Think you’re good at classic arcade games such as Space Invaders, Breakout and Pong? Think again. In a groundbreaking paper published today in Nature, a team of researchers led by DeepMind co-founder Demis Hassabis reported developing a deep neural network that was able to learn to play such games at an expert level. What makes this achievement all the more impressive is that the program was not given any background knowledge about the games. It just had access to the score and the pixels on the screen. It didn’t know about bats, balls, lasers or any of the other things we humans need to know about in order to play the games. But by playing lots and lots of games many times over, the computer learnt first how to play, and then how to play well. A machine that learns from scratch This is the latest in a series of breakthroughs in deep learning, one of the hottest topics today in artificial intelligence (AI). Actually, DeepMind isn’t the first such success at playing games. Twenty years ago a computer program known as TD-Gammon learnt to play backgammon at a super-human level also using a neural network. But TD-Gammon never did so well at similar games such as chess, Go or checkers (draughts). In a few years time, though, you’re likely to see such deep learning in your Google search results. Early last year, inspired by results like these, Google bought DeepMind for a reported UK£500 million. Many other technology companies are spending big in this space. Baidu, the “Chinese Google”, set up the Institute of Deep Learning and hired experts such as Stanford University professor Andrew Ng. Facebook has set up its Artificial Intelligence Research Lab which is led by another deep learning expert, Yann LeCun. And more recently Twitter acquired Madbits, another deep learning startup. What is the secret sauce behind deep learning? Geoffrey Hinton is one of the pioneers in this area, and is another recent Google hire. In an inspiring keynote talk at last month’s annual meeting of the Association for the Advancement of Artificial Intelligence, he outlined three main reasons for these recent breakthroughs. First, lots of Central Processing Units (CPUs). These are not the sort of neural networks you can train at home. It takes thousands of CPUs to train the many layers of these networks. This requires some serious computing power. In fact, a lot of progress is being made using the raw horse power of Graphics Processing Units (GPUs), the super fast chips that power graphics engines in the very same arcade games. Second, lots of data. The deep neural network plays the arcade game millions of times. Third, a couple of nifty tricks for speeding up the learning such as training a collection of networks rather than a single one. Think the wisdom of crowds. What will deep learning be good for? Despite all the excitement though about deep learning technologies there are some limitations over what it can do. Deep learning appears to be good for low level tasks that we do without much thinking. Recognising a cat in a picture, understanding some speech on the phone or playing an arcade game like an expert. These are all tasks we have “compiled” down into our own marvellous neural networks. Cutting through the hype, it’s much less clear if deep learning will be so good at high level reasoning. This includes proving difficult mathematical theorems, optimising a complex supply chain or scheduling all the planes in an airline. Where next for deep learning? Deep learning is sure to turn up in a browser or smartphone near you before too long. We will see products such as a super smart Siri that simplifies your life by predicting your next desire. But I suspect there will eventually be a deep learning backlash in a few years time when we run into the limitations of this technology. Especially if more deep learning startups sell for hundreds of millions of dollars. It will be hard to meet the expectations that all these dollars entail. Nevertheless, deep learning looks set to be another piece of the AI jigsaw. Putting these and other pieces together will see much of what we humans do replicated by computers. If you want to hear more about the future of AI, I invite you to the Next Big Thing Summit in Melbourne on April 21, 2015. This is part of the two-day CONNECT conference taking place in the Victorian capital. Along with AI experts such as Sebastian Thrun and Rodney Brooks, I will be trying to predict where all of this is taking us. And if you’re feeling nostaglic and want to try your hand out at one of these games, go to Google Images and search for “atari breakout” (or follow this link). You’ll get a browser version of the Atari classic to play. And once you’re an expert at Breakout, you might want to head to Atari’s arcade website. This article was originally published on The Conversation. Read the original article.
