A major breakthrough in artificial intelligence has been announced, with a computer beating the world's best players at competitive eSports for the first time.
OpenAI is a non-profit research company established in 2015 whose founders include Elon Musk. They report that, for the first time, a computer program has beaten the world's best human players at competitive eSports – in this case, the game Defence of the Ancients 2 (aka Dota 2).
Earlier this year, Google's AlphaGo defeated the world's number one player at Go, a board game similar to chess but with far more complexity. Dota 2, however, is orders of magnitude more complex than even Go. This multiplayer battle arena pits two teams of five against each other, with each team occupying and defending their own separate base on a map. Each of the ten players independently controls a powerful character, known as a "hero", who all have unique abilities and styles of play. There are 113 to choose from. During a match, a player and their team collects experience points and items for their heroes in order to fight the opposing team's heroes and other defences. A team wins by being the first to destroy a large structure in the enemy base called the "Ancient".
This week, at the annual tournament hosted in Seattle, OpenAI has been demonstrating its bot, which mastered the game from scratch by self-play and does not use imitation learning or tree search. According to Greg Brockman, the company's Chief Technology Officer, self-playing is a more effective way for AI to learn complex tasks – as opposed to fighting much weaker enemies, or overwhelmingly strong ones. By playing against a copy of itself, the bot always has a worthy opponent and therefore remembers more useful information. After many thousands of practice runs, it gradually learned which moves worked best and which to avoid. It developed a number of behaviours that are demonstrated in the video below.
Whereas chess uses an 8 x 8 board, and the game Go has a 19 x 19 board (with each cell being either blank or occupied), Dota 2 is vastly more complex with a gaming "board" of 15,000 x 15,000 units, a lot of hidden information, and many more variables with every action. The OpenAI bot was able to handle this level of detail, however.
In a series of 1v1 matches, it went up against the top human champions in the world, beating them all. This included SumaiL (world's best 1v1 player) and Arteezy (top overall player in the world), as well as 27-year-old Danil "Dendi" Ishutin. At times, the movements of the bot were eerily human-like, with Dendi saying it "feels a little like [a] human, but a little like something else."
Elon Musk today praised OpenAI on Twitter. He has previously warned that AI represents a "fundamental risk to the existence of civilisation" and hopes the organisation will foster the creation of safer, more benevolent forms of machine intelligence.
OpenAI first ever to defeat world's best players in competitive eSports. Vastly more complex than traditional board games like chess & Go.
OpenAI is now developing its bot further, and hopes it can take part in a proper five-on-five match next year, as opposed to just 1v1. The company will also explore using its neural network for other games.
A large survey of experts in artificial intelligence suggests there is a 50% chance of AI outperforming humans in all tasks within 45 years and of automating all human jobs in 120 years.
A team from the University of Oxford and Yale University has published a new survey that reveals experts' opinions on the likely timing of future milestones in artificial intelligence (AI). For this paper, they wrote to all researchers who had published at the 2015 NIPS and ICML conferences (two of the premier venues for peer-reviewed research in machine learning). Of the 1,634 authors contacted, a total of 352 (21%) responded.
Questions given to participants concerned the timing of specific AI capabilities (e.g. folding laundry, language translation), superiority at specific occupations (e.g. truck driver, surgeon), superiority over humans at all tasks, and the social impacts of advanced AI.
According to this survey, experts believe that AI will outperform humans in many activities in the near future, such as translating languages (by 2024), writing a high-school essay (by 2026), driving a truck (by 2027), working in retail (by 2031), writing a bestselling book (by 2049), and working as a surgeon (by 2053). Researchers believe there is a 50% chance of AI outperforming humans in all tasks within 45 years (shown in the graph below), and of automating all human jobs within 120 years.
Aggregate subjective probability of 'High-Level Machine Intelligence' (HLMI) arrival by future years. Each respondent provided three data points for their forecast.
