ISC (RotM) Artificial intelligence is about to transform society in the same way electricity did 100 years ago, but researchers are nowhere near producing the sort of self-aware sociopathic systems beloved of sci-fi writers.
At least that’s what Andrew Ng, Silicon Valley-based chief scientist at Chinese Web giant Baidu, when he kicked off the International Supercomputing Conference, by sketching the progress of neural networks, or deep learning platforms over the last decade.
Ng said that in 2007, researchers were working on the CPU level, and were making networks with one million connections. As technology has progressed through the use of GPUs, and onto the cloud, and into the realms of HPC technology, networks were being constructed with 100 million connections.
At the same time, he said, researchers were able to use much larger data sets. Whereas academic research projects on speech recognition had worked with data sets of 2000 hours of speech, Baidu’s own speech recognition project was using 40,000 hours, he said, resulting in something close to a game-changing 99 per cent accuracy.
One of the biggest brakes right now was the ability to hard scale the systems, he said, with performance typically plateauing as systems hit 16 cores. However, he outlined a series of techniques Baidu had applied which had pushed their systems up to 128 GPUs. These included slashing the amount of time spent loading up parameters – from 80 per cent to 1.5 per cent of totally processor time – dramatically cutting time spent loading data, and squashing other latencies.
Other issues were less easily tackled, said Ng, including the fact that most training of AI systems is done using labelled or tagged data – so machines aren’t exactly learning all on their own-some yet.
Nevertheless, AI was set to transform society, though this being an HPC conference, Ng said the AI community could really do with collaboration with the broader community, as it sought improvements to help with the incredibly computer intensive elements of the technology.
Ultimately in the AI world, he said it was data that would allow companies to win out – those that had access to mountains of the stuff would be able to train their systems better and faster. “Data is the defensible barrier to entry.”
He said that with with the onset of electricity many companies had vps of electricity who introduced individual motors to electrify elements of the production process. In the long run though, the real winners were companies that redesigned whole business model around electricity
Rise of the Machines.... can we diarise?
However, it’s all very well talking algorithms, scaling and the defensibility of data. What about the defensibility of humans? As soon as the floor was thrown open to questions, people wanted to know – when will the machines become self-aware and turn on us?
Asked how far we were from constructing something on the level of Clarke’s morally conflicted know-it-all auto-pilot cum assassin, Hal Ng said:
“I wish I knew. I do not see a clear path to getting there. We need a breakthrough in algorithms. If we make a lucky breakthrough, in the next decade. If we're not lucky, it could be the 500 years from now.”
As for the likelihood of AI turning on us, Ng was relaxed saying that “worrying about AI turning around and killing us all is like worrying about overpopulation on the planet Mars.” We have to get there first.
There were indeed ethical issues, he continued, such as the possibility of militarization of the technology. He questioned why researchers would want to work on ever smarter cruise missiles, for example.
At the same time, he said, AI would make for a better society by freeing us from repetitive tasks, as it found its way into areas like healthcare, manufacturing, autonomous vehicles, and of course advertising and marketing.
But this raised what is, in Ng’s view, a far more pressing problem in the shape of “labour displacement”.
“If as technologists we put people out of a job, we have a responsibility.” He added that he, along with much of Silicon Valley supports the concept of a guaranteed basic income.
Of course, in a society where an elite caste of computer engineers tend the machines that do all the work, while the rest of us are paid to sit around all day dreaming up new coffee shop concepts, many may end up wishing the machines do go on the rampage, if only to relieve the boredom. ®