Serious thinking on AI


Difficult ethical questions have been raised this month after artificial intelligence (AI) was shown to be 91% accurate at guessing whether somebody was gay or straight from a photograph of their face. If a madman with a nuclear weapon is a 20th century apocalypse plot, climate change and AI are the two blockbuster contenders of the 21st. Both play to the tantalisingly ancient theme of humanity’s hubristic desire to be greater, bringing about their own downfall in the process. Barring recent political setbacks, the risks of climate change are no longer controversial. AI, on the other hand, seems to still be.

In Oxford, an ancient city of spires and philosophers, seemingly standing out against technological advance, there is the Future of Humanity Institute. The institute is headed by Nick Bostrom, a philosopher who believes that the advent of artificial intelligence could well bring about the destruction of civilisation. Meanwhile, in the New World, the top ‘schools’ are turning overwhelmingly to the study and development of artificial intelligence. At Stanford University, about 90% of undergraduates now take a computer science course. In Silicon Valley, the religion is one of self-improvement. Optimization (with a ‘z’) is their religion. With a strong culture of obsessing over their own ‘productivity’, it’s little wonder that the promises of AI, the ultimate optimiser, have such a powerful draw on so many brilliant brains on the West Coast.

Back home, Bostrom is a leading source of warnings against AI. His papers have titles like “Existential Risks: Analyzing Human Extinction Scenarios and Related Hazards”. It’s clear that he is a man not shy of staring back at the abyss, though I’m not entirely sure what ‘Related Hazards’ we need to be concerned about post-Human Extinction. He’s said that our work on artificial intelligence is “like children playing with a bomb”.

Bostrom extrapolates from the way that humans have dominated the world to highlight the risks of AI. Our brains have capabilities beyond that of the other animals on this planet, and that advantage alone has been distinctive in making us so overwhelmingly the dominant species, and in a relatively tiny amount of time. The dawn of civilisation could be described as a biological singularity, an intelligence explosion. If an artificial brain could be made more powerful than our own, why should we not see a second ‘intelligence explosion’? This brings up the technological singularity – a paradigm in which humans might come to have as little power over our lives as battery farmed chickens do now over theirs. The difference, I suppose, is that chickens didn’t inadvertently create humanity, so Bostrom sees a chance for us to control our creation – we choose how and when to turn it on.

However, AI need not be inherently malignant in order to destroy us. By illustration, one of Bostrom’s fun apocalypse scenarios (another of his light reads is Global Catastrophic Risks) is that of the end of the world by a runaway artificial intelligence algorithm initially designed to help a paperclip factory manufacture as many paperclips as possible. Within seconds, the machine has quite logically reasoned that this would be best achieved by wiping out the human race in order to make a bit more space for paperclip manufacturing, and cheerfully, obediently embarked on its task.

My engineering degree at Oxford is definitely a bit backward. Most of the course seems not to have changed since Alan Turing cracked the Enigma Machine at Bletchley Park. My decision to focus my final year on AI thus makes me – I would like to think – a dangerous maverick. Probably, the professors discuss me in hushed tones, fear mixed equally with reverence. I’m actually extremely grateful for this environment.

Away from the fervent centre of the Religion of Optimisation, it’s far easier to see the bigger picture, without being blinded by the light of enthusiasm. The Laboratory of Applied Artificial Intelligence which I just became part of sits within the Oxford Robotics Institute. The important nuance in this hierarchy is that at Oxford, artificial intelligence is more a prosaic tool to be applied to an end, than a quasi-religious holy grail in itself. Say, making better driverless cars. It is only in very specific, tailored ways like this that artificial intelligence is, and can be, currently used. This concrete embodiment of AI is called Machine Learning, and is nothing more glamorous than particularly clever statistics, run by relatively powerful computers.

It is these mundane algorithms that optimise all online ads to their audience, determine your sexuality from a photo, get your Über driver to you within 2 minutes, or even replace that Über driver altogether. Long before Bostrom’s artificial superintelligence surpasses the human brain and crushes us like ants, civilisation will be tested by the extreme turbulence in lifestyle and employment that will be brought by this far more mundane embodiment of computer intelligence. Besides a bit of philosophy and working out how to unplug a monster brain, we should be considering a far closer future, in which boring machines that can’t even hold a conversation will have nonetheless put most of us out of work.

more here

thoughts about industry 4.0

As Industry 4.0 technology becomes smarter and more widely available, manufacturers of any size will be able to deploy cost-effective, multipurpose and collaborative machines as standard. This will lead to industrial growth and market competitiveness, with a greater understanding of production processes leading to new high-quality products and digital services.

