Category Archives: Ai

In BriefA startup operating out of Tel Aviv, Israel is looking tochange the way artificial intelligence is able to perceive theworld. Cortica boasts of having a transparent and verifiablemachine learning algorithm that makes it possible for machines tolearn, classify, and represent things without guidance. Seeing the World as We Do

On the second floor of a small office building in the middle of Tel Aviv, the bustling heart of Israels booming tech industry sits the world headquarters for Cortica, an AI company with the ambitiousgoal of getting machines to see the world as well as we do.Click to View Full Infographic

They are one of hundreds of AI startups that have sprouted up all over the world in the last few years. The global AI market has now exceeded a billion dollarsandthe tech giants are racing to acquire them as they are re-positioning themselves as AI companies first.

But its a convoluted space with a high knowledge barrier of entry. Most of these companies use buzz words like machine learning, deep learning or neural networks knowing that most consumers and investors have no idea what they really mean.For someone without a background in the field it can be hard to distinguish substance from snake oil.

What makes Cortica different is that unlike most machine learning algorithms that are essentially blackboxes, systems where even the programmers dont know what the system is doing, Corticas program is transparent and verifiable.

It also relies on a branch of AI research called unsupervised learning. Most AI companies rely on algorithms that have to be taught by humans how to characterize and represent whatever they are learning. It is a much slower and more tedious process that has far less chance of revealing anything novel. Unsupervised learning algorithms, such as the ones Cortica uses, are capable of learning, classifying and representing things without guidance.Cortica was founded by Professor Josh Zeevi and two of his doctoral graduates, Igal Raichelgauz and Karina Odinaev from Technion Israel Institute of Technology. Israel, a tiny country of just over 8 million people, has emerged as a global hub of technical innovation. Israel now has more companies listed on the New York stock exchange than any country besides America and China and 50% of those Israeli firms can trace their origins to Technion. If Corticas aspirations come to fruition, it will catapult itself to the top of a long list of successful

Cortica was founded by Professor Josh Zeevi and two of his doctoral graduates, Igal Raichelgauz and Karina Odinaev from Technion Israel Institute of Technology. Israel, a tiny country of just over 8 million people, has emerged as a global hub of technical innovation. Israel now has more companies listed on the New York stock exchange than any country besides America and China and 50% of those Israeli firms can trace their origins to Technion. If Corticas aspirations come to fruition, it will catapult itself to the top of a long list of successfulIsraeli startups that have earned the country its moniker as the startup nation.

As Prof. Zeevi states, There is no objective reason why computers should not do better than the human brain. Although no one is there yet, Cortica will get there. Their team, with offices in New York and Beijing, believe they have figured out how to reverse engineer the biological visual cortexto enable machines to see the world as well as we can.

Doing so is no trivial task as an incredible amount of processing goes on in your head whenever you open your eyes. As soon as your eyelids flicker open receptors in the back of your eye take in the visible light waves and convert it to electrical signals that relay that message to the back of your brain to sort and analyze the immense amount of information contained into the size, shape, depth, and color of all the objects in view. All that data then gets sent for further processing to your cortex which classifies everything into objects by comparing them to every other object you have ever encountered. This is how you recognize things, make sense of them and determine their function. This happens tens of times per second and gives you a sense of motion as each frame is represented by a selected group of neurons that constitutes a clique a fundamental concept in Corticas technology that was motivated by the neurophysiological function exhibited by cortical networks of neurons. This visual response is triggered instantaneously and gets compared with Corticas extensive data base of previously triggered clustered and stored image concepts represented by highly-compressed signatures of neural cliques. All told, it is a mind boggling amount of processing that happens at every moment your eyes are open and is a process that Cortica believes they have managed to accurately simulate in silicon.

A number of companies now claim to be on the verge of getting machines to do this task with the same ease and fluidity that humans can. Cortica believes that their unique approach to the problem puts them ahead of the crowd and that they will be the first to come up with a visual system on par with our own.

This will have a wide range of applications and revolutionize a number of different fields. Everything from security cameras to autonomous vehicles to satellite imagery to medical diagnostics will be vastly improved by the application of this technology.

But that is just the beginning of what the people at Cortica believe their system will be capable of.

Endowing computers with a sense of sight will be a monumental step forward in AI research. Coupled with advances in natural language processing, AI will then have the fundamental pieces needed to understand and interact with the world. This will enable AI research to move into its next phase which is the search for the holy grail of computer science; AGI, artificial general intelligence.