Interestingly, Asian respondents predict these dates much sooner than North Americans. There was no correlation between the seniority of a researcher and the predictions they made. A majority of researchers believe the field of machine learning has accelerated in recent years, with 67% saying progress was faster in the second half of their career.
While there is much hype and fear about the danger of robots, depicted in movies such as Terminator, the researchers take a more optimistic view about the longer-term future of AI. The majority believe it will benefit humanity, with only a 5% chance of being "extremely bad" (i.e. causing human extinction).
"These results will inform discussion amongst researchers and policymakers about anticipating and managing trends in AI," the survey team concludes.
Most of the questions focus on the cognitive aspects of intelligence that fit well-defined tasks. "But parts of intelligence – such as emotional intelligence – go beyond cognition," says Georgios Yannakakis, an Associate Professor at the Institute of Digital Games, University of Malta. "It would be interesting to ask when AI will surpass humans at being art or movie critics."
The full survey results can be downloaded as a PDF at arXiv.org.
The vessel "YARA Birkeland" will be the world's first fully electric and autonomous container ship, with zero emissions. Operation is planned to start in Norway during the latter half of 2018, with fully autonomous journeys from 2020.
"YARA Birkeland" will be the world's most advanced container feeder ship – integrating sensor, control, communications and power into a fully autonomous and 100% electric system.
Named after the company YARA's founder – the scientist and innovator Kristian Birkeland – it will be the world's first fully electric container feeder, reducing NOx and CO2 emissions and improving road safety by removing the need for up to 40,000 truck journeys in populated urban areas.
YARA Birkeland will initially operate as a manned vessel, shifting to remote operations in 2019 and performing fully autonomous operations from 2020. This new zero-emission ship could be a game-changer for global maritime transport, helping to meet the UN sustainability goals for 2030 and beyond. Shipping is currently responsible for 2.5% of global human-made CO2 emissions, but this could rise to 17% by 2050 if no action is taken. There are now over 100,000 transport ships at sea, of which about 6,000 are large container ships.
"As a leading global fertiliser company with a mission to feed the world and protect the planet, investing in this zero emission vessel to transport our crop nutrition solutions fits our strategy well. We are proud to work with KONGSBERG to realise the world's first autonomous, all-electric vessel to enter commercial operation," says Svein Tore Holsether, President and CEO of YARA.
"Every day, more than 100 diesel truck journeys are needed to transport products from YARA's Porsgrunn plant to ports in Brevik and Larvik where we ship products to customers around the world. With this new autonomous battery-driven container vessel, we move transport from road to sea and thereby reduce noise and dust emissions, improve the safety of local roads, and reduce NOx and CO2 emissions," says Holsether.
KONGSBERG is responsible for development and delivery of all key enabling technologies on the YARA Birkeland, including the sensors and integration required for remote and autonomous operations, in addition to the electric drive, battery and propulsion control systems.
"By moving container transport from land to sea, YARA Birkeland is the start of a major contribution to fulfilling national and international environmental impact goals. The new concept is also a giant step forward towards increased seaborne transportation in general," says Geir Håøy, CEO of KONGSBERG.
As a leading global maritime technology company, KONGSBERG's integrated control and monitoring systems are already capable of facilitating remote and unmanned operations. YARA Birkeland will benefit from competence and new technologies developed across KONGSBERG. In addition to being autonomous and fully electric, it will also be ballast-free, to further reduce its impact on the marine environment. This lack of water discharge will help in preventing the spread of non-native, nuisance, or exotic species of plants, animals, viruses and bacteria, which can often cause extensive ecological and economic damage.
"Developing systems for autonomous operations is a major opening and natural step for KONGSBERG, considering our decades of expertise in the development and integration of advanced sensors, control and communication systems for all areas of ship operations," says Håøy. "YARA Birkeland will set the benchmark for the application of innovative maritime technology for more efficient and environmentally friendly shipping."