Exactly what impact a smarter robotic workforce with the potential to operate on its own will have on the manufacturing industry, is still widely disputed. Artificial intelligence as we know it from science fiction is still in its infancy. It could well be the 22nd century before robots really have the potential to make human labour obsolete by developing not just deep learning but true artificial understanding that mimics human thinking.

Ideally, Industry 4.0 will enable human workers to achieve more in their jobs by removing repetitive tasks and giving them better robotic tools. In theory, this would allow us humans to focus more on business development, creativity and science, which it would be much harder for any robot to do. Technology that has made humans redundant in the past has forced us to adapt, generally with more education.

But because Industry 4.0 robots will be able to operate largely on their own, we might see much greater human redundancy from manufacturing jobs without other sectors being able to create enough new work. Then we might see more political moves to protect human labour, such as taxing robots.

Again, in an ideal scenario, humans may be able to focus on doing the things that make us human, perhaps fuelled by a basic income generated from robotic work. Ultimately, it will be up to us to define whether the robotic workforce will work for us, with us, or against us.

uncertainty and the path to prototype warfare…

This article is primarily about war planning.  However the description of uncertainty applies to everything. Well worth a read 

Dealing with uncertainty is not a new problem for military planners, though, and research on military innovation may provide an answer for the way forward. For example, Stephen Rosen’s classic book on military innovation, Winning the Next War, identifies two types of flexibility for dealing with uncertainty. According to Rosen, “Type I” flexibility relies upon developing capabilities, such as the aircraft carrier, that have great utility over time, particularly as such can be modified as mission certainty increases. “Type II” flexibility involves buying information on weapon systems and then deferring large-scale production decisions. This usually involves bringing systems to the prototype stage and permitting military testing in field or fleet exercises. Rosen describes how this strategy was used successfully in the development of guided missile programs. At the end of World War II, it was not clear how to proceed with missile technology, as this was a period of great uncertainty both technically and politically. The Joint Chiefs of Staff adopted a hedging strategy in the late 1940s that focused on investing in the basic research and, as the operational demands of the Korean War increased, the Pentagon was able to quickly shift into full-scale missile production.

Can the Type II model of flexibility be turned into an operational advantage? This is not a new idea, but has yet to be given a fair shake by the U.S. military. In the 1990s, one U.S. Army officer predicted that this approach to capability development would result in an operationally significant concept he termed prototype warfare. In The Principles of War for the Information Age, published in 2000 and prior to the onset of the trends that Hammes discusses, Robert Leonhard argues that to create or to maintain a technical advantage in the information age, successful militaries need to adapt their economies and military doctrine to prototype warfare. He observes that technological change in the industrial age occurred at a moderate tempo and as a result, military doctrine was based on the fundamental premise of mass production. He concludes, “future warfare will feature a constant myriad of technological advances that come at a tempo that disallows mass production.”

AI researcher worries… intelligently… Could Marx still be right?

fascinating issue  raised at the end of the article. Could Marx and Lenin still be right? Could the elites own all the means of production  through AI slavery?

I’m not very concerned about unintended consequences in the types of AI I am developing, using an approach called neuroevolution. I create virtual environments and evolve digital creatures and their brains to solve increasingly complex tasks. The creatures’ performance is evaluated; those that perform the best are selected to reproduce, making the next generation. Over many generations these machine-creatures evolve cognitive abilities.
Right now we are taking baby steps to evolve machines that can do simple navigation tasks, make simple decisions, or remember a couple of bits. But soon we will evolve machines that can execute more complex tasks and have much better general intelligence. Ultimately we hope to create human-level intelligence.
Along the way, we will find and eliminate errors and problems through the process of evolution. With each generation, the machines get better at handling the errors that occurred in previous generations. That increases the chances that we’ll find unintended consequences in simulation, which can be eliminated before they ever enter the real world.
Another possibility that’s farther down the line is using evolution to influence the ethics of artificial intelligence systems. It’s likely that human ethics and morals, such as trustworthiness and altruism, are a result of our evolution – and factor in its continuation. We could set up our virtual environments to give evolutionary advantages to machines that demonstrate kindness, honesty and empathy. This might be a way to ensure that we develop more obedient servants or trustworthy companions and fewer ruthless killer robots.
While neuroevolution might reduce the likelihood of unintended consequences, it doesn’t prevent misuse. But that is a moral question, not a scientific one. As a scientist, I must follow my obligation to the truth, reporting what I find in my experiments, whether I like the results or not. My focus is not on determining whether I like or approve of something; it matters only that I can unveil it.

