On the Digital Age
The robots are late
Take a look at popular entertainment and it seems the robots are running behind schedule. Stanley Kubrick's 1968 film, 2001: A Space Odyssey, has HAL the computer in fluent conversation at the turn of the century. Blade Runner (1982) chases androids in the year 2019. The Terminator (1984) brings lifelike robots back in time from the year 2029. The timescales become progressively less optimistic – 32 years, 37, 45 – yet reality never matches the promise.
We can't rely on Hollywood for our science predictions, although science fiction writers do better than most. Arthur C Clarke is famous for nailing geosynchronous satellites. Isaac Asimov's 1964 musings on a 2014 World Trade Fair show great prescience on flat screens, microwave ovens, microprocessors, videoconferencing and computerised books and documents. Yet they miss the mark on levitating cars, underwater cities and moon colonies. Our relationship with the moon is a telling one. First man on the moon 1969. Last, 1972. Who would have guessed at that combination?
My own impression, as a technical journalist and software engineer, is that technology is progressing at a much slower rate than we imagine. And of course in a slightly different direction. Hardly anybody foresaw the Internet, though Mark Twain (in 1898!), Jules Verne, Jorge Luis Borges and Paul Otlet made a decent stab at it. And it's robots in particular that I see arriving at the pace of a snail. Speech recognition is not going according to plan, it should have been cracked in the early 2000s, and that's a sign of our failings.
It's the intelligence element of robotics that's lagging behind. Movement, dexterity and sensors aren't in bad shape, but in intelligence terms we're struggling around cockroach level.
Here's a rough list of artificial intelligence levels:
- Small scale dedicated task intelligence (play chess)
- Mimic animals (eg a dog)
- Large scale task intelligence (drive a car or fly a plane fully independently, or process language)
- Mimic humans (the Turning Test)
- Equal animals (thinking not mimicking)
- Equal humans
- Exceed humans (sometimes called the Singularity)
Deep Blue proved itself a chess champion in 1997, so we've passed Stage One. Sony's Aibo dog in 1999 just about qualified for Stage Two. We are now tackling Stage Three. With a few modifications a commercial airliner could make a full flight without a pilot, as long as nothing too unusual happened. Google's driverless car is starting to worry the insurance industry, not because it's fallible but quite the opposite. IBM's Dr Watson won the quiz show Jeopardy in 2011. Clearly it had the knowledge and could make decisions, but it still needed to receive the questions as text not speech. On balance, we still haven't fully cracked Stage Three, large scale tasks, sixteen years after Stage Two, the robot dog.
Other observers are more optimistic. If you believe the experts canvassed by Pew Research, AI is just around the corner and will be a big influence by 2025. The problem is that most of these experts have jobs that rely on the progression of technology. They'll earn less if it goes slowly.
Which brings us on to Ray Kurzweil, director of engineering at Google and author of The Singularity is Near. When asked in 1990 for the year when a machine would beat the world's chess champions he said 1998 – a year late. He currently has a $10,000 bet on a machine passing the Turning Test by 2029, with The Singularity pencilled in for 2045. He's a real engineer who was at the forefront of flatbed scanning, optical character recognition and speech synthesis. Not only does he have a good track record but he now has influence over the massive resources of Google, the company that developed the driverless car and also owns Boston Dynamics and DeepMind – it's a prime mover in AI. There is some slight moral question over whether you can place a technology bet then become the driving force of that technology, but it's nothing more than an amusing aside. Kurzweil is good and his predictions should be taken seriously.
Viewed in a slightly cynical light, you could say that all we really need to pass the Turing Test is good speech recognition. After all, the average human isn't all that clever so anything too bright would probably fail the Turing Test and need to be dumbed down. I'll wager that Kurzeil's Turing Test bet is safe. He has the resources for Google to build a giant computer that does excellent speech recognition by 2025, and mimicking average human conversation will be relatively trivial. Simply collect a million examples of dialogue and regurgitate them.
But after the mimicking comes the thinking, and that's a challenge on an entirely different scale. One of the major problems is that we can't replicate the workings of the human brain because we don't know what they are. Kurzweil's twenty years from Turing to Singularity is a tight call. Noam Chomsky, who may not be a computer engineer but certainly knows a lot about language and thinking, believes that Deep Blue and Dr Watson were nothing more than large data stores combined with good search mechanisms. He doubts whether our current style of computer programming can ever show real intelligence. The Singularity remains a pipe-dream.
Certainly we have a history of over-optimism. Wikipedia runs a page on the Future of Robotics, with timelines from respected government research. Here are some fine examples:
- 2015-2020 - every South Korean and many European households will have a robot, The Ministry of Information and Communication (South Korea), 2007
- 2015 - one third of US fighting strength will be composed of robots, US Department of Defense, 2006
- 2013-2017 — robots that care for the elderly, Japan NISTEP report, 2000
Maybe when thinking robots finally do arrive they can help improve our technology predictions. That wouldn't require much intelligence.