<h2>Internal Safeguards</h2>
DeepMind, the artificial intelligence startup purchased by Google in 2014, is on a mission to build the world&#8217;s first <a href="https://futurism.com/why-build-human-level-ai-experts">artificial general intelligence</a> (AGI) — the sort of all-encompassing, superhuman AI we see in science fiction.
To make sure that AGI is used responsibly, DeepMind founder Demis Hassabis protected his company&#8217;s independence from Google and parent company Alphabet by building internal safeguards, according to a <a href="https://www.1843magazine.com/features/deepmind-and-google-the-battle-to-control-artificial-intelligence">new story</a> by The Economist&#8216;s cultural publication, 1843 Magazine — including an ethics board controlled by Hassabis and his original team, rather than Google.
<h2>Power Grabs</h2>
Though Google has long <a href="https://futurism.com/google-dont-be-evil">branded itself as an ethical leader</a> within the world of AI, 1843&#8216;s report details a history of power grabs and questionable management.
For instance, DeepMind started a new healthcare division in 2016 that would use AI technology to streamline hospital operations and monitor patients&#8217; vitals.
Just two years later, months after DeepMind Health had closed contracts with partner hospitals, Google <a href="https://futurism.com/google-health-deepmind-streams-app">announced Google Health</a> and absorbed DeepMind&#8217;s program into it, giving little notice to DeepMind or the hospitals. Sources close to the matter, granted anonymity by 1843 because they signed non-disclosure agreements, shared that the move led to resentment and anger among DeepMind&#8217;s team.
<h2>The Big &#8220;If&#8221;</h2>
Some of those same anonymous sources also shared doubts with 1843 that DeepMind would ever crack the code of artificial general intelligence, pointing to little-known weaknesses and caveats to the company&#8217;s <a href="https://futurism.com/google-deepmind-ai-beat-starcraft-ii-players">highly-publicized AI success stories</a>.
The algorithms developed at DeepMind can do fascinating things, like <a href="https://futurism.com/google-deepmind-ai-finds-way-doom-like-3d-maze-sight">demolish certain video games</a> and <a href="https://futurism.com/artificial-intelligence-remember-agi">retain knowledge better than other systems</a>, but there&#8217;s no clear roadmap to the fundamentally-new technology necessary for developing AGI.
But should Hassabis ever figure it out, the safeguards and ethical review boards that he put in place when his company was bought could mean that the technology will remain in the hands of him and his team instead of becoming just another Google technology.
READ MORE: <a href="https://www.1843magazine.com/features/deepmind-and-google-the-battle-to-control-artificial-intelligence">DeepMind and Google: the battle to control artificial intelligence</a> [1843 Magazine]
More on artificial intelligence: <a href="https://futurism.com/artificial-intelligence-hype">You Have No Idea What Artificial Intelligence Really Does</a>

The ethics review board in charge of evaluating any future AGI is controlled by DeepMind staff.

Google's DeepMind is grappling with how to control superhuman AI

Bureaucratic Takeover

Google has a history of absorbing DeepMind's AI technology into its own projects. To protect the goldmine of AGI, DeepMind has safeguards in place.

There’s a Power Struggle Inside Google to Control Superhuman AI

If DeepMind builds a superhuman AI, who will control it?

Working on artificial intelligence can be a vicious cycle. It&#8217;s easy to lose track of the bigger picture when you spend an entire career developing a niche, hyper-specific AI application. An engineer might finally step away and realize that the public never actually needed such a robust system; each of the marginal improvements they&#8217;ve spent so much time on didn&#8217;t mean much in the real world.
Still, we need these engineers with lofty, yet-unattainable goals. And one specific goal still <a href="https://futurism.com/human-level-artificial-intelligence-agi">lingers in the horizon</a> for the more starry-eyed computer scientists out there: building a human-level artificial intelligence system that could change the world.
<a href="https://futurism.com/artificial-intelligence-hype">Coming up with a definition</a> of human-level AI (HLAI) is tough because so many people use it interchangeably with artificial general intelligence (AGI) – which is the thoughtful, emotional, creative sort of AI that exists only in movie characters like C-3PO and “Ex Machina’s” Ava.
Human-level AI is similar, but not quite as powerful as AGI, for the simple reason that many in the know expect AGI to surpass anything we mortals can accomplish. Though some see this as an argument against building HLAI, some experts believe that only an HLAI could ever be clever enough to design a true AGI – human engineers would only be necessary up to a certain point once we get the ball rolling. (Again, neither type of AI system exists nor will they anytime soon.)
At a <a href="https://www.hlai-conf.org/">conference on HLAI</a> held by Prague-based AI startup GoodAI in August, a number of <a href="https://futurism.com/should-evil-ai-research-be-published-five-experts-weigh-in">AI experts and thought leaders</a> were asked a simple question: “Why should we bother trying to create human-level AI?”
For those AI researchers that have detached from the outside world and gotten stuck in their own little loops (yes, of course we care about your AI-driven digital marketplace for farm supplies), the responses may remind them why they got into this line of work in the first place. For the rest of us, they provide a glimpse of <a href="https://futurism.com/advanced-artificial-intelligence-better-humans">the great things to come</a>.
For what it’s worth, this particular panel was more of a lightning round — largely for fun, the experts were instructed to come up with a quick answer rather than taking time to deliberate and carefully choose their words.
<q class="pullquote">&#8220;Why should we bother trying to create human-level AI?”</q>
<a href="http://goertzel.org/">Ben Goertzel</a>, CEO at SingularityNET and Chief Scientist at Hanson Robotics
<blockquote>AI is a great intellectual challenge and it also has more potential to do good than any other invention. Except superhuman AI which has even more.</blockquote>
<a href="https://research.fb.com/people/mikolov-tomas/">Tomas Mikolov</a>, Research Scientist At Facebook AI
<blockquote>[Human-Level AI will give us] ways to make life more efficient and basically guide [humanity.]</blockquote>
<a href="https://www.hlai-conf.org/stanley">Kenneth Stanley</a>, Professor At University Of Central Florida, Senior Engineering Manager And Staff Scientist At Uber AI Labs
<blockquote>I think we’d like to understand ourselves better and how to make our lives better.</blockquote>
<a href="https://www.researchgate.net/profile/Pavel_Kordik">Pavel Kordik</a>, Associate Professor at Czech Technical University and Co-founder at Recombee
<blockquote>To create a singularity, perhaps.</blockquote>
<a href="https://www.araya.org/archives/418">Ryota Kanai</a>, CEO at ARAYA
<blockquote>To understand ourselves.</blockquote>
More on the future of AI: <a href="https://futurism.com/artificial-intelligence-experts-fear">Five Experts Share What Scares Them the Most About AI</a>

Some think only human-level AI could make other AI that's smarter than humans.

Why should we bother building human-level AI? Five experts weigh in

CLEAR AIs, FULL HEARTS

Whether it's to manage the world better than we do or give us new insight into how our own minds work, many experts agree we should make a human-level AI.

Why Should We Bother Building Human-Level AI? Five Experts Weigh In

Machines as smart as people could build a much better future.