Unlike narrow AI that is designed to take in and spit out specific information, AGI will be able to take in a variety of inputs and give a variety of outputs. In many ways this is what separates humans from computers, computers need specific instructions and can only output what they are directed to output. But humans can take in a wide range of inputs through our various senses and then do different tasks with that information. In the eyes of programmers this makes us general purpose computers.

But we too are limited, mostly by the amount of information that we can process and the speed at which we can output information. People can really only do one task effectively at a time and can only send information at the speed at which we can talk or write. AGI has the potential to far exceed us in both.

Cortica believes that in time their technology will be a vital part of bringing about AGI. If successful, it will be the last tool we will ever need to create.

Read more:

An Israeli AI Company Is Giving Machines The Gift Of Sight - Futurism

Theres no question that the Mario Karts AI is a big cheater. Anyone who plays the game has probably noticed how the computer-controlled racers tend to catch up to you no matter what and always have an item in hand during the races final moments. But what advantages, exactly, do the robot racers get in comparison to their human opponents? The question has become relevant once more with the new Mario Kart 8 Deluxe for the Switch, in which the artificial intelligence appears to have snapped the rules of fair play in two. Or it could all be a matter of psychology, as some experts have argued.

Ive been trying to unlock every last vehicle and accessory in Deluxe. As I plowed through, I played against a lot of AI competitors. As I played, I planned to earn all three stars for every single prize cup in every level, which you can only achieve through first-place finishes. Heres the thing, though: that Mario Kart AI really doesnt want a lowly human like me to place first.

I mean, sure, human competitors dont want me to win, either. But, I swear, the AI racers in Mario Kart 8 Deluxe appear to ignore the other AI drivers and save their items just for me. Ive seen robot racers hold onto a red shell for ages, only to toss it my way as soon as I finally try to pass them.

It feels personal.

Back in the 1990s and early 2000s, people who played Mario Kart 64 and Mario Kart: Double Dash noticed that the AI had the ability to speed up after falling far behind. It was easy to spot this rubber-band effect in action in those games. More recent Mario Karts have gotten more subtle about providing a helping hand to AI racers. When Mario Kart 8 came out for the Wii U in 2014, some players noticed that the AI was using classic shortcuts on tracks. A computer controlled Mario or Bowser or whoever would use a speed boost right before cutting across a grassy inlet. Human players have done this for years, but AI racers never seemed smart enough to do it before MK8.

Some Mario Kart 8 players have also noticed that the computers speed varies based on how fast the human player goes at the outset of the race. For example, if you hang back and wait for several seconds at the start of a race, your AI opponents will go slower throughout the rest of their laps. Once you finally head out, youll easily be able to outrun them, as this video shows:

That seems only fair, though, right? If you were a bad player, youd be grateful for the AIs rubber-band speed alterations, since it would allow you to catch up and feel like a contender.

With the new version of 8, something seems to be going on with how the computer racers use items. Or its my imagination. Somethings up, though. Heres the deal:

As usual with Mario Kart, item drops in MK8D become less and less useful the longer you stay in the first position. Youll pick up a lot of low-value items like coins and bananas. Every now and then, youll get the best defensive item of all: the Super Horn, which can knock out a Blue Shell if you time it right. MK8D brings back two legacy items from earlier in the series: the Cape Feather and the Boo. That pesky little ghost is a great get for your opponents, because the Boo can steal items. If youre in first, you might have to waste your Super Horn on an incoming Boo, thereby leaving yourself open to a shell takedown. If youre doing well, youre not supposed to get these items, but guess who does get these items when youre doing well?

The higher in speed you go in Mario Kart 8 Deluxe, the more petty the AI gets and the luckier it seems to get with items. Its easy enough to place first in every 50cc and 100cc level and clean up plenty of stars and coins in the process, although the AIs anti-human opportunism will still stick out to you in these early tracks. The 150cc tracks, however? Youll have to get mighty clever in order to beat the AI in those stages. Youll hit the final few turns before crossing the third laps finish line and, somehow, the second-place AI rival will have a Boo and a shell. As soon as you get knocked back to second, youre screwed, because every other AI next in line has also been saving their items for you, and as soon as youre in striking distance, they deploy the shell onslaught. Its as though the robot racers know that theres just one human in the running and theyve got it in for her.