We all, individually and as a society, need to prepare for that nightmare scenario, using the time we have left to demonstrate why our creations should let us continue to exist. Or we can decide to believe that it will never happen, and stop worrying altogether. But regardless of the physical threats superintelligences may present, they also pose a political and economic danger. If we don’t find a way to distribute our wealth better, we will have fueled capitalism with artificial intelligence laborers serving only very few who possess all the means of production.

Future of jobs… it’s more complicated than you think…
This video is the best argument I have seen in a easily understood context about the coming challenges of employability and future jobs with the rise of automation. Although I have a problem with the use of horses as a metaphor to understand future human jobs. Particularly in that the horses were never the creator of their own jobs where as we humans are the creator of our jobs. I do believe there will be tremendous deflation in the future that currently is expressing itself in some type of technological deflation which is hard to measure using the tools of an industrial time such as GDP. This deflation will make so many of our living expenses smaller. This will probably first be seen in the tremendous lowering of the cost of transportation. And has already been seen as mentioned above in the decreasing cost of technological gadgets and services. This will leave our largest cost as housing and healthcare. It is possible that a universal basic income will become increasingly likely. Indeed we live in interesting times.

AI checkup…

This conference adds much needed perspective on the AI revolution and the current state of the art. Indeed we are many breakthroughs away from artificial general intelligence.  The current level of technology does not quite justify the current level of hype. However my interest in AI is related to the fact that of all the current narratives that can drive investors to want to put their money in a stock, AI and it’s many sub-fields (such as transportation as a service, big data, etc) it’s by far the most beguiling story.

While the growth of deep neural networks has helped propel the field of machine learning to new heights, there’s still a long road ahead when it comes to creating artificial intelligence. That’s the message from a panel of leading machine learning and AI experts who spoke at the Association for Computing Machinery’s Turing Award Celebration conference in San Francisco today.

We’re still a long way off from human-level AI, according to Michael I. Jordan, a professor of computer science at the University of California, Berkeley. He said that applications using neural nets are essentially faking true intelligence but that their current state allows for interesting development.

“Some of these domains where we’re faking intelligence with neural nets, we’re faking it well enough that you can build a company around it,” Jordan said. “So that’s interesting, but somehow not intellectually satisfying.”

Those comments come at a time of increased hype for deep learning and artificial intelligence in general, driven by interest from major technology companies like Google, Facebook, Microsoft, and Amazon.

Fei-Fei Li, who works as the chief scientist for Google Cloud, said that she sees this as “the end of the beginning” for AI, but says there are still plenty of hurdles ahead. She identified several key areas where current systems fall short, including a lack of contextual reasoning, a lack of contextual awareness of their environment, and a lack of integrated understanding and learning.

“This kind of euphoria of AI has taken over, and [the idea that] we’ve solved most of the problem is not true,” she said.

One pressing issue identified by Raquel Urtasun, who leads Uber’s self-driving car efforts in Canada, is that the algorithms used today don’t model uncertainty very well, which can prove problematic.

“So they will tell you that there is a car there, for example, with 99 percent probability, and they will tell you the same thing whether they are wrong or not,” she said. “And most of the time they are right, but when they are wrong, this is a real issue for things like self-driving [cars].”

The panelists did concur that an artificial intelligence that could match a human is possible, however.

“I think we have at least half a dozen major breakthroughs to go before we get close to human-level AI,” said Stuart Russell, a professor of computer science and engineering at the University of California, Berkeley. “But there are very many very brilliant people working on it, and I am pretty sure that those breakthroughs are going to happen.”

New Republic on conspiracies… must read…

Conspiracy theories spread like measles: First they infect the weak and vulnerable; then they spread like wildfire among the entire population. Researchers have found that if a person believes in one conspiracy, he or she is more likely to believe in others—even those unrelated to the initial theory. Which is to say, once conspiracy becomes part of our beliefs, it can be harder to see the world as it truly is. Conspiracy depends on a rejection of the world as it appears to be. Once this belief takes root, it becomes harder and harder to differentiate truth from fiction.

In other words, it is not the methodology of conspiracy that’s the problem. When paranoid thinking opens up possibilities, it can serve a useful function. The danger comes when conspiracists remain wedded to their theories in the face of conflicting information, when they refuse to do the hard work of confirming and substantiating their own assumptions and beliefs. Woodward and Bernstein did not simply point to a trail of shady campaign contributions and tweet that Nixon was behind it all. They followed the facts, step by painstaking step, all the way to the Oval Office.

The promise of conspiracy—that it will assuage our anxiety—is a false one. Watching Donald Trump from the social media sidelines, expecting at any minute that the Deep State will appear and fire a single magic bullet from the Grassy Knoll and put everything right again, is a dangerous delusion. It offers false assurance that you, as one lone individual, can’t do anything, even though American democracy has never needed you more.