WHEN SOPHIA THE ROBOT first switched on, the world couldn&#8217;t get enough. It had a cheery personality, it joked with late-night hosts, it had facial expressions that echoed our own. Here it was, finally — a robot plucked straight out of science fiction, the closest thing to true artificial intelligence that we had ever seen.
There&#8217;s no doubt that Sophia is an impressive piece of engineering. Parents-slash-collaborating-tech-companies Hanson Robotics and SingularityNET equipped Sophia with sophisticated neural networks that give Sophia the ability to learn from people and to <a href="https://www.theverge.com/2017/11/10/16617092/sophia-the-robot-citizen-ai-hanson-robotics-ben-goertzel">detect and mirror emotional responses</a>, which makes it seem like the robot has a personality. It didn&#8217;t take much to convince people of Sophia&#8217;s apparent humanity — <a href="https://futurism.com/sophiabot-asks-you-to-be-nice-so-she-can-learn-human-compassion/">many</a> of <a href="https://futurism.com/videos/sophia-robot-says-will-destroy-humankind/">Futurism’s own articles</a> <a href="https://futurism.com/saudi-arabia-made-robot-citizen-calling-womens-rights/">refer to the robot</a> as “her.” Piers Morgan even decided to <a href="https://www.inverse.com/article/33229-piers-morgan-sexually-harasses-a-robot">try his luck</a> for a date and/or sexually harass the robot, depending on how you want to look at it.
“Oh yeah, she is basically alive,&#8221; Hanson Robotics CEO David Hanson said of Sophia during a <a href="https://www.youtube.com/watch?v=Bg_tJvCA8zw">2017 appearance on Jimmy Fallon’s Tonight Show</a>. And while Hanson Robotics never officially claimed that Sophia contained artificial general intelligence — the comprehensive, life-like AI that we see in science fiction — the adoring and uncritical press that followed all those public appearances only helped the company grow.
But as Sophia became more popular and people took a closer look, <a href="https://www.adweek.com/digital/how-sophia-the-robot-is-used-to-market-the-future-of-artificial-intelligence/">cracks emerged</a>. It became harder to believe that Sophia was the all-encompassing artificial intelligence that we all wanted it to be. Over time, articles that might have once oohed and ahhed about Sophia&#8217;s conversational skills became more focused on the fact that they were partially scripted in advance.
<iframe width="500" height="281" src="https://www.youtube.com/embed/S5t6K9iwcdw?feature=oembed" frameborder="0" allow="autoplay; encrypted-media" allowfullscreen></iframe>
<a href="http://goertzel.org/">Ben Goertzel</a>, CEO of SingularityNET and Chief Scientist of Hanson Robotics, isn&#8217;t under any illusions about what Sophia is capable of. &#8220;Sophia and the other Hanson robots are not really &#8216;pure&#8217; as computer science research systems, because they combine so many different pieces and aspects in complex ways. They are not pure learning systems, but they do involve learning on various levels (learning in their neural net visual systems, learning in their OpenCog dialogue systems, etc.),&#8221; he told Futurism.
But he&#8217;s interested to find that Sophia inspires a lot of different reactions from the public. &#8220;Public perception of Sophia in her various aspects — her intelligence, her appearance, her lovability — seems to be all over the map, and I find this quite fascinating,&#8221; Goertzel said.
Hanson finds it unfortunate when people think Sophia is capable of more or less than she really is, but also said that he doesn’t mind the benefits of the added hype. Hype which, again, has been bolstered by the two companies’ repeated publicity stunts.
<q class="pullquote">&#8220;Sophia and the other Hanson robots are not really &#8216;pure&#8217; as computer science research systems&#8230;&#8221;</q>
Highly-publicized projects like Sophia convince us that true AI — human-like and perhaps even conscious — is right around the corner. But in reality, we&#8217;re not even close.
The true state of AI research has fallen far behind the technological fairy tales we&#8217;ve been led to believe. And if we don&#8217;t treat AI with a healthier dose of realism and skepticism, the field may be stuck in this rut forever.
NAILING DOWN A TRUE definition of artificial intelligence is tricky. The field of AI, constantly reshaped by new developments and changing goalposts, is sometimes best described by explaining what it is not.
&#8220;People think AI is a smart robot that can do things a very smart person would — a robot that knows everything and can answer any question,&#8221; <a href="https://www.linkedin.com/in/miremadmousavi/detail/background-image/">Emad Mousavi</a>, a data scientist who founded a platform called QuiGig that connects freelancers, told Futurism. But this is not what experts really mean when they talk about AI. &#8220;In general, AI refers to computer programs that can complete various analyses and use some predefined criteria to make decisions.&#8221;
Among the ever-distant goalposts for human-level artificial intelligence (HLAI) are the ability to communicate effectively — chatbots and machine learning-based language processors <a href="https://www.livescience.com/61818-ai-fueled-robot-learns-how-to-love-on-paper-at-least.html">struggle to infer meaning</a> or to understand nuance — and the ability to continue learning over time. Currently, the AI systems with which we interact, including those <a href="https://theoutline.com/post/5964/the-self-driving-car-that-will-never-arrive?zd=6&amp;zi=5efcby5f">being developed for self-driving cars</a>, do all their learning before they are deployed and then stop forever.
“They are problems that are easy to describe but are unsolvable for the current state of machine learning techniques,” <a href="https://research.fb.com/people/mikolov-tomas/">Tomas Mikolov</a>, a research scientist at Facebook AI, told Futurism.
Right now, AI doesn&#8217;t have free will and certainly isn&#8217;t conscious — two assumptions people tend to make when faced with advanced or over-hyped technologies, Mousavi said. The most advanced AI systems out there are merely products that follow processes defined by smart people. They can&#8217;t make decisions on their own.
In machine learning, which includes deep learning and neural networks, an algorithm is presented with boatloads of training data — examples of whatever it is that the algorithm is learning to do, labeled by people — until it can complete the task on its own. For facial recognition software, this means feeding thousands of photos or videos of faces into the system until it can reliably detect a face from an unlabeled sample.
Our best machine learning algorithms are generally just memorizing and running statistical models. To call it &#8220;learning&#8221; is to anthropomorphize machines that operate on a very different wavelength from our brains. Artificial intelligence is now such a big catch-all term that practically any computer program that automatically does something is referred to as AI.
<q class="pullquote">Artificial intelligence is now such a big catch-all term that practically any computer program that automatically does something is referred to as AI.</q>
If you train an algorithm to add two numbers, it will just look up or copy the correct answer from a table, Mikolov, the Facebook AI scientist, explained. But it can&#8217;t generalize a better understanding of mathematical operations from its training. After learning that five plus two equals seven, you as a person might be able to figure out that seven minus two equals five. But if you ask your algorithm to subtract two numbers after teaching it to add, it won’t be able to. The artificial intelligence, as it were, was trained to add, not to understand what it means to add. If you want it to subtract, you’ll need to train it all over again — a process that notoriously <a href="https://futurism.com/artificial-intelligence-remember-agi/">wipes out whatever the AI system had previously learned</a>.
“It’s actually often the case that it’s easier to start learning from scratch than trying to retrain the previous model,” Mikolov said.
These flaws are no secret to members of the AI community. Yet, all the same, these machine learning systems are often touted as the cutting edge of artificial intelligence. In truth, they&#8217;re actually quite dumb.
Take, for example, an image captioning algorithm. A few years back, one of these got some <a href="https://www.nytimes.com/2014/11/18/science/researchers-announce-breakthrough-in-content-recognition-software.html">wide-eyed coverage</a> because of the sophisticated language it seemed to generate.
“Everyone was very impressed by the ability of the system, and soon it was found that 90 percent of these captions were actually found in the training data,” Mikolov told Futurism. “So they were not actually produced by the machine; the machine just copied what it did see that the human annotators provided for a similar image so it seemed to have a lot of interesting complexity.&#8221; What people mistook for a robotic sense of humor, Mikolov added, was just a dumb computer hitting copy and paste.
“It’s not some machine intelligence that you’re communicating with. It can be a useful system on its own, but it’s not AI,” said Mikolov. He said that it took a while for people to realize the problems with the algorithm. At first, they were nothing but impressed.
<figure id="attachment_139378" style="width: 1200px" class="wp-caption alignnone"><img class="wp-image-139378 size-large" src="https://wordpress.futurism.com/wp-content/uploads/2018/10/artificial-intelligence-hype-deception2-1200x630.png" alt="" width="1200" height="630" /><figcaption class="wp-caption-text">Image Credit: Victor Tangermann</figcaption></figure>
WHERE DID WE GO so off course? The problem is when our present-day systems, which are so limited, are <a href="https://futurism.com/artificial-intelligence-experts-fear/">marketed and hyped up</a> to the point that the public believes we have technology that we have no goddamn clue how to build.
&#8220;I am frequently entertained to see the way my research takes on exaggerated proportions as it progresses through the media,&#8221; <a href="https://cs.byu.edu/parttime_faculty/info/50504">Nancy Fulda</a>, a computer scientist working on broader AI systems at Brigham Young University, told Futurism. The reporters who interview her are usually pretty knowledgeable, she said. &#8220;But there are also websites that pick up those primary stories and report on the technology without a solid understanding of how it works. The whole thing is a bit like a game of &#8216;telephone&#8217; — the technical details of the project get lost and the system begins to seem self-willed and almost magical. At some point, I almost don&#8217;t recognize my own research anymore.&#8221;
<q class="pullquote">&#8220;At some point, I almost don&#8217;t recognize my own research anymore.&#8221;</q>
Some researchers themselves are guilty of fanning this flame. And then the reporters who don&#8217;t have much technical expertise and don&#8217;t look behind the curtain are complicit. Even worse, some journalists are happy to play along and add hype to their coverage.
Other problem actors: people who make an AI algorithm present the back-end work they did as that algorithm’s own creative output. Mikolov calls this a dishonest practice akin to sleight of hand. “I think it’s quite misleading that some researchers who are very well aware of these limitations are trying to convince the public that their work is AI,” Mikolov said.
That&#8217;s important because the way people think AI research is going will depend on whether they want money allocated to it. This unwarranted hype could be preventing the field from making real, useful progress. <a href="https://futurism.com/artificial-general-intelligence-governments-invest/">Financial investments in artificial intelligence</a> are inexorably linked to the level of interest (read: hype) in the field. That interest level — and corresponding investments — fluctuate wildly whenever Sophia has a stilted conversation or some new machine learning algorithm accomplishes something mildly interesting. That makes it hard to establish a steady, baseline flow of capital that researchers can depend on, Mikolov suggested.
Mikolov hopes to one day create a genuinely intelligent AI assistant — a goal that he told Futurism is still a distant pipedream. A few years ago, Mikolov, along with his colleagues at Facebook AI, <a href="https://arxiv.org/pdf/1511.08130.pdf">published a paper</a> outlining how this might be possible and the steps it might take to get there. But when we spoke at the <a href="https://www.hlai-conf.org/">Joint Multi-Conference on Human-Level Artificial Intelligence</a> held in August by Prague-based AI startup GoodAI, Mikolov mentioned that many of the avenues people are exploring to create something like this are likely dead ends.
One of these likely dead ends, unfortunately, is reinforcement learning. Reinforcement learning systems, which teach themselves to complete a task through trial and error-based experimentation instead of using training data (think of a dog fetching a stick for treats), are often oversold, according to <a href="https://www.microsoft.com/en-us/research/people/jcl/">John Langford</a>, Principal Researcher for Microsoft AI. Almost anytime someone brags about a reinforcement-learning AI system, Langford said, they actually gave the algorithm some shortcuts or limited the scope of the problem it was supposed to solve in the first place.
The hype that comes from these sorts of algorithms helps the researcher sell their work and secure grants. Press people and journalists use it to draw audiences to their platforms. But the public suffers — this vicious cycle leaves everyone else unaware as to what AI can really do.
There are telltale signs, Mikolov says, that can help you see through the misdirection. The biggest red flag is whether or not you as a layperson (and potential customer) are allowed to demo the technology for yourself.
“A magician will ask someone from the public to test that the setup is correct, but the person specifically selected by the magician is working with him. So if somebody shows you the system, then there’s a good likelihood you are just being fooled,” Mikolov said. “If you are knowledgeable about the usual tricks, it’s easy to break all these so-called intelligent systems. If you are at least a little bit critical, you will see that what [supposedly AI-driven chatbots] are saying is very easy to distinguish from humans.”
Mikolov suggests that you should question the intelligence of anyone trying to sell you the idea that they’ve beaten the Turing Test and created a chatbot that can hold a real conversation. Again, think of Sophia&#8217;s prepared dialogue for a given event.
&#8220;Maybe I should not be so critical here, but I just can’t help myself when you have these things like the Sophia thing and so on, where they’re trying to make impressions that they are communicating with the robot at so on,” Mikolov told Futurism.&#8221;Unfortunately, it&#8217;s quite easy for people to fall for these magician tricks and fall for the illusion, unless you&#8217;re a machine learning researcher who knows these tricks and knows what&#8217;s behind them.&#8221;
Unfortunately, so much attention to these misleading projects can stand in the way of progress by people with truly original, revolutionary ideas. It&#8217;s hard to get funding to build something brand new, something that might lead to AI that can do what people already expect it to be able to do, when venture capitalists just want to <a href="https://futurism.com/artificial-general-intelligence-governments-invest">fund the next machine learning solution</a>.
If we want those projects to flourish, if we ever want to take tangible steps towards artificial general intelligence, the field will need to be a lot more transparent about what it does and how much it matters.
“I am hopeful that there will be some super smart people who come with some new ideas and will not just copy what is being done,” said Mikolov. “Nowadays it’s some small, incremental improvement. But there will be smart people coming with new ideas that will bring the field forward.”
More on the nebulous challenges of AI: <a href="https://futurism.com/artificial-consciousness/">Artificial Consciousness: How To Give A Robot A Soul</a>