Nintendo didnt respond to my asking if the item distribution in MK8D offers any advantage to the AI racers. I have no proof that my AI rival always gets the Lightning Strike, the Boo, and/or a shell in the final lap. And perhaps its just my imagination, anyway. Perhaps, as a human player, Im biased against robot racers because I feel like theyve got a one-up on me.

Could it all be in my mind?

If a human competitor in Mario Kart gets an amazing item, I chalk that up to the games internal probability and I blame myself for not having a more precise banana-throwing accuracy. But if the computer gets great items? Well, that feels like the game is cheating. The computer is part of the game! It has more information than I do! The AI Mario has a car phone with a direct line to Nintendo and he can get any item he wants!

This episode of Game Psych, which zooms in on Mario Kart 8 Deluxes item distribution, subscribes to the theory that MK8Ds AI racers do indeed get better items, but that this is part of the games attempt to balance out human racers ability to problem-solve and find counter strategies:

According to Game Psych, the computer deserves better items: Humans are better drivers than AIs in Mario Kart. Humans can pull off moves AIs never will. To keep up with the humans, the AI drivers have to rely on the power items they receive from boxes, and that makes them feel cheap or unfair.

Even if thats true, its a delicate balance to strike. If the AIs items are too good, then the game swings back to feeling unfair again. I want to feel challenged, sure, but I dont want to feel like earning three stars in each race is impossible.

Other game developers have discussed this problem with designing AI competitors before. Take Civilization lead creator Sid Meier, who gave a talk at GDC in 2010 titled The Psychology of Game Design (Everything You Know Is Wrong). In the talk, he explained that players want the AI to feel like a human competitor, but that if the AI is too good, the human opponent will suspect the robot peeked behind the curtain.

I rewatched several of my replays to find more definitive proof that the AI players tend to gang up on me. I can only observe what the robots do when theyre close to me, but you know what? They totally hate me. They aim their items at me whenever possible. However, I did notice one instance when AI players too far down the track to identify lobbed shells at one another. So I guess the robots do fight when Im not in shooting distance, but if they can see me, they do all they can to take me out.

But lets look at it from the robots perspective, shall we? Im an unfair opponent myself! I have the ability to drag a banana behind me or save a Super Horn for the perfect moment. Ive got those killer benefits that come with being human: creativity and unpredictability.

Philip Tan, creative director for the MIT Game Lab, described this phenomenon to Compete as an example of how players can do things against AI that AI cant do to players, and players dont consider that cheating. Although the AI has become increasingly humanlike since Mario Kart 8, having gained more drifting and shortcut capabilitiesthe robot racers still cant match my cleverness with counter-items and blocking. If they could, then perhaps they wouldnt need the extra Boo and Lightning Bolt drops. Not that I can prove theyre getting more Boos and Lightning Bolts, because I cant, but come on. They totally are.

Tan also remarked that players rarely notice when AI racers attack other AI racers, which my own experiences have borne out. Ive listed plenty of moments here when the AI racers seemed to gang up on me, but the rest of the time, I pay them no mind because Im focused on my own strategies and racing. I only notice the AI behavior when it affects me.

Above all, I remember the times that the AI felt unfair the most. Those moments stick out the most in my mind.

I dont think about all the races that I won with no trouble. After all, those races were perfectly fair.

See the rest here:

Either I'm Paranoid Or Mario Kart 8 Deluxe's AI Is Cheating - Kotaku

Forbes

AI Influencer Andrew Ng Plans The Next Stage In His Extraordinary Career Forbes Andrew Ng is one of the foremost thinkers on the topic of artificial intelligence. He founded and led the Google Brain project which developed massive-scale deep learning algorithms. In 2011, he led the development of Stanford University's main ...

Continued here:

AI Influencer Andrew Ng Plans The Next Stage In His Extraordinary Career - Forbes

An attack by artificial intelligence on humans, said Google software engineerand University of Michigan professor Igor Markov, would be sort of like when the Black Plague hit Europe in the 14th century, killing up to 50percent of the population.

Virus particles were very small and there were no microscopes or notion of infectious diseases, there was no explanation, so the disease spread for many years, killed a lot of people, and at the end no one understood what happened, he said. This would be illustrative of what you might expect if a superintelligent AI would attack. You would not know precisely whats going on, there would be huge problems, and you would be almost helpless.