If you thought it has been a while since you heard any more rumours about the long-awaited Apple TV, they are about to be replaced by even more exciting possibility that Apple may be about to build an electric car. The Wall Steet Journal kicked things off with a report that Apple had been hiring “hundreds” of staff with automotive design skills to work on a project called “Titan” that may be a self-driving electric vehicle configured in a (not-so-exciting) mini-van design. There are several back-stories to this potential move by Apple. In one, we see continuing competition with rival Google, who has been working on a driverless car for some time and are saying that they will be launching a commercial version onto the market between 2017 and 2020. Google’s motivation behind the self-driving car has been the development of the artificial intelligence software capable of pulling off this feat. Even if the car is not successful, the AI software will have a range of applications and possibility that would make the project still worthwhile. Increasingly, Apple has shown its willingness to develop its own capability in a range of competitive technologies that it can incorporate into products. In another back-story, there is electric car company Tesla whose CEO, Elon Musk, has claimed that it will be as big financially, as Apple, within a decade. This will in part be based on the release of the Model 3, an affordable (US $35,000) family car with a range of 200 miles. Part of Tesla’s strategy appears to include the poaching of numerous Apple staff. Although it seems that Apple has been reciprocating by offering Tesla staff large signing bonuses to move to Apple. And finally there is the view that electric cars, self-driving or otherwise, represent the future of transportation, especially a climate-friendly and sustainable one. At first sight, this may be a bit hard to believe when you consider that the top 3 selling vehicles in the US in 2014 were “pickup trucks”. At the same time, hybrid electric vehicles represented less than 3% of all cars sold. Still, there is continuing interest by the car manufacturers in producing electric cars, if only as a hedge. GM has announced their new 200 mile range Chevy Bolt that will retail at around the same price as Tesla’s Model 3. There is little doubt that Apple could move into car manufacturing. With US $180 billion in cash, it could buy Fiat Chrysler, Tesla, General Motors and Ford outright. There is also no doubt that with its ability to bring design and innovative computing to an industry employing technology that significantly lags that found in an iPhone. Apple and Google have both made moves to create in-vehicle media interfaces based on their systems. Apple’s CarPlay will start to appear in cars this year. Customers who can’t wait can buy after-market devices from Pioneer. Apple’s motivation to build an electric car may be driven by competition with Google, Tesla and others. It may be also finding a new business that doubles its value to $1.3 trillion as predicted by Carl Icahn. Alternatively however, it may be genuinely interested in building a technology that makes driving more sustainable and less dependent on oil. Apple is set to invest $3 billion in new solar farms in California and Arizona to provide energy for its operations there. Apple CEO Tim Cook recently told investors: “We know that climate change is real,” Cook said on Tuesday. “Our view is that the time for talk has passed, and the time for action is now. We’ve shown that with what we’ve done.” Whether Apple’s electric cars are aimed at combating climate change will depend on how they are manufactured and how the recharging infrastructure, which is still largely to be built in the US and globally, is run. Apple throwing its weight behind this infrastructure being built at all would certainly help making electric cars a more popular possibility. This article was originally published at The Conversation.
Autonomous vehicles, or self-driving cars, are likely to be seen more widely on roads in 2015. Already, legislation authorising the use of autonomous vehicles has been introduced in the US states of Nevada, Florida, California and Michigan, with similar legislation being planned for the UK. To date, these laws have focused on legalising the use of autonomous vehicles and dealing, to an extent, with some of the complex issues relating to liability for accidents. But as with other emerging disruptive technologies, such as drones and wearables, it is essential that issues relating to user privacy and data security are properly addressed prior to the technologies being generally deployed. Understanding autonomous vehicles There is no single, uniform design for autonomous vehicles. Rather, it is best to understand an autonomous vehicle as a particular configuration of a combination of applications, some of which – such as adaptive cruise control, lane departure warnings, collision avoidance and parking assistance – are already part of current car design. The most well-known prototype, Google’s self-driving car, uses a variety of technologies, including: a laser range finder (LIDAR) that generates a detailed 3D map of the environment; radars; cameras for detecting traffic lights; and a GPS. Other projects, including prototypes being developed by Mercedes-Benz, Volkswagen, Toyota and Oxford University, use different combinations of technologies. This means that the privacy and data security problems arising from autonomous vehicles depend upon the precise technologies applied in any particular design. Some generalisations are, however, possible. The relationship between the virtual and the real The rules (or “code”) governing the online world have been different to those that apply offline. For example, online activities invariably generate digital traces, including metadata, which can be used to build profiles of users. With emerging technologies, such as drones, wearables and autonomous vehicles, we are increasingly seeing the transposition of virtual models onto the real. One consequence of the range of sensors and data collection devices being deployed (and interconnected) is that our offline activities can leave traces at least as extensive as those generated online. One way to understand types of autonomous vehicles is by reference to the kind of data collected and the ways in which that data is processed. For instance, autonomous vehicles often incorporate event recorders, or “black boxes”, to provide essential information in the event of an accident. This raises questions about who has rights to this data and about who can have access to the data. Anonymising data There is an overlap here with questions of liability, as insurance companies have clear incentives to collect as much data about user behaviour as possible. The potential for intrusive surveillance of personal activities is particularly jarring, as the car has been an archetypal space of personal privacy and freedom. A fundamental distinction must be drawn between self-contained autonomous vehicles, in