The day of the truly intelligent AI system has not yet arrived, no matter what anyone says.

You have no idea what artificial intelligence really does

Artificial Stupidity

It's easy to fall for all of the marketing and hype, but artificial intelligence isn't half as robust as many want you to believe.

You Have No Idea What Artificial Intelligence Really Does

The world of AI is full of hype and deception.

“Star Wars,” “Her,” and “iRobot.” What do all these movies have in common? The artificial intelligence (AI) depicted in there is crazy-sophisticated. These robots can think creatively, <a href="https://futurism.com/artificial-intelligence-remember-agi/">continue learning over time</a>, and maybe even pass for conscious.
Real-life artificial intelligence experts have a name for AI that can do this — it&#8217;s Artificial General Intelligence (AGI). For decades, scientists have tried all sorts of approaches to create AGI, using techniques such as reinforcement learning and machine learning. No approach has proven to be much better than any other, at least not yet.
Indeed, there&#8217;s a catch here: despite all the excitement, <a href="https://futurism.com/darpa-ai-mimic-human-brain/">we have no idea</a> how to build AGI.
At <a href="https://www.hlai-conf.org/">The Joint Multi-Conference on Human-Level Artificial Intelligence</a> held last month in Prague, AI experts and thought leaders from around the world <a href="https://futurism.com/advanced-artificial-intelligence-better-humans/">shared their goals</a>, hopes, and progress towards human-level AI (HLAI), which is the last stop before true AGI or the same thing, depending on who you ask.
Either way, most experts think it&#8217;s coming — sooner rather than later. In a poll of conference attendees, AI research companies GoodAI and SingularityNet found that <a href="https://docs.google.com/document/d/1jvfZId7Z41hTnW6shiZZAmur_ZqyHuoSKnWQyypgOyY/edit?usp=sharing">37 percent of respondents</a> think people will create HLAI within 10 years. Another 28 percent think it will take 20 years. Just two percent think HLAI will never exist.
Almost every expert who had an opinion hedged their bets — most responses to the question were peppered with caveats and “maybes.” There’s a lot that we don’t know about the path towards HLAI, such as questions over <a href="https://futurism.com/artificial-general-intelligence-governments-invest/">who will pay for it</a> or how we’re going to combine our algorithms that can think or reason but can’t do both.
Futurism caught up with a number of AI researchers, investors, and policymakers to get their perspective on when HLAI will happen. The following responses come from panels and presentations from the conference and exclusive interviews.
<h3><a href="https://www.itu.int/en/fnc/Pages/bios/BERIDZEIrakli.aspx">Irakli Beridze</a>, Head of the Centre for Artificial Intelligence and Robotics, UNICRI, United Nations</h3>
At the moment, there is absolutely no indication that we are anywhere near AGI. And no one can say with any kind of authority or conviction that this would happen within a certain time frame. Or even worse, no one can say this can even happen period. We may never have AGI, so we need to take that into account when we are discussing anything.
<h3><a href="http://www.crassh.cam.ac.uk/people/profile/sean-o-heigeartaigh">Seán Ó hÉigeartaigh</a>, Executive Director of the Cambridge Center for the Study of Existential Risk</h3>
There’s still a lot of work to be done; there are still many things we don’t understand. Given we have this understanding, maybe it’s possible that it happens within 50 years.
<h3><a href="https://www.microsoft.com/en-us/research/people/jcl/">John Langford</a>, Principal Researcher at Microsoft AI</h3>
I think we should enjoy the technology while it advances. We should be looking out for where to go in the future. But on the other hand, it’s not like we have human-level AI right now and I don’t think it’s going to happen very quickly. I think that if I’m lucky it’ll happen in my lifetime.
A worm’s level of intelligence is actually pretty doable. If you try to look at vision and planning, this is kind of narrowly doable. The integration of planning and learning, planning as its own thing is pretty well solved. But planning in a way which works with [machine learning] is not very well solved.
<h3><a href="https://blog.marekrosa.org/">Marek Rosa</a>, CEO and CTO at GoodAI</h3>
I think we are almost there. I am not predicting we will have general AI in three years, 30 years. But I am confident it can happen any day.
<h3><a href="http://goertzel.org/">Ben Goertzel</a>, CEO at SingularityNET and Chief Scientist at Hanson Robotics</h3>
I don’t think we need fundamentally new algorithms. I think we do need to connect our algorithms in different ways than we do now. If I’m right, then we already have the core algorithms that we need… I believe we are less than ten years from creating human-level AI.
<h3><a href="https://www.darpa.mil/staff/dr-hava-siegelmann">Hava Siegelmann</a>, Program Manager at DARPA</h3>
I don’t think we’re almost there in the technology for General AI. I think general AI is almost a branding for a very general idea. Lifelong learning is an example of that — it’s a very particular type of AI. We know the theoretical foundation of that already, we know how nature does it, and it’s very well defined. There is a very clear direction, there is a metric. I think we can reach it in a close time.
On the last day of the conference, a number of participants participated in a lightning round of sorts. Almost entirely for fun, these experts were encouraged to throw out a date at which they expected us to figure out how to make HLAI. The following answers, some of whom were given by the same people who already answered the question, should be taken with an entire shaker of salt — some were meant as jokes and others are total guesses.
<h3>John Langford</h3>
Maybe 20 [years]?
<h3>Marek Rosa</h3>
I really have no idea which year, but if I have to say one year I’d say ten years in the future. The reason is its kind fo vague, you know like anything can happen in ten years.
<h3><a href="https://www.araya.org/archives/418">Ryota Kanai</a>, CEO at ARAYA</h3>
2030.
<h3><a href="https://www.researchgate.net/profile/Pavel_Kordik">Pavel Kordik</a>, Associate Professor at Czech Technical University and Co-founder at Recombee</h3>
I think it’s very hard to define human-level AI, so it might come in five years or later.
<h3><a href="https://www.cs.ucf.edu/~kstanley/">Kenneth Stanley</a>, Professor at University of Central Florida and Senior Research Scientist at Uber AI Labs</h3>
I’m between 20 and 2,000 years.
<h3>Tomas Mikolov </h3>
Based on the recent progress, I would guess it will come in 50 years.
<h3>Ben Goertzel</h3>
It will occur on December 8, 2026, which will be my 60th birthday. I will delay it until then just to have a great birthday party.
More On expert predictions of AI: <a href="https://futurism.com/artificial-intelligence-experts-fear/">Five Experts Share What Scares Them the Most About AI</a>