At the Future of Go Summit in China, Google's DeepMind AlphaGo AI program has beaten the world's number one Go player, Ke Jie.
Yet another milestone in the rapidly advancing field of artificial intelligence was achieved this week, as a software program designed to play Go was able to defeat the reigning world champion. Computer scientists and futurists had predicted that AI would need at least another two decades before it could master the game.
Despite its relatively simple rules, Go is extremely complex – even more so than chess – and has more possible combinations of moves than the total number of atoms in the visible universe. Compared to chess, Go has both a larger board with more scope for play and longer games, and, on average, many more alternatives to consider per move.
AlphaGo was developed by DeepMind, a London-based software company acquired by Google for more than $500m in 2014. In October 2015, it became the first computer program to beat a human professional Go player without handicaps on a full-sized 19×19 board. In March 2016, it beat the legendary Lee Sedol in four of five games, the first time a computer program has beaten a 9-dan professional without handicaps. In December 2016, it was chosen by the prestigious journal Science as one of the Breakthrough of the Year runners-up. And now, it has beaten Ke Jie, who has continuously held the world No. 1 ranking for two years, winning three of three games against him. The final match, which lasted for 209 moves, is pictured above.
"I'm very sorry I lost," said the 19-year-old, during a post-match press conference in Wuzhen, Zhejiang province. "I wish I could have done better."
AlphaGo's algorithm uses a Monte Carlo tree search to decide its moves based on knowledge previously "learned" by machine learning; specifically by an artificial neural network (a deep learning method) by extensive training, both from human and computer play.
DeepMind founder, Demis Hassabis, said Ke Jie had played "perfectly" and "pushed AlphaGo right to the limit." The company says the eventual plan is to deploy its artificial intelligence "in areas of medicine and science." They plan to publish an academic paper later this year that will detail the extensive set of improvements made to the algorithms' efficiency and potential to be generalised across a broader set of problems.
Urgent reform is needed to deal with the rapid rise of automation, a leading Scottish think-tank has said.
Urgent reform is needed to deal with the rapid rise of automation, which threatens nearly half of Scottish jobs by 2030, a leading think-tank has said. The stark warning comes in a new report published by the Institute for Public Policy Research (IPPR) in Scotland, a leading progressive think tank, and supported by the JPMorgan Chase Foundation.
The report, Scotland's Skills 2030, outlines the need to reskill Scotland's workforce for the world of work in the coming decades. With greater numbers of workers working for longer, due to demographic change, and in multiple jobs, multiple careers and for multiple employers, due to technological change, Scotland will need to retrofit the workforce with the skills required to compete in the future.
There are 2.5m working age adults today (78%) that have left compulsory education, that will still be of the working age by 2030, the study notes – adding they are likely to experience significant changes to the economy over this time, and will need support to learn new skills, retrain and upskill.
Meanwhile, just under half (46.1%) of jobs in Scotland, about 1.2m jobs, are at "high risk" of automation over the next couple of decades. The sectors most likely to be affected are transport, manufacturing and retail, the report states. This brings the need for a skills system that is able to work with people in jobs, throughout their careers, rather than solely at the start or before their careers have begun, the researchers warn.
Scotland has a clear gap in training and learning for people who have already started their careers, with a greater focus on younger people, and full-time provision in recent years. Employers are not plugging this gap, and too often pursue a low-skill business model. IPPR Scotland is calling for a new mid-career learning route, called the Open Institute of Technology, to sit alongside apprenticeships and further education, to help train the current workforce to be ready for the future challenges Scotland's economy faces, the report concludes.
Russell Gunson, Director of IPPR Scotland, said: "There are more than 2.5 million people already in the workforce today that will still be working by 2030. There are also 1.2m jobs in Scotland at risk of automation over the same time. Scotland urgently needs to design a skills system better able to work with people already into their careers to help them to retrain, reskill and respond to world of work of 2030.