Rather than devising technological solutions, in a recent talk about how to keepsuperintelligent AI from harming humans, Markov looked to lessons from ancient history.

Markov joined sci-fi author David Brin and other influential names in the artificial intelligence community Friday at The AI Conference in San Francisco.

One lesson fromearly humans that could help in the fight againstAI: make friends. Domesticate AI the same way Homo sapiens turned wolves into their protectors and friends.

If you are worried about potential threats, then try to use some of them for protection or try to adapt or domesticate those threats. So you might develop a friendly AI that would protect you from malicious AI or track unauthorized accesses, he said.

Markovbegan and ended his presentationby calling himself an amateur and saying he doesnt have all the answers, but he also said he has been thinking about ways to prevent an AI takeover for more than a year.He now believes the most important way for humans to prevent the rise of malicious AIis to put in a series of physical world restraints.

The bottom line here is that intelligence either hostile or friendly would be limited by physical resources, and we need to think about physical resources if we want to limit such attacks, he said. We absolutely need to control access to energy sources of all kinds, and we need to be very careful about physical and network security of critical infrastructure because if that is not taken care of, then disasters can obviously happen.

Calling upon a background in hardware design, Markov suggested steps be taken to separate powerful systems and have deficiencies built in to act as a kill switch, because if superintelligent AI ever arises, it will likely be by accident.

He also implores thatlimits be placed on self repair, replication, or improvement of AI, and urges that specific scenarios be considered, such as a nuclear weapons attack or use of biological weapons.

Generally, each agent, each part of your AI ecosystem needs to be designed with some weakness. You dont want agents to be able to take over everything, right? So you would control agents through these weaknesses and separation of powers, he said. In the discipline of electronic hardware design, we use obstruction hierarchies. We go from transistors to CPUs to data centers, and each level typically has a well-defined function, so if youre looking at this from the perspective of security, if you are defending against something, you would want to limit or regulate every level, and you would want the same type of limitations for AI.

Markovs presentation relies on predictions made by Ray Kurzweil, who believes that in a decade, virtual reality will be indistinguishable from real life, after which computers will surpass humans. Then, through augmentation, humans will become more machine-like until we reach the Singularity.

Markov also pointed out that there is a range ofopinions on malicious AI. Stephen Hawking believes AI will eventually supersede humankind, telling the BBC, The development of full artificial intelligence could spell the end of the human race.

In contrast, former Baidu AI head Andrew Ng said last year that people should be as concerned about malicious AI as they are about overpopulation on Mars.

View original post here:

Watch Google's Igor Markov explain how to avoid the AI apocalypse - VentureBeat

While technology develops at exponential speed, transforming how we go about our everyday tasks and extending our lives, it also offers much to worry about. In particular, many top minds think that automation will cost humans their employment, with up to 47% of all jobs gone in the next 25 years. And chances are, this number could be even higher and the massive job loss will come earlier.

So when will your job become obsolete? Researchers at the University of Oxford surveyed the world's top artificial intelligence experts to find out when exactly machines will be better at humans in various occupations.

Katja Grace from Oxford's Future of Humanity Instituteled the team that gathered responses from 352 academics and industry experts in machine learning. Then they calculated the median responses to come up with some concrete numbers.

In the next 10 years, we should have AI do better than humans in translating languages (by 2024), writing high-school-level essays (by 2026), writing top 40 songs (by 2028) and driving trucks. And while the consensus may be that driving trucks may come by 2027, it's easy to predict that this could happen even sooner, with top tech entrepreneurs like Elon Musk constantly pushing the envelope and promising these innovations earlier.

A chore that would take less time - folding laundry should be a breeze for AI by 2022. Other tasks might take longer, but still within the foreseeable future. It's likely you'll be around for these. We should get AI-driven machines in retail by 2031. By 2049, AI should be writing New York Times bestsellers and performing surgeries by 2053.

Overall, AI shoud be better than humans at pretty much everything in about 45 years.

Here's the full chart:

As MIT Technology Review points out, these predictions have a way of coming earlier. AI wasn't supposed to beat humans at the game of Go until 2027 and that happened back in 2015. In fact, it took Google's DeepMind just two years to come up with the necessary tech, instead of the 12 that was predicted.

On the other hand, as 40 years is an average person's working life, predictions that extend past that might be unreliable as they are not based on technology in which the experts might not have enough practical knowledge.