Those who think we will have human-level AI predict it'll arrive anywhere from five to 2,000 years from now.

When will we have artificial intelligence as smart as a human? Here’s what experts think

Artificial Intelligence

Human-level artificial intelligence (HLAI) is a holy grail of AI research. But no one knows how to make it. Here's when the experts think it'll happen.

When Will We Have Artificial Intelligence As Smart as a Human? Here’s What Experts Think

Robots in the movies can think creatively, continue learning over time, and maybe even pass for conscious. Why don't we have that yet?

Sophisticated AI could make the world a better place. It might let us fight cancer and improve healthcare around the world, or simply free us from the menial tasks that dominate our lives.
That was the primary topic of conversation last month when engineers, investors, researchers, and policymakers got together at <a href="https://www.hlai-conf.org/">The Joint Multi-Conference on Human-Level Artificial Intelligence</a>.
But there was an undercurrent of fear that ran through some of the talks, too. Some people are anxious about losing their jobs to a robot or line of code; others fear a robot uprising. Where&#8217;s the line between fearmongering and legitimate concern?
In an effort to separate the two, Futurism asked five AI experts at the conference about what they fear most about a future with advanced artificial intelligence. Their responses, below, have been lightly edited.
Hopefully, with their concerns in mind, we&#8217;ll be able to steer society in a better direction — one in which we use AI for all the good stuff, like fighting global epidemics or granting more people an education, and less of the bad stuff.
<figure id="attachment_136825" style="width: 1200px" class="wp-caption alignnone"><img class="wp-image-136825 size-large" src="https://wp-assets.futurism.com/2018/09/artificial-intelligence-experts-fear2-1200x681.png" alt="" width="1200" height="681" /><figcaption class="wp-caption-text">Image Credit: <a href="https://www.gettyimages.ca/license/1017194038">Getty Images</a></figcaption></figure>
<q class="pullquote">Q: When you think of what we can do — and what we will be able to do — with AI, what do you find the most unsettling?</q>
<h2><a href="https://www.hlai-conf.org/stanley">Kenneth Stanley</a>, Professor at University of Central Florida, Senior Engineering Manager and Staff Scientist at Uber AI Labs</h2>
<blockquote>I think that the most obvious concern is when AI is used to hurt people. There are a lot of different applications where you can imagine that happening. We have to be really careful about letting that bad side get out. [Sorting out how to keep AI responsible is] a very tricky question; it has many more dimensions than just the scientific. That means all of society does need to be involved in answering it.</blockquote>
On how to develop safe AI:
<blockquote>All technology can be used for bad, and I think AI is just another example of that. Humans have always struggled with not letting new technologies be used for nefarious purposes. I believe we can do this: we can put the right checks and balances in place to be safer.
I don’t think I know what exactly we should do about it, but I can caution us to take [our response to the impacts of AI] very carefully and gradually and to learn as we go.</blockquote>
<h2><a href="https://www.itu.int/en/fnc/Pages/bios/BERIDZEIrakli.aspx">Irakli Beridze</a>, Head of the Centre for Artificial Intelligence and Robotics at UNICRI, United Nations</h2>
<blockquote>I think the most dangerous thing with AI is its pace of development. Depending how quickly it will develop and how quickly we will be able to adapt to it. And if we lose that balance, we might get in trouble.</blockquote>
On terrorism, crime, and other sources of risk:
<blockquote>I think the dangerous applications for AI, from my point of view, would be criminals or large terrorist organizations using it to disrupt large processes or simply do pure harm. [Terrorists could cause harm] via digital warfare, or it could be a combination of robotics, drones, with AI and other things as well that could be really dangerous.
And, of course, other risks come from things like job losses. If we have massive numbers of people losing jobs and don&#8217;t find a solution, it will be extremely dangerous. Things like lethal autonomous weapons systems should be properly governed — otherwise there’s massive potential of misuse.</blockquote>
On how to move forward:
<blockquote>But this is the duality of this technology. Certainly, my conviction is that AI is not a weapon; AI is a tool. It is a powerful tool, and this powerful tool could be used for good or bad things. Our mission is to make sure that this is used for the good things, the most benefits are extracted from it, and most risks are understood and mitigated.</blockquote>
<h2><a href="https://www.microsoft.com/en-us/research/people/jcl/">John Langford</a>, Principal Researcher at Microsoft</h2>
<blockquote>I think we should watch out for drones. I think automated drones are potentially dangerous in a lot of ways.The computation on board unmanned weapons isn&#8217;t efficient enough to do something useful right now. But in five or ten years, I can imagine that a drone could have onboard computation sufficient enough that it could actually be useful. You can see that drones are already getting used in warfare, but they’re [still human-controlled]. There’s no reason why they couldn’t be carrying some kind of learning system and be reasonably effective. So that’s something that I worry about a fair bit.</blockquote>
<h2><a href="https://www.darpa.mil/staff/dr-hava-siegelmann">Hava Siegelmann</a>, Microsystems Technology Office Programs Manager at DARPA</h2>
<blockquote>Every technology can be used for bad. I think it&#8217;s in the hands of the ones that use it. I don’t think there is a bad technology, but there will be bad people. It comes down to who has access to the technology and how we use it.</blockquote>
<h2><a href="https://research.fb.com/people/mikolov-tomas/">Tomas Mikolov</a>, Research Scientist at Facebook AI</h2>
<blockquote>When there’s a lot of interest and funding around something, there are also people who are abusing it. I find it unsettling that some people are selling AI even before we make it, and are pretending to know what [problem it will solve].
These strange startups are also promising things that are some great AI examples when their systems are basically over-optimizing a single path that maybe anyone didn&#8217;t even care about before [such as a chatbot that&#8217;s just a little better than the last version]. And maybe after spending tens of thousands of hours of work, by over-optimizing a single value, some of these startups come in with these big claims that they did achieve something that nobody could previously do.
But come on, let&#8217;s be honest, many of the recent breakthroughs of these groups that I don&#8217;t want to name, nobody cared before, and they are not generating any money. They are more like magician tricks. Especially ones that see AI as just over-optimizing a single task that is very narrow and there&#8217;s no way they can scale to pretty much anything else other than very simple problems.
Someone who’s even a little bit critical of these systems would quickly encounter problems that go against the company&#8217;s lofty claims.</blockquote>
More on developing safe AI: <a href="https://futurism.com/should-evil-ai-research-be-published-five-experts-weigh-in/">Should Evil AI Research Be Published? Five Experts Weigh In.</a>

Don't worry, killer robots only came up twice.

