globe. Its front panel lights blink with growing intelligence as each processing node is added. Finally, the machinery appears fully conscious, towering over the white coated scientists who gave it life. The scientists stare at each other, reaching a silent consensus as to whether to pose the ultimate question. “Computer,” one of them asks, “Is there a God?”

Despite the front panel’s blinking lights, the computer appears dark and sinister. “Now there is,” the machine replies.

 

Computers, suggest the academic books of the time, were something to be feared. There would be dire consequences if you let these beasts into your life, the books insisted. Social philosopher Jacques Ellul, for instance, in “The Technological Society,” argued that technology (translated from the French as “an obsession with ‘technique’”) would eventually overrun religion, a sense of higher purpose, art, literature and all that humans regard as important. Mathematician Norbert Weiner, who coined the term “Cybernetics,” was among the first to envision the possibilities of extending human potential with massively networked computing machinery, but the fictional “cyborgs” that seemed to flow from his vision — half man, half machine — were decidedly evil and destructive..

Consider the same computer joke, 30 years later, this one told by Stanford professor Edward Feigenbaum who implored American computer scientists to emulate Japanese efforts to get computing machinery to think and act like human beings.

 

Scientists are once again assembling a global computer network, with the idea of constructing an all-seeing/all-knowing machine... Only… this construction is sponsored by the U.S. military, ostensibly the only organization in the world with the economic resources to fund the construction of the world’s largest and most intelligent super computer. The generals and Joint Chiefs of Staff gather around as the machinery comes to life, front panel lights blinking. The generals stare at each other, coming to a silent consensus as to what sort of question to ask. Finally, one of the generals steps forward. “Computer,” he asks, “what do you see?”

The computer’s lights blink as it surveys its most distance nodes.

 

Solemnly, it answers: “Missiles are coming.”

 

The generals stare at each other in shock and disbelief. "Computer,” one asks, “Are they coming over land? Or over sea?”

 

The computer’s lights blink, before it answers, “Yes.”

 

The generals’ eyes widen, as they begin to sense their multi-billion dollar project may have generated some extreme silliness. “Yes?!” the top general replies, his voice rising in anger, “Yes WHAT?!”

 

The lights blink once more, as if the machine were considering its best response. Finally, they align in a row as if the machine were snapping to attention. It responds with confidence and authority: “Yes SIR!”

 

 

Computer Science 101

 

With the increasing reliance on unseen computing resources accessed almost seamlessly on the web (conversely, with the increasing suspicion that it’s a runaway microcontroller that has caused Toyota’s acceleration problems), it may be interesting to review what these machines actually do. The four-phase computer clock, whose tempo goes “Instruction Fetch, Instruction Execute,” a couple of billion times a second, controls a complex network of transistor switches. Picture a vast railroad switching yard, but with microscopic switches. Intel’s latest processor, formally introduced last week, is a six-core processor with roughly 1.2 billion of these switches. All these switches do is enable pathways for the flow of “ones-and-zeros” in and out of the machine. That’s ALL they do!

 

Sure:  There are complications, to be sure. The machine has to keep track of addresses in memory; those addresses tell it where to find instructions and data. The “Instruction Fetch/Instruction Execute” then becomes a process of generating an address in memory, opening the switches to that memory location, letting the instruction — itself a pattern of “ones-and-zeros” — flow into the processor, and latching it place. In the next cycle, you’re generating even more addresses: finding the data on which the instruction will be executed, letting the data flow into the machine, generating an address to which to send the results and opening the switches to that location.

 

As described, there’s nothing even vaguely human in this process. It’s just a matter of moving binary numbers around — albeit, at super-human speeds. If there is any seeming “intelligence” in the process, it comes from the programmers and their tools. Yes: there are software compilers and high-level language programming tools that will translate statements like “add the contents of register 5 to the contents of register 7” into the machine language — the sequence of “ones-and-zeros” — that controls the operation of the switchyard. But the sequence of events — what happens when you ask Google to search for “Intel’s x86” embodiments, or when your girlfriend’s picture pops up on your cell phone screen when she calls you, or when the outer wheels on your Audi spin faster than the inner wheels as you go through a high speed turn —has to be thought out by humans in elaborate detail. If a computer steers a course to Jupiter, someone has painstakingly taught it to do that. And if a computer refuses to obey your instructions, someone somewhere has taught it to do that too.

 

What makes this so interesting and useful is that these software decision sequences are cumulative. Google’s indexing, for example, is repeated over and over for damn-near every search term you can think of. Behind every search there are literally many thousands of Intel processors and tons of memory. The typical Google server — each card — is built around two Intel quad-core processors (8 CPUs), a half-terabyte of memory, and lots of high-speed I/O. The city-sized data centers being constructed by Google (like the one in the Columbia River basin) use thousands of these cards. And pulling up previously indexed search terms in a matter of seconds may call a thousand or more for each search.

 

Some have likened the new “cloud computing” model, to the mainframe-connected-to-dumb-terminals model, that characterized computing before the spread of PCs. There is certainly a lot of similarities. The difference with cloud computing is that the mainframe is no longer a single entity connected to dumb terminals on a network. Rather, it is a vast and redundant network of servers, scattered in space, and connecting users wirelessly with cell phones and mini-notebook computers. The vision is to increase the mobility and portability of the notebook by reducing its functionality. You can, for example, reduce space and power consumption of your laptop by foregoing a disk drive — letting your applications and essential data reside on a server somewhere. Thus, software applications become a “service” you subscribe to, rather than a piece of software that you buy and make resident on your machine. And Apple’s construction of their own city-sized data centers represents a transformation of their business from “hardware” to “service.” Despite Microsoft’s loss of user data, despite the creation of ultra-portable computers (as a reaction to the limited functionality of mini-notebooks), Gartner believes the long term model for cloud computing is viable.

 

 

See the world for me, Richard

 

This is far from the last word on anthropomorphic computers. In the heartbreaking “Galatea 2.2,” Richard Powers describes a novelist narrator who, on a bet, sets out to teach a super computer to perform literary criticism — or at least sound like it knows what it is talking about. As part of the lessons, he reads the computer novels and plays and examples of literary criticism. High-brow stuff. This is a Pygmalion drama: the novelist becomes fascinated with the remarkable associations the machinery seems to be making, and, to the extent that the computer speaks in a female voice, he falls in love with her. The novelist has his own psychological torments to deal with, and these surface as the computer insists he stop reading her literature, and give her instead a city newspaper — the kind with daily doses of ugly politics, war and murder. Read the response and ask yourself what are the chances of a computer actually saying this? Powers, who admits he’s only scratched the surface of artificial intelligence, has nonetheless made a believable and poignant character out of this machine.

In the meantime, we can add some new wrinkles to the forty-year-old joke about all-seeing/all-knowing computers:

 

Scientists assemble the most powerful computer known to man, ostensibly with the idea of getting it to help them resolve cosmic questions about the origins of the universe. They utilize what we might call “massively parallel” resources, linking together processing resources from around the globe. Its front panel lights blink with growing intelligence as each processing node is added. Finally, the machinery appears fully conscious, towering over the white coated scientists who gave it life. The scientists stare at each other, reaching a silent consensus as to whether to pose the ultimate question. “Computer,” one of them asks, “Is there a God?”

 

The front panel’s blinking lights appear happy and triumphant. “Now there is,” the machine replies. “And its name is Google.” q