which the data collected from sensor devices installed in the car are stored and processed in the vehicle itself, and interconnected vehicles, in which data is shared with a centralised server and, potentially, with other vehicles. Regardless of whether a vehicle is self-contained or interconnected, design decisions have to be made about whether or not the data collected is anonymised or linked to individual users. If the data is not anonymised, especially with interconnected vehicles, this poses serious surveillance threats. After all, once the data exists, and especially if it is connected to a server, it is vulnerable to access by third parties. It is possible to envisage implementations of autonomous vehicles where data about a particular user is linked to other data sources, such as an online profile, for purposes such as tracking or marketing. This might take the form of personalised advertising displayed in the car, or even adjusting a vehicle’s route so that it passes retail outlets which match a user’s imputed preferences. What else is at stake: human autonomy and hacking We are now familiar with technologies, such as predictive search, which in the online context, attempt to predict what we want to do and make more or less persuasive suggestions. It is likely that some versions of autonomous vehicles will implement predictive technologies. In any case, the progressive delegation of human decisions to machines raises system-wide questions about the cumulative impact on human autonomy: the more people are habituated to decisions being made for them, the less likely they may be to make their own decisions. We are also now depressingly familiar with the vulnerability of computer systems to malicious third parties. Just as effective data security is essential to online safety, autonomous vehicles must be designed with a high level of data security, especially given the potentially calamitous consequences of hacked vehicles. As interconnected data processing systems are progressively rolled out in applications such as wearables and autonomous vehicles, we seem likely to see an offline version of the same sort of perpetual guerrilla warfare played out online between information security and hackers. Protecting privacy at the design stage Autonomous vehicles promise significant social and economic benefits, especially in potential improvements to road safety. There are, nevertheless, considerable legal and regulatory challenges. As with other emerging disruptive technologies, it is vital that privacy and anonymity be properly protected at the design stage. To date, in the face of significant challenges relating to the legality of autonomous vehicles and liability issues, the privacy rights of users have been relatively neglected. But unless the era of artificial intelligence is to be accompanied by us sleepwalking into ubiquitous surveillance, we must recognise that safety and security needs to be balanced against the legitimate rights of people to control their own data and to retain their fundamental rights to privacy. David Lindsay is a board member of the Australian Privacy Foundation. This article was originally published on The Conversation. Read the original article.
The cofounder of a pioneering Sydney-based robotics startup, with a Powerhouse Museum display and a successful crowdfunding campaign under its belt, says the sector is set to get much bigger but finance for projects remains an issue. Robological cofounder Damith Herath told Private Media there are a number of exciting robotics startups founded by Australians, including Marathon Robotics and Navisens, and the sector is gaining momentum globally. “It’s kinda like the ‘70s in computing and the ‘90s in the web. It’s the same feeling in the robotics community and the general consensus is it’s getting a lot bigger,” Herath says. “A few good examples are some of the startups Google has recently purchased, or Baxter, or Cynthia Breazeal, who quit her job at MIT to do a startup called Jibo and raised $2 million on Indiegogo. “But we have to be careful, because a lot of people over-promise and under deliver. Robots will move into other spaces, though not in the anthropomorphic sense. “One of the issues is finding people to finance you is tricky, especially for hardware. People are more comfortable with apps and things that get a quick return on their investment.” In January, Robological raised $3031 on Indiegogo for Ro-buddy, a pre-built board that integrates with an Android app, making it easy to build a robot without needing to learn a programming language such as C. Herath says the startup is finalising the board for fabrication in China. “You can build a Raspberry Pi robot straight away, plug in a camera and motors, and within 10 to 20 minutes you have a spy cam working with the Android app,” he says. “We think it’s useful because it’s in the pro-maker space, but it’s not as complex as Arduino. So if you’re building something in home automation, you can get something going with Android.” Aside from Ro-buddy, Herath says Robological does consulting and research work, including working as a research partner with the Australian distributor for Rethink Robotics’ Baxter robot and on Curtin University’s Alternative Anatomies project. It is also “chipping away” on a variation of the cloud-based internet of things robotics ideas put forward by UC Berkley professor Ken Goldberg, although Herath is remaining tight-lipped about what the project involves. The startup began with a robotics display called the Articulated Head, which was on exhibit for two years at Sydney’s Powerhouse Museum. “The three founders – Zhengzhi Zhang, Christian Kroos and I – met at the University of Western Sydney six years ago on a project called Thinking Ahead, which was a project of the Australian Research Council into AI (artificial intelligence). “We each had a slightly different background, myself with robotics engineering, Zhang with software engineering and Christian with linguistic and cognitive science. “Stelarc is one of the top performing artists in the world; an Australian artist who’s done a lot of work with robotics on stage and theatre. And the project I worked on was conceived of by Stelarc.” The project ended when funding ended, but this allowed the team to develop valuable intellectual property on robots and human interaction. The founders decided to form Robological to continue their research. One of its first projects was called Adopt a Robot, a research project looking into interactions between humans and robots. “It got a lot of good publicity because it captured the public imagination. We gave away seven robots and over six months we changed its behaviour and added a face… Each person who got a robot had to care for it and fill out a questionnaire every four to six weeks,” Herath says. Next month, Robological will jointly organise a workshop on robots and art with Curtin University as part of the Sixth International Conference on Social Robotics in Sydney. Follow StartupSmart on Facebook, Twitter, and LinkedIn.