Five experts share what scares them the most about AI

Artificial intelligence will change life for the better. But there are reasons to be cautious as well. Five experts tell Futurism what most worries them.

Five Experts Share What Scares Them the Most About AI

A future governed by AI is promising. But it's also scary.

When you return to school after summer break, it may feel like you forgot everything you learned the year before. But if you learned like an AI system does, you actually would have — as you sat down for your first day of class, your brain would take that as a cue to wipe the slate clean and start from scratch.
AI systems&#8217; tendency to forget the things it previously learned upon taking on new information is called <a href="https://futurism.com/a-new-computing-technology-mimics-aspects-of-human-thought/">catastrophic forgetting</a>.
That&#8217;s a big problem. See, cutting-edge algorithms learn, so to speak, after analyzing countless examples of what they&#8217;re expected to do. A facial recognition AI system, for instance, will analyze thousands of photos of people&#8217;s faces, likely photos that have been manually annotated, so that it will be able to detect a face when it pops up in a video feed. But because these AI systems don&#8217;t actually comprehend the underlying logic of what they do, teaching them to do anything else, even if it&#8217;s pretty similar — like, say, recognizing specific emotions — means training them all over again from scratch. Once an algorithm is trained, it&#8217;s done, we can&#8217;t update it anymore.
For years, scientists <a href="https://arxiv.org/abs/1612.00796">have been trying to figure out how to work around</a> the problem. If they succeed, AI systems would be able to learn from a new set of training data without overwriting most of what they already knew in the process. Basically, if the robots should someday rise up, our new overlords would be able to conquer all life on Earth and chew bubblegum at the same time.
But still, catastrophic forgetting is one of the major hurdles preventing scientists from building an artificial general intelligence (AGI) — AI that&#8217;s all-encompassing, empathetic, and imaginative, like the ones <a href="https://futurism.com/artificial-consciousness/">we see in TV and movies</a>.
In fact, a number of AI experts who attended <a href="https://www.hlai-conf.org/">The Joint Multi-Conference on Human-Level Artificial Intelligence</a> last week in Prague said, in private interviews with Futurism or during panels and presentations, that the problem of catastrophic forgetting is one of the top reasons they don’t expect to see AGI or human-level AI anytime soon.
<q class="pullquote">Catastrophic forgetting is one of the top reasons experts don’t expect to see  human-level AI anytime soon.</q>
But <a href="https://www.hlai-conf.org/higgins">Irina Higgins</a>, a senior research scientist at Google DeepMind, used her presentation during the conference to announce that her team had begun to crack the code.
She had developed an AI agent — sort of like a video game character controlled by an AI algorithm — that could think more creatively than a typical algorithm. It could &#8220;imagine&#8221; what the things it encountered in one virtual environment might look like elsewhere. In other words, the neural net was able to disentangle certain objects that it encountered in a simulated environment from the environment itself.
This isn&#8217;t the same as a human&#8217;s imagination, where we can come up with new mental images altogether (think of a bird — you can probably conjure up an image of what a fictional spherical, red bird might look like in your mind&#8217;s eye.) The AI system isn&#8217;t that sophisticated, but it can imagine objects that it&#8217;s already seen in new configurations or locations.
“We want a machine to learn safe common sense in its exploration so it&#8217;s not damaging itself,” said Higgins in her speech at the conference, which had been organized by GoodAI. She had published <a href="https://arxiv.org/pdf/1808.06508.pdf">her paper</a> on the preprint server arXiv earlier that week, describing work that allows <a href="https://openreview.net/pdf?id=rkN2Il-RZ">previously-developed</a> <a href="https://openreview.net/pdf?id=Sy2fzU9gl">AI agents</a> to continuously learn without forgetting earlier training.
Let’s say you’re walking through the desert (as one does) and you come across a cactus. One of those big, two-armed ones you see in all the cartoons. You can recognize that this is a cactus because you have probably encountered one before. Maybe your office bought some succulents to liven up the place. But even if your office is cactus-free, you could probably imagine what this desert cactus would look like in a big clay pot, maybe next to Brenda from accounting’s desk.
Now Higgins’ AI system can do pretty much the same thing. With just five examples of how a given object looks from various angles, the AI agent learns what it is, how it relates to the environment, and also <a href="https://futurism.com/google-ai-surroundings-3d-model/">how it might look from other angles it hasn&#8217;t seen </a>or in different lighting. The paper highlights how the algorithm was trained to spot a white suitcase or an armchair. After its training, the algorithm can then <a href="https://deepmind.com/blog/imagine-creating-new-visual-concepts-recombining-familiar-ones/#gif-96">imagine how that object would look</a> in an entirely new virtual world and recognize the object when it encounters it there.
“We run the exact setup that I used to motivate this model, and then we present an image from one environment and ask the model to imagine what it would look like in a different environment,” Higgins said. Again and again, her new algorithm excelled at the task compared to AI systems with entangled representations, which could predict fewer qualities and characteristics of the objects.
<figure id="attachment_136721" style="width: 1200px" class="wp-caption alignnone"><img class="wp-image-136721 size-large" src="https://wp-assets.futurism.com/2018/08/cactus-setting-2-1200x630.png" alt="" width="1200" height="630" /><figcaption class="wp-caption-text">Image Credit: Emily Cho</figcaption></figure>
In short, the algorithm is able to note differences between what it encounters and what it has seen in the past. Like most people but unlike most other algorithms, the new system Higgins built for Google can understand that it hasn&#8217;t come across a brand new object just because it&#8217;s seeing something from a new angle. It can then use some spare computational power to take in that new information; the AI system updates what it knows about the world without needing to be retrained and re-learn everything all over again. Basically, the system is able to transfer and apply its existing knowledge to the new environment. The end result is a sort of spectrum or continuum showing how it understands various qualities of an object.
Higgins’ model alone won’t get us to AGI, of course. But it marks an important first step towards AI algorithms that can continuously update as they go, learning new things about the world <a href="https://futurism.com/we-created-ai-and-now-they-are-teaching-us/">without losing what they already had</a>.
“I think it’s very crucial to reach anything close to artificial general intelligence,” Higgins said.
<q class="pullquote">“I think it’s very crucial to reach anything close to artificial general intelligence.”</q>
And this work is all still in its early stages. These algorithms, like many other object recognition AI tools, excel at a rather narrow task with a <a href="https://www.theverge.com/2018/8/21/17761424/ai-algorithm-art-machine-vision-perception-tom-white-treachery-imagenet">constrained set of rules</a>, such as looking at a photo and picking out a face among many things that are not faces. But Higgins&#8217; new AI system is doing a narrow task in such a way that more closely resembles creativity and some digital simulation of an imagination.
And even though Higgins’ research didn’t immediately bring about the era of artificial general intelligence, her new algorithm already has the ability to improve the existing AI systems we use all the time. For instance, Higgins tried her new AI system on a major set of data used to train facial recognition software. After analyzing the thousands and thousands of headshots found in the dataset, the algorithm could create a spectrum of any quality with which those photos have been labeled. As an example, Higgins presented the spectrum of faces ranked by skin tone.
Higgins then revealed that her algorithm was able to do the same for the subjective qualities that also find their ways into these datasets, ultimately <a href="https://futurism.com/microsoft-announces-tool-catch-biased-ai/">teaching human biases</a> to facial recognition AI. Higgins showed how images that people had labeled as “attractive” created a spectrum that pointed straight towards the photos of young, pale women. That means any AI system that had been trained with these photos — and there are many of them out there — now hold the same racist views as do the people who labeled the photos in the first place: that white people are more attractive.
This creative new algorithm is already better than we are when it comes to finding new ways to <a href="https://futurism.com/trust-artificial-intelligence-ibm/">detect human biases</a> in other algorithms so engineers can go in and remove them.
So while it can’t replace artists quite yet, Higgins&#8217; team&#8217;s work is a pretty big step towards getting AI to imagine more like a human and less like an algorithm.
More on Artificial General Intelligence: <a href="https://futurism.com/advanced-artificial-intelligence-better-humans/">Advanced Artificial Intelligence Could Run The World Better Than Humans Ever Could</a>

It's a really important first step towards artificial general intelligence, algorithms that can do more than a single narrowly-defined task.