"Scotland has a really strong record on skills in many ways, and in this report we find that Scotland is the highest skilled nation in the UK. However, our system has a clear gap in that we don't have enough provision for people who have already started their careers, and employers are not investing to fill this gap. To respond to the huge changes facing Scotland around demographic, technological and climate change – and of course Brexit – we're going to have to focus on retrofitting the current workforce to provide them with the skills they need, to deliver the inclusive economic growth we wish to see.
"Our report makes a number of recommendations to help Scotland plot a path through these challenges, to reform the skills system in Scotland, to help to secure an economy that delivers fairness and reduces inequality. Without reform of the skills system we could see changes to the economy harm whole sections of population, and whole communities, leaving many behind."
The University of Utah has revealed a new robotic drill system for greatly speeding up surgical procedures. One type of complex cranial surgery could be done in a fiftieth of the normal time, decreasing from two hours to just two and a half minutes.
A computer-driven automated drill, similar to those used to machine auto parts, could play a pivotal role in future surgical procedures. The new machine can make one type of complex cranial surgery 50 times faster than standard procedures, decreasing from two hours to two and a half minutes. Researchers at the University of Utah developed a drill that produces fast, clean and safe cuts – reducing the time the wound is open and the patient is anesthetised, thereby decreasing the incidence of infection, human error, and surgical cost. The findings are reported in Neurosurgical Focus.
To perform complex surgeries – especially cranial surgeries – surgeons typically use hand drills to make intricate openings, adding hours to a procedure: "It was like doing archaeology," said William Couldwell, study author and neurosurgeon at the University of Utah Health. "We had to slowly take away the bone to avoid sensitive structures."
Couldwell saw a need for a device that could alleviate this burden and make the process more efficient: "We knew the technology was already available in the machine world, but no one ever applied it to medical applications."
"My expertise is dealing with the removal of metal quickly, so a neurosurgical drill was a new concept for me," explained A. K. Balaji, associate professor in mechanical engineering. "I was interested in developing a low-cost drill that could do a lot of the grunt work to reduce surgeon fatigue."
Credit: University of Utah
The team developed the drill from scratch, as well as new software to calculate the safest cutting path. First, the patient is imaged using CT scans to gather bone data and identify the exact location of sensitive structures, such as nerves, veins and arteries that must be avoided. Surgeons then use this information to program a cutting path for the drill: "The software lets the surgeon choose the optimum path from point A to point B, like Google Maps," says Balaji. In addition, the surgeon can program safety barriers along the cutting path within 1 mm of sensitive structures. "Think of the barriers like a construction zone," says Balaji. "You slow down to navigate it safety."
The translabyrinthine surgery is performed thousands of times a year to expose slow-growing, benign tumours that can form at auditory nerves. This cut must avoid several sensitive features, including facial nerves and the venous sinus, a large vein that drains blood from the brain. Risks of this surgery include loss of facial movement. The system developed at Utah has an automatic emergency shut-off switch. During surgery, facial nerves are monitored for any signs of irritation: "If the drill gets too close to the facial nerve and irritation is monitored, the drill automatically turns off," says Couldwell.
The new drill could reduce the duration of this complex procedure from two hours for hand-drilling by an experienced surgeon to two and a half minutes. The shorter surgery is expected to lower the chance of infection and improve post-operative recovery. It also has potential to substantially reduce the cost of surgery, because it shaves hours from operating room time.
The team has now demonstrated the safety and speed of the drill by performing this complex cut – but Couldwell stresses that it can be applied to many other procedures: "This drill can be used for a variety of surgeries, like machining the perfect receptacle opening in the bone for a hip implant," he said.
The varied application of the drill highlights another factor that drew Balaji to the project: "I was motivated by the fact that this technology could democratise health care by levelling the playing field so more people can receive quality care," he said. The team is now examining opportunities to commercialise the drill to ensure that it is more widely available for other surgical procedures.