One of the issues of self-driving vehicles is legal liability for death or injury in the event of an accident. If the car maker programs the car so the driver has no choice, is it likely the company could be sued over the car’s actions. One way around this is to shift liability to the car owner by allowing them to determine a set of values or options in the event of an accident. People are likely to want to have the option to choose how their vehicle behaves, both in an emergency and in general, so it seems the issue of adjustable ethics will become real as robotically controlled vehicles become more common. Self-drive is already here With self-driving vehicles already legal to drive on public roads in a growing number of US states, the trend is spreading around the world. The United Kingdom will allow these vehicles from January 2015. Before there is widespread adoption, though, people will need to be comfortable with the idea of a computer being in full control of their vehicle. Much progress towards this has been made already. A growing number of cars, including mid-priced Fords, have an impressive range of accident-avoidance and driver-assist technologies like adaptive cruise control, automatic braking, lane-keeping and parking assist. People who like driving for its own sake will probably not embrace the technology. But there are plenty of people who already love the convenience, just as they might also opt for automatic transmission over manual. Are they safe? After almost 500,000km of on-road trials in the US, Google’s test cars have not been in a single accident while under computer control. Computers have faster reaction times and do not get tired, drunk or impatient. Nor are they given to road rage. But as accident-avoidance and driver-assist technologies become more sophisticated, some ethical issues are raising their heads. The question of how a self-driven vehicle should react when faced with an accident where all options lead to varying numbers of deaths of people was raised earlier this month. This is an adaptation of the “trolley problem” that ethicists use to explore the dilemma of sacrificing an innocent person to save multiple innocent people; pragmatically choosing the lesser of two evils. An astute reader will point out that, under normal conditions, the car’s collision-avoidance system should have applied the brakes before it became a life-and-death situation. That is true most of the time, but with cars controlled by artificial intelligence (AI), we are dealing with unforeseen events for which no design currently exists. Story continues on page 2. Please click below. Who is to blame for the deaths? If car makers install a “do least harm” instruction and the car kills someone, they create legal liability for themselves. The car’s AI has decided that a person shall be sacrificed for the greater good. Had the car’s AI not intervened, it’s still possible people would have died, but it would have been you that killed them, not the car maker. Car makers will obviously want to manage their risk by allowing the user to choose a policy for how the car will behave in an emergency. The user gets to choose how ethically their vehicle will behave in an emergency. As Patrick Lin points out the options are many. You could be: democratic and specify that everyone has equal value pragmatic, so certain categories of person should take precedence, as with the kids on the crossing, for example self-centred and specify that your life should be preserved above all materialistic and choose the action that involves the least property damage or legal liability. While this is clearly a legal minefield, the car maker could argue that it should not be liable for damages that result from the user’s choices – though the maker could still be faulted for giving the user a choice in the first place. Let’s say the car maker is successful in deflecting liability. In that case, the user becomes solely responsible whether or not they have a well-considered code of ethics that can deal with life-and-death situations. People want choice Code of ethics or not, in a recent survey it turns out that 44% of respondents believe they should have the option to choose how the car will behave in an emergency. About 33% thought that government law-makers should decide. Only 12% thought the car maker should decide the ethical course of action. In Lin's view it falls to the car makers then to create a code of ethical conduct for robotic cars. This may well be good enough, but if it is not, then government regulations can be introduced, including laws that limit a car maker’s liability in the same way that legal protection for vaccine makers was introduced because it is in the public interest that people be vaccinated. In the end, are not the tools we use, including the computers that do things for us, just extensions of ourselves? If that is so, then we are ultimately responsible for the consequences of their use. David Tuffley does not work for, consult to, own shares in or receive funding from any company or organisation that would benefit from this article, and has no relevant affiliations. This article was originally published on The Conversation. Read the original article. Follow StartupSmart on Facebook, Twitter, and LinkedIn.