New artificial intelligence does something extraordinary — it remembers

A new kind of artificial intelligence is able to recognize and remember objects in a way that's a lot more like humans and a lot less like algorithms.

New Artificial Intelligence Does Something Extraordinary — It Remembers

A new AI system can recognize a cactus in a desert and a cactus in an office. That's actually a really big deal.

There are fears that tend to come up when people talk about futuristic artificial intelligence — say, one that could teach itself to learn and become more advanced than anything we humans might be able to comprehend. In the wrong hands, perhaps even on its own, such an advanced algorithm might dominate the world’s governments and militaries, impart <a href="https://futurism.com/facial-recognition-us-borders/">Orwellian levels of surveillance</a>, manipulation, and social control over societies, and perhaps even control entire battlefields of <a href="https://futurism.com/pledge-autonomous-killing-machines-ignore-problem/">autonomous lethal weapons</a> such as military drones.
But some artificial intelligence experts don&#8217;t think those fears are well-founded. In fact, highly-advanced artificial intelligence could be better at managing the world than humans have been. These fears themselves are the real danger, because they may hold us back from making that potential a reality.
“Maybe not achieving AI is the danger for humanity,” <a href="https://research.fb.com/people/mikolov-tomas/">Tomas Mikolov</a>, a research scientist for Facebook AI, said at <a href="https://www.hlai-conf.org/program">The Joint Multi-Conference on Human-Level Artificial Intelligence</a>, organized by GoodAI, in Prague on Saturday.
<q class="pullquote">“Maybe not achieving AI is the danger for humanity.”</q>
As a species, Mikolov explained, humans are pretty terrible at making choices that are good for us in the long term. People have carved away rainforests and other ecosystems to harvest raw materials, unaware of (or uninterested in) how they were contributing to the slow, maybe-irreversible degradation of the planet overall.
But a sophisticated artificial intelligence system might be able to protect humanity from its own shortsightedness.
“We as humans are very bad at making predictions of what will happen in some distant timeline, maybe 20 to 30 years from now,” Mikolov added. “Maybe making AI that is much smarter than our own, in some sort of symbiotic relationship, can help us avoid some future disasters.”
Granted, Mikolov may be in the minority in thinking a superior AI entity would be benevolent. Throughout the conference, many other speakers expressed these common fears, mostly about AI used for dangerous purposes or misused by malicious human actors. And we shouldn&#8217;t laugh off or downplay those concerns.
We don’t know for sure whether it will ever be possible to create artificial general intelligence, often considered the holy grail of sophisticated AI that&#8217;s capable of doing pretty much any cognitive task humans can, maybe even doing it better.
The future of advanced artificial intelligence is promising, but it comes with a lot of ethical questions. We probably don&#8217;t know all the questions we&#8217;ll have to answer yet.
But most of the panelists at the HLAI conference agreed that we still need to decide on the rules before we need them. The time to create international agreements, ethics boards, and regulatory bodies across governments, private companies, and academia? It&#8217;s now. Putting these institutions and protocols in place would reduce the odds that a hostile government, unwitting researcher, or even a cackling mad scientist would unleash <a href="https://futurism.com/should-evil-ai-research-be-published-five-experts-weigh-in/">a malicious AI system</a> or otherwise weaponize advanced algorithms. And if something nasty did get out there, then these systems would ensure we&#8217;d have ways to handle it.
With these rules and safeguards in place, we will be much more likely to usher in a future in advanced AI systems live harmoniously with us, or perhaps even save us from ourselves.
More on the future of artificial intelligence: <a href="https://futurism.com/artificial-general-intelligence-governments-invest/">If We Ever Want Artificial General Intelligence, Governments Need To Invest In It</a>

We shouldn't be afraid of AI that's smarter than us — we should create it so it can help us.

Advanced artificial intelligence could run the world better than humans ever could

Humanity is terrible at planning for the distant future. If we build advanced artificial intelligence, it might save us from ourselves.

Advanced Artificial Intelligence Could Run The World Better Than Humans Ever Could

Humans are pretty terrible at making choices that are good for us in the long term. AI could do better.