Interestingly, there is a difference in how experts from different parts of the world view the future. Asian researchers put the AI takeover in just 30 years, while their North American counterparts see that happening in about 74 years. Full automation of labor is expected in under 125 years.

You can check out Grace's paper here.

Excerpt from:

Here's When Machines Will Take Your Job, as Predicted by AI Gurus - Big Think

Apple's big developer powwow, WWDC, kicks off Monday with a keynote that'll set the stage for software updates, developer platforms and the launch of the next iPhone. But the elephant in the room will be artificial intelligence and Apple's ability (or inability) to keep up with rivals.

The company got a jump on digital assistants and AI with the 2011 launch of Siri. Since that time though, Apple hasn't kept up with the AI pace as companies such as Alphabet (Google Assistant), Amazon (Alexa), Facebook (Messenger bots), Microsoft (Cortana) and IBM (Watson) have surged forward on the technology.

At the last two Google I/O conferences (2017 and 2016), executives couldn't stop talking about how they were sprinkling AI everywhere. Google is going for the AI-first company and CEO Sundar Pichai sees the transition as important as the one to mobile. See: Google I/O 2017: Here's what we learned

AWS and Amazon are using Alexa and AI in multiple areas.

Turns out everyone has an AI story--except for Apple. Consider that in Apple's latest quarterly and annual regulatory filings the terms machine learning and artificial intelligence weren't mentioned once.

Google's regulatory filings sound like this:

But here's the catch. Apple doesn't necessarily need to talk AI in abstract terms. Apple needs to productize AI and use it as a tool. Apple seems to get the message.

To wit:

So yes, Apple is likely to talk up its AI efforts and may position itself as a player, but there are multiple questions ahead.

Is Apple positioned for an AI-first technology world and how long does it have? Apple has plenty of cash and a product cycle that means it can put its own spin on AI. Apple doesn't have to chase AI leaders as much as utilize the technology.

Will AI be a standalone technology category or enabler? The best AI will run in the background and just do its thing. If you buy into that vision, then Apple doesn't have to take a Google-ish detour in an AI arms race.

Does Apple need its own AI technology and research? Yes. And no. Most companies will use multiple AI technologies just like they do cloud services. Couldn't Apple leverage the AI services from AWS, Google and Microsoft to create something new. You bet. I see the best Apple AI strategy as one that takes the same approach as SAP, which has stated it will use its own technology as well as allow customers to use other AI intellectual property.

How important is it that AI isn't cloud powered? Apple is likely to note that its AI will be device centric. Part of that reality is due to Apple's privacy pitch. Part of that pitch is because it doesn't have the AI, machine learning or cloud platform on the back-end.

We could go on, but rest assured that Apple's AI strategy and technology will remain a work in progress. Here's to hoping that WWDC sheds some light on what's around the corner.

ZDNet Monday Morning Opener

The Monday Morning Opener is our opening salvo for the week in tech. Since we run a global site, this editorial publishes on Monday at 8:00am AEST in Sydney, Australia, which is 6:00pm Eastern Time on Sunday in the US. It is written by a member of ZDNet's global editorial board, which is comprised of our lead editors across Asia, Australia, Europe, and the US.

Previously on Monday Morning Opener:

Read the original:

Apple's WWDC to-do list needs to include a dose of AI - ZDNet

Artificial intelligence technology is accelerating forward at a blistering pace, and a trio of scientists are calling for more accountability and transparency in AI, before it's too late.

In their paper, the UK-based researchers say existing rules and regulations don't go far enough in limiting what AI can do and recommend that robots be held to the same standards as the humans who make them.

There are a number of issues, say the researchers, that could lead to problems down the line, including the diverse nature of the systems being developed and a lack of transparency about the inner workings of AI.

"Systems can make unfair and discriminatory decisions, replicate or develop biases, and behave in inscrutable and unexpected ways in highly sensitive environments that put human interests and safety at risk," the team reportsin their paper.

In other words: how do we know we can trust AI?

Even before we get to the stage of the robots rising up, AI that's unaccountable and impossible to decipher is going to cause issues from problems working out the cause of an accident between self-driving cars, to understanding the reasons why a bank's computer has turned you down for a loan.

Another issue raised concerns systems that are mostly AI but have some human inputs, which may exclude them from regulation.