No sleep needed: New technologies are emerging that could radically reduce our need to sleep - if we can bear to use them, writes Jessa Gamble for aeon magazine. Imagine a disease that cuts your conscious life by one-third. You would clamour for a cure. We’re talking about sleep. There may be no cure yet for sleep, but the palliatives are getting better. “Work, friendships, exercise, parenting, eating, reading — there just aren’t enough hours in the day,” Gamble writes. “To live fully, many of us carve those extra hours out of our sleep time. Then we pay for it the next day. A thirst for life leads many to pine for a drastic reduction, if not elimination, of the human need for sleep. Little wonder: if there were a widespread disease that similarly deprived people of a third of their conscious lives, the search for a cure would be lavishly funded. It’s the Holy Grail of sleep researchers, and they might be closing in.” Dilbert does startup: When Dilbert cartoonist Scott Adams turned himself to entrepreneurship, he wasn’t prepared for some of the weirder ways of Silicon Valley. Describing himself as an “embedded journalist” this week he takes on the pivot. “The Internet is no longer a technology,” Adams writes. “The Internet is a psychology experiment. Building a product for the Internet is the easy part. “Getting people to understand the product and use it is the hard part. The only way to make the hard part work is by testing one hypothesis after another. Every entrepreneur is a behavioral psychologist with the tools to pull it off.” And he’s distilled it all down in “the system”, which looks like this: 1. Form a team 2. Slap together an idea and put it on the Internet. 3. Collect data on user behavior 4. Adjust, pivot, and try again What the gospel of innovation gets wrong: “In the last years of the nineteen-eighties, I worked not at startups but at what might be called finish-downs,” write Jill Lepore in a piece titled ‘The Disruption Machine’ in The New Yorker. Lepore’s thesis is that Clayton Christensen’s theory of disruption, accepted across American industry as “the gospel of innovation”, is wobbly at best because it rests on a group of handpicked case studies that prove little or nothing. “Most of the entrant firms celebrated by Christensen as triumphant disrupters no longer exist, their success having been in some cases brief and in others illusory,” writes Lepore. Anyone who has anything to do with the startup industry will relate to this point: “Ever since “The Innovator’s Dilemma,” everyone is either disrupting or being disrupted,” she writes. “There are disruption consultants, disruption conferences, and disruption seminars. This fall, the University of Southern California is opening a new program: “The degree is in disruption,” the university announced. “Disrupt or be disrupted,” the venture capitalist Josh Linkner warns in a new book, “The Road to Reinvention,” in which he argues that “fickle consumer trends, friction-free markets, and political unrest,” along with “dizzying speed, exponential complexity, and mind-numbing technology advances,” mean that the time has come to panic as you’ve never panicked before.” Don’t worry about the robots: Venture capitalist Marc Andreessen does not believe that robots will eat jobs. “Robots and AI are not nearly as powerful and sophisticated as people are starting to fear, writes Andreessen, “With my venture capital hat on I wish they were, but they’re not. There are enormous gaps between what we want them to do, and what they can do. There is still an enormous gap between what many people do in jobs today, and what robots and AI can replace. There will be for decades.” Image credit: Flickr/jdhancock
The struggles of the Australian toy market have been put into sharp focus by a US-based start-up founded by a former Pixar executive, which has raised $11.5 million for an internet-connected, artificially intelligent teddy bear.
David Urpani doesn’t like to stay still for long. He went from being an architect to a doctor of artificial intelligence to the founder of insurance comparison giant iSelect in 2000.
A new recycling “ATM” will take an old mobile phone and pay an agreed price on the spot, taking the concept of bartering to a new level.
Concerns were raised by economic soothsayers this week when a new report predicted the end of the mining boom – Australia’s runaway success story – within two years.
The director of Saber Astronautics has highlighted opportunities in Australia’s fledgling space industry, after his company was chosen as a finalist in the NewSpace Business Plan Competition.
Artificially intelligent machines that can converse and argue with humans are just years away, according to scientists in the United Kingdom, as speech technology starts to take off.