Pretty much every tech startup that boasts its use of artificial intelligence is actually focused on an ultra-specific problem. Visual effects company Digital Domain has an AI algorithm that <a href="https://futurism.com/artificial-intelligence-automating-hollywood-art/">automates and enhances video editing</a>; healthcare startup Babylon developed <a href="https://www.forbes.com/sites/forbesdigitalcovers/2018/06/29/this-startup-wants-to-replace-your-doctor-with-a-chatbot/">an AI chatbot</a> to answer the constant barrage of patient questions.
These tools are fine. They do what they&#8217;re supposed to. But they&#8217;ll never lead to something more, something bigger. No matter how great these AI systems sound, no startup will ever stumble upon artificial general intelligence (AGI), which is essentially <a href="https://futurism.com/the-most-important-question-underlying-artificial-general-intelligence-research-is-math-real/">a complete, human-like AI system</a> that has become truly intelligent. No amount of optimizing systems to get better at a particular task would ever lead to AGI.
<q class="pullquote">No startup will ever stumble upon artificial general intelligence.</q>
It makes sense, of course, that these companies are so narrowly focused. Mastering those specific applications means these companies get to keep doing business.
But it&#8217;s also a real shame. Because the emergence of an AGI would revolutionize everything we know about technology.
An artificial intelligence system capable of abstract reasoning, creativity, and problem-solving would inspire large-scale, rapid changes in the way we tackle problems in space exploration, healthcare, security, and many other aspects of the economy — and our lives. To whatever investors owned a piece of the pie, such an algorithm would be priceless. But based on how investors and venture capitalists approach their business, the people who might want to build AGI would never get the private funding they needed. So if AGI stands any chance of happening, that money will have to come from somewhere else — likely a major government.
“Investors only fund something when they see the end of the tunnel, and in AI it’s very far,” said <a href="https://www.ipmllp.com/team/marian-gazdik/">Marian Gazdik</a>, Managing Director of Startup Grind Europe. He thinks governments should do that investing instead. As a venture capitalist, Gazdik doesn’t necessarily look for AI companies, but rather chooses to invest in companies with amazing technology that could actually solve a real-world problem.
Gazdik spoke at the “Investing in General AI” panel of <a href="https://www.hlai-conf.org">The Joint Multi-Conference </a><a href="https://www.hlai-conf.org">on Human-Level Artificial Intelligence</a> organized by GoodAI last week in Prague. He, along with the other five tech investors on the panel, arrived at two reasons for a venture capitalist to invest in a new AI startup emerged. And neither of them are conducive to any of the big-picture innovation that would give rise to an AGI.
“I very much like General AI as an intellectual, but as an investor not as much,” said <a href="https://www.crunchbase.com/person/tak-lo">Tak Lo</a>, a partner at Zeroth.ai, an Asian accelerator that invests in technology startups, who also spoke on the panel.
<figure id="attachment_136430" style="width: 1200px" class="wp-caption alignnone"><img class="wp-image-136430 size-large" src="https://wp-assets.futurism.com/2018/08/artificial-general-intelligence2-1200x800.jpg" alt="" width="1200" height="800" /><figcaption class="wp-caption-text">Image Credit: <a href="https://www.pexels.com/photo/lights-abstract-curves-long-exposure-1944/">Little Visuals/Adobe Stock/Victor Tangermann</a></figcaption></figure>
Lo, like other venture capitalists who back AI-focused startups, usually finds it more profitable to invest in companies with great business models that use AI to solve a big problem, or companies that got their hands on a large, valuable dataset for training algorithms.
AGI, on the other hand, is just too darn far away to promise a return on investment in a reasonable timeframe. The panelists agreed that venture capitalists want to see their money back (and then some) in five to 10 years, but certainly no longer than that. And for many private companies, including giants like the Googles and IBMs of the world, the uncertainty surrounding AGI makes for too long of a wait before seeing any returns, said <a href="https://github.com/jkot">panelist Jakub Kotowski</a>, the Artificial Intelligence Lead at MSD Global Software Engineering.
And any sort of truly intelligent AI system like an AGI is very far off in the future, if it will ever exist at all, as experts (such as Facebook AI research scientist <a href="https://research.fb.com/people/mikolov-tomas/">Tomas Mikolov</a> and The U.N.’s head of Robotics and AI <a href="https://www.itu.int/en/fnc/Pages/bios/BERIDZEIrakli.aspx">Irakli Beridze</a>) told Futurism in private interviews outside of this panel. Not exactly a promising investment.
Private industry, then, isn&#8217;t going to get us to our AGI future. So the panelists called on governments and other forms of public funding to get the ball rolling. For government funds, they say, shareholders or an investor&#8217;s timeline are irrelevant — instead, these funders can throw sizable sums of money at these distant, lofty dreams in the interest of inching us closer to it for the common good.
Once the government has thrown enough money at the problem to have a few solutions to show for it, startups and private investors can pounce on the most viable ones, should AGI become a feasible, short-term goal.
Take flying cars. Companies like <a href="https://futurism.com/volvo-acquired-flying-car-company/">Volvo</a>, <a href="https://futurism.com/uber-ceo-flying-cars-only-decade-away/">Uber</a>, and <a href="https://futurism.com/boeing-gofly-competition-personal-flying-transport-vehicle/">Boeing </a>are all racing to be the first to develop flying personal vehicles. But those companies are depending on <a href="https://sanfrancisco.cbslocal.com/2018/05/08/uber-nasa-flying-taxis/">partnerships with NASA</a> and UC Berkeley, a public university, in order to <a href="https://futurism.com/uber-flying-taxis/">actually figure out how to get there</a>.
“Since investors are like sheep, they like to invest in areas in which others are investing. They always wait for the first big move,” said Gazdik.
<q class="pullquote">“Since investors are like sheep, they like to invest in areas in which others are investing. They always wait for the first big move.”</q>
The U.S. government might not be throwing enough money at AGI to ensure that the strongest competitor will develop there. Russia and China are investing heavily in artificial intelligence, and are <a href="https://futurism.com/chinese-ai-race-winning-china/">relying on their federal infrastructures</a> to do it. In the U.S., much of that government investment flows through research-focused universities and partnerships with tech giants like Facebook and Google.
So how, exactly, should governments invest in AGI to help make the revolutionary technology a reality? None of the panelists could come up with a concrete suggestion. Instead, they vaguely suggested that governments, companies, and academics ought to all “play to their strengths,” and “focus on what they should be doing.” No one elaborated on what exactly that means, but Lo did suggest that government agencies should focus their efforts on thinking about where society will be in the coming decades and craft appropriate policy, standards, and protocols.
But as far as government investments go, some of the panelists suggested that federal agencies ought to put their money in the hands of private managers, such as the panelists themselves, who might be able to invest it more strategically than a bureaucrat would know how to (this would, of course, benefit the panelists a great deal, so take it with a grain of salt).
For any of this to happen, a government has to decide the goal of developing the world’s first general artificial intelligence system is a high-ranking priority that deserves a slot on a competitive budget sheet. So far, none has. And until one does, AGI will always lie just beyond any tech company’s grasp, always lurking just beyond the timeframe for commercial investment.
More on how to develop AI: <a href="https://futurism.com/trust-artificial-intelligence-ibm/">To Build Trust In Artificial Intelligence, IBM Wants Developers To Prove Their Algorithms Are Fair</a>

No startup, no private company can afford to look that far ahead.

If we ever want artificial general intelligence, governments need to invest in it

A panel of venture capitalists and other experts on financing AI startups agree: to get a truly general AI system, governments will need to step up.

If We Ever Want Artificial General Intelligence, Governments Need To Invest In It

Startups with super-narrow algorithms aren't ever going to make AGI happen.

A rhetorical question for you. Let’s say you’re an AI scientist, and you’ve found the holy grail of your field — you figured out how to build an artificial general intelligence (AGI). That&#8217;s a truly intelligent computer that could pass as human in terms of cognitive ability or emotional intelligence. AGI would be creative and find links between disparate ideas — things no computer can do today.
That&#8217;s great, right? Except for one big catch: your AGI system is evil or could only be used for malicious purposes.
So, now a conundrum. Do you publish your white paper and tell the world exactly how to create this unrelenting force of evil? Do you file a patent so that no one else (except for you) could bring such an algorithm into existence? Or do you sit on your research, protecting the world from your creation but also passing up on the astronomical paycheck that would surely arrive in the wake of such a discovery?
Yes, this is a rhetorical question — for now. But some top names in the world of AI are already thinking about their answers. On Friday, speakers at the “AI Race and Societal Impacts” panel of <a href="https://www.hlai-conf.org/program">The Joint Multi-Conference on Human-Level Artificial Intelligence</a>, which was organized by GoodAI, in Prague gave their best responses after the question was posed by an audience member.
Here’s how five panelists, all experts on the future of AI, responded.
<a href="https://www.darpa.mil/staff/dr-hava-siegelmann">Hava Siegelmann</a>, Program Manager at DARPA
Siegelmann urged the hypothetical scientist to publish their work immediately. Siegelmann had earlier told Futurism that she believes there is no evil technology, but there are people who would misuse it. If that AGI algorithm was shared with the world, people might be able to find ways to use it for good.
But after Siegelmann answered, the audience member who posed the hypothetical question clarified that, for the purposes of the thought experiment, we should assume that no good could ever possibly come from the AGI.
<a href="https://www.itu.int/en/fnc/Pages/bios/BERIDZEIrakli.aspx">Irakli Beridze</a>, Head of the Centre for Artificial Intelligence and Robotics, UNICRI, United Nations
Easy one: “Don’t publish it!”
Beridze otherwise stayed out of the fray for this specific question, but throughout the conference he highlighted the importance of setting up strong ethical benchmarks on how to develop and deploy AGI. Apparently, deliberately releasing an evil super-intelligent entity into the world would go against those standards.
<a href="https://www.general-ai-challenge.org/ai-race">Alexey Turchin</a>, author and finalist in GoodAI&#8217;s &#8220;<a href="https://www.general-ai-challenge.org/ai-race">Solving the AI Race</a>&#8221; challenge
Turchin believes there are responsible ways to handle such an AI system. Think about a grenade, he said — one should not hand it to a small child, but maybe a trained soldier could be trusted with it.
But Turchin&#8217;s example is more revealing than it may initially appear. A hand grenade is a weapon created explicitly to cause death and destruction no matter who pulls the pin, so it’s difficult to imagine a so-called responsible way to use one. It&#8217;s not clear whether Turchin intended his example to be interpreted this way, but he urged the AI community to make sure dangerous algorithms were left only in the most trustworthy hands.
<a href="https://www.crunchbase.com/person/tak-lo">Tak Lo</a>, a partner at Zeroth.ai, an accelerator that invests in AI startups 
Lo said the hypothetical computer scientist should sell the evil AGI to him. That way, they wouldn&#8217;t have to hold onto the ethical burden of such a powerful and scary AI — instead, you could just pass it to Lo and he would take it from there. Lo was likely (at least half-)kidding, and the audience laughed. Earlier that day, Lo said that private capital and investors should be used to push AI forward, and he may have been poking fun at his own obviously capitalistic stance. Still, someone out there would absolutely try to buy such an AGI system, should it arrive.
But what Lo suggests, in jest or not, is one of the most likely results, should this actually come to pass. While hobbyists can develop truly valuable and innovative algorithms, much of the top talent in the AI field is scooped up by large companies who then own the products of their labor. The other likely scenario is that the scientist would publish their paper on an open-access preprint server <a href="https://arxiv.org/list/cs.AI/recent">like arXiv</a> to help promote transparency.
<a href="http://www.crassh.cam.ac.uk/people/profile/sean-o-heigeartaigh">Seán Ó hÉigeartaigh</a>, Executive Director of the Cambridge Center for the Study of Existential Risk
Ó hÉigeartaigh agreed with Beridze: you shouldn&#8217;t publish it. “You don’t just share that with the world! You have to think about the kind of impact you will have,” he said.
And with that, the panel ended. Everyone went on their merry way, content that this evil AGI was safe in the realm of the hypothetical.
In the &#8220;real world,&#8221; though, ethics often end up taking a back seat to more earthly concerns like money and prestige. Companies like <a href="https://futurism.com/darpa-stop-deepfakes-facebook-google/">Facebook, Google</a>, and <a href="https://futurism.com/amazon-rekognition-employees/">Amazon</a> regularly publish facial recognition or other surveillance systems, often selling them to <a href="https://futurism.com/lapd-documents-show-their-policing-algorithms-continue-to-target-minorities-and-past-offenders/">police or the military</a> which uses it to monitor everyday people. Academic scientists are trapped in the “publish or perish,” cycle — publish a study, or risk losing your position. So ethical concerns are often relegated to a paper&#8217;s conclusion, as a factor for someone else to sort out at some vague point in the future.
For now, though, it&#8217;s unlikely that anyone will come up with AGI — much less evil AGI — anytime soon. But the panelists&#8217; wide-ranging answers means that we are still far from sorting out what should be done with unethical, dangerous science.
More about evil AI thought experiments: <a href="https://futurism.com/grimes-musk-rococo-basilisk/">Grimes, Elon Musk, and the Supposedly Trauma-Inducing A.I. Theory That Brought Them Together</a>