Airport security is one area where more transparency is needed, say researchers. Image: Wachter, Mittelstadt, Floridi

Among the suggestions put forward by the researchers is the idea of having a set of guidelines that covers robotics, AI, and decision-making algorithms as a group, though they admit that these diverse areas are hard to regulate as a whole.

The scientists also acknowledge that adding in extra transparency into AI systems can negatively affect their performance and competing tech companies might not be too willing to share their various secret sauces.

What's more, with AI now essentially teaching itself in some systems, we may even be beyond the stage where we can explain what's happening.

"The inscrutability and the diversity of AI complicate the legal codification of rights, which, if too broad or narrow, can inadvertently hamper innovation or provide little meaningful protection," the researchers write.

It's a delicate balancing act.

It's not the first time that these three researchers have criticised current AI rules: in January they called for an artificial intelligence watchdog in response to the General Data Protection Regulation drawn up by the EU.

"We are already too dependent on algorithms to give up the right to question their decisions," one of the researchers, Luciano Floridi from the University of Oxford in the UK, told Ian Sample at The Guardian.

Floridi and his colleagues quoted cases from Austria and Germany where they felt people hadn't been given enough information on how AI algorithms had reached their decisions.

Ultimately, AI is going to be a boon for the human race, whether it's helping elderly people keep their independence through self-driving cars, or spotting early signs of disease before doctors can.

Right now, though, experts are scrambling to put safety measures in place that can stop these systems getting out of control or becoming too dominant, and that's where the scientists behind this new paper want to see us putting our efforts.

"Concerns about fairness, transparency, interpretability, and accountability are equivalent, have the same genesis, and must be addressed together, regardless of the mix of hardware, software, and data involved," they write.

The paper has been published in Science Robotics.

View post:

Scientists want to set some ground rules to stop AI taking over the world - ScienceAlert

My first encounter with Siri was back in 2011, when Apples voice-activated artificial intelligence was first released.

I was completing university; the BlackBerry was still popular and I was one of a handful in my class with the new iPhone that had the robot assistant built in.

The technology was pretty basic at the time: it struggled with some accents and the commands it could respond to were fairly limited.

So early on Siri was something of a novelty, rather than genuinely useful. Six years later, BlackBerry is all but gone and the iPhone reigns, but what hasnt changed is Siris novelty feel.

Its voice recognition has improved dramatically (I suspect even the thickest regional accent wouldnt trip it up), and Apple has added lots of new features, but I doubt most iPhone users take advantage of it regularly.

Excerpt from:

As AI moves on, it's time for Apple's Siri to grow up - Telegraph.co.uk

Artificial intelligence is slowly proving that that video games aren't a total waste of time, at least for machines: It's through learning to play games that AI algorithms can acquire all sorts of generalizable skills, like problem-solving.

Now, computer scientists from RWTH Aachen University in Germany and Microsoft Research have released the largest-ever database of human playthroughs for some of the most popular games for the Atari 2600. Artificial agents using deep learning techniques will be able to pull patterns out of these playthroughs and learn from them.

Read More: Why Artificial Intelligence Researchers Love 'Super Mario Bros.'

According to a paper posted to the arXiv preprint server this week, which is undergoing peer review, the database contains more than 45 hours of gameplay from five games: Q*Bert, Ms. Pacman, Space Invaders, Video Pinball, and Montezuma's Revenge. Video games are an increasingly popular training ground for AI to solve general problems, like how to quickly arrive at a course of action, or how to effectively learn in an environment where the rewards for learning are sparse, which is notoriously the case for Montezuma's Revenge .

A massive database of human playthroughs could accelerate this kind of research, making researchers' lives easier and potentially producing more capable AI.

"When a person learns to play a game, you have lots of prior information about the world around you; but when an AI agent learns, it does it from scratch," said Vitaly Kurin, a Master's student at RWTH Aachen University and co-author of the paper, in an interview. "When an AI agent learns from human demonstration, it's like implicitly giving the bot all of the information we have about the world, and optimum strategies and behaviour. Everybody can use this dataset and check their ideas and models."

Kurin and his colleagues collected the database with a browser-based Atari 2600 emulator paired with an app that captures the player's input. They posted a link to the emulator to Reddit and asked people to help with the experiment by playing some video games, Kurin said. To make things faster and easier, the app only recorded the initial game state and every player input after that. The entire playthrough was then reconstructed offline using the player's recorded actions on the starting game state.

That might seem like a lot of work, but the payoff for research using the database could be huge. Machine learning using human demonstrations has already shown itself to be an impressive technique for quick learning. Researchers from the OpenAI institute recently unveiled an AI system that uses a single human demonstration in virtual reality to teach itself how to stack toy blocks.

"Let's imagine you have an autonomous car that you want to train to drive in the real world," Kurin said. "What if we gave the car a model that teaches it how to behave like a human? It suddenly becomes less dangerous because it won't do some crazy stuff after training completely from scratch." It's enough to make me glad for my peanut human brainwhen I play video games, it's just to relax. And that's fine with me.

Subscribe to Science Solved It , Motherboard's new show about the greatest mysteries that were solved by science.

See more here:

Scientists Made a Huge Library of Atari Gameplay to Give AI a Power Up - Motherboard

According to a survey of artificial intelligence experts, AI will probably be good enough to take on pretty much most of our jobs within half a century.

While there's plenty of room for debate on the details, the predicted applications of AI could serve as an alarm bell for us to consider how our economy and job market will adapt to ever smarter technology.

A team of researchers from the University of Oxford and Yale University received 352 responses to a survey they'd sent out to over 1,600 academics who had presented at conferences on machine learning and neural information processing in 2015.

The survey asked the experts to assign probabilities to dates in the future that AI might be capable of performing specific tasks, from folding laundry to translating languages.

They also asked for predictions on when machines would be superior to humans in fulfilling certain occupations, such as surgery or truck-driving; when they thought AI would be better than us at all tasks; and what they thought the social impacts could be.

The researchers then combined the results to determine a range of time stretching from a low 25 percent confidence to 75 percent certain, calculating a median point when most experts were hedging their bets.

You can check out the results in the table below.

Grace, Salvatier, Dafoe, Zhang, Evans

Most of the academics seem fairly confident that we'll have an AI be better than all humans at playing Angry Birds within the next seven years, and that we can start to place bets on AI winning World Series Poker within a decade.

We can bet there's a 50 percent chance robots will be better than us at folding laundry in about six years, followed very soon by an AI winning the strategy computer game StarCraft.

If you drive a truck for a living, there's a slim chance you'll be competing with automated drivers in just over five years, but you can be fairly sure there'll giving up the road to driverless trucks in just over 20 years.

There's a good chance we'll see a book written by an AI in the New York Times bestseller list in 26 years and a top 40 pop song in maybe 12 years.

And just in case you think you'll play it smart and develop the AI that is going to take over the world, the experts think there's a slim chance that machines will be the ones developing AI within half a century, and odds-on they'll be running the show in about 80 years.

The researchers set the 50 percent chance line for artificial intelligence being better at just about everything a point they describe as High Level Machine Intelligence in just under 50 years. At which point they think there's a likelihood that AI will be capable of doing just about any job you can imagine by 2140.

Just how much should we trust the experts on this one?

It might help to know that when they completed the survey back in 2015, on average they estimated there was only half a chance AI would beat the world champion at the game Go in about 12 years.

That's a skill we can now pretty much set to 100 percent accomplished, with the recent news of Google's DeepMind AlphaGo AI beating world champion Ke Jie in the first of a three-match series.

So it's possible the academics might simply be a little conservative in their estimates. In addition, experts in Asia saw AI progress occurring on average much sooner than those in North America, suggesting culture will affect our ability to predict.

While some jobs and skills are clearly on the horizon, it's a big call to say AI will probably be good at just about everything in 45 years.

An article at MIT Technology Review makes a compelling argument on why any prediction of "40 years in the future" should ring alarm bells: it is also the length of a single career, putting it outside of the span of any one person's current job.

In other words, we tend to optimistically imagine the technological advances required for some things occurring after our time.

The research is available on the pre-publish website arXiv.org, so hasn't yet been peer reviewed.

Even if the details are still up for discussion, we can be fairly confident that as time goes on, technology will at least be capable of doing a better job than your average human.

"While individual breakthroughs are unpredictable, longer term progress in R&D [research and development] for many domains (including computer hardware, genomics, solar energy) has been impressively regular," the researchers write.

As determined recently by a pair of US economists, we should expect significant impacts to particular fields in industry.

That isn't to say we can expect economics collapse and universal joblessness, either; going by history, technology creates more jobs than it destroys.

Give it a few decades and we'll be able to ask AI what they think we should do.

Original post:

Experts think this is how long we have before AI takes all of our jobs - ScienceAlert