Should you sacrifice your livelihood to save the world? Or is there a responsible way to bring your work to life?

Should evil AI research be published? Five experts weigh in.

Five experts speaking at a conference on Human-Level Artificial Intelligence weigh in on whether scientists should create AI they know will be evil.

Should Evil AI Research Be Published? Five Experts Weigh In.

Upside: you created artificial general intelligence! Downside: it's evil. 

<h2>Putting the &#8220;AI Hype&#8221; in Perspective</h2>
Today, many of the world&#8217;s leading companies are in a one-of-a-kind race: To bring artificial intelligence (AI) to life. Already, machine learning systems are the core of many businesses, so it&#8217;s no surprise that updates about this AI or that neural net often pop up on our newsfeed. Such headlines typically read along the lines of, &#8220;AI beats human players in video game&#8221; or &#8220;AI mimics human speech&#8221; and even sometimes things like &#8220;AI detects cancer using machine learning.&#8221;
But just how close are we to having machines with the intelligence of a human—machines that we can talk with and work with like we do any other individual? Machines that are conscious?
While all of the aforementioned developments are real, Yann LeCun, Director of AI Research at Facebook and a professor of computer science at NYU, thinks that we may be overestimating the abilities of today&#8217;s AI, and, thus building up a bit of hype. &#8220;We&#8217;re very far from having machines that can learn the most basic things about the world in the way humans and animals can do,&#8221; LeCun told The Verge in <a href="https://www.theverge.com/2017/10/26/16552056/a-intelligence-terminator-facebook-yann-lecun-interview" target="_blank" rel="noopener">an interview published last week</a>. &#8220;Like, yes, in particular areas machines have superhuman performance, but in terms of general intelligence we&#8217;re not even close to a rat.&#8221;
<figure style="width: 360px" class="wp-caption alignright"><a href="https://futurism.com/images/the-evolution-of-brain-computer-interfaces-infographic/"><img src="https://wp-assets.futurism.com/2016/11/Brain-Computer-Interface.png" alt="The Evolution of Brain-Computer Interfaces [INFOGRAPHIC]" ></a><figcaption class="wp-caption-text center"><a href="https://futurism.com/images/the-evolution-of-brain-computer-interfaces-infographic/" target="_blank">Click to View Full Infographic</a></figcaption></figure>
This so-called artificial general intelligence (AGI) refers to an AI operator capable of performing virtually every task a human being could. Conversely, today&#8217;s AI specialize in particular tasks: for example, image or speech recognition, or identifying patterns by sifting through tons of data that machine learning AIs have been trained on. These specialized AIs are also called &#8220;applied AI&#8221; or &#8220;narrow AI&#8221; to highlight their rather limited intelligence.
Speaking to Futurism via email, Manuel Cebrian, one of the MIT researchers that developed Shelley, an <a href="https://futurism.com/a-new-ai-is-writing-perverse-horror-fiction/" target="_blank" rel="noopener">AI horror storyteller</a>, agreed with LeCun&#8217;s sentiments. &#8220;AI is just a great tool,&#8221; he said, adding that, &#8220;it seems to me, based on my work with Shelley, that AI is very far from being able to create professional-level horror fiction.&#8221; And thus, still quite far from human levels of intelligence.
LeCun clarifies that we shouldn&#8217;t devalue the significant work that AI researchers have made in recent months and years, but that work in machine learning and neural networks is not the same as developing true artificial intelligence. &#8220;So for example, and I don&#8217;t want to minimize at all the engineering and research work done on AlphaGo by our friends at DeepMind, but when [people interpret the development of AlphaGo] as significant progress towards general intelligence, it&#8217;s wrong,&#8221; LeCun added. &#8220;It just isn&#8217;t.&#8221;
Pierre Barreau, CEO of Aiva Technologies, the company behind the <a href="https://futurism.com/a-new-ai-can-write-music-as-well-as-a-human-composer/">music-composing AI Aiva</a>, also thinks that the advancements that we have made towards synthetic intelligence are overstated. &#8220;AGI is a very hyped topic,&#8221; he noted via email. &#8220;I am, in general, quite optimistic about how fast tech develops, but I think a lot of people don’t realize the complexity of our own brain, let alone creating an artificial one.&#8221;
<h2>Making Artificial General Intelligence</h2>
People often use AI-related terms as synonymous with true artificial intelligence. News coverage drops terms like machine learning or deep learning together with artificial neural networks whenever AI is discussed. While each of these have something to do with AI, these aren&#8217;t exactly AI per se.
Machine learning is a tool: a set of algorithms that learn by ingesting huge amounts of data, from which an intelligent system is constructed. Similarly, deep learning refers to a kind of machine learning that is not task-specific. An artificial neural network, on the other hand, is a system that mimics the way the human brain works, and upon which machine learning algorithms are built.
All of these, AI experts believe, are the foundation for a synthetic intelligence with truly human cognition. But this is just the nascent stage; we have made a lot of progress, but current research isn&#8217;t really close to creating true intelligence.
So the big question is, when can we expect to have this type of intelligent AI? What&#8217;s the specific timeline?
For Luke Tang, general manager of AI startup accelerator <a href="http://www.techcode.com/index.php?m=wap">TechCode</a>, the shift will start with a &#8220;breakthrough in unsupervised learning algorithms.&#8221; Once this is accomplished, &#8220;machine intelligence can quickly surpass human intelligence,&#8221; he said in a statement sent to Futurism.
Needless to say, the path to this will be quite challenging. &#8220;In order to achieve AGI, there will need to be major breakthroughs not just in software, but also in Neuroscience and Hardware,&#8221; Barreau explained. He clarified, &#8220;We are starting to hit the <a href="https://futurism.com/new-discovery-breaks-the-boundaries-of-moores-law/">ceiling of Moore’s law</a>, with transistors being as small as they can physically get. New hardware platforms like quantum computing have not yet shown that they can beat performances of our usual hardware in all tasks.&#8221;
Indeed, for an AI to be considered truly intelligent, most agree that it has <a href="http://intelligence.org/2013/08/11/what-is-agi/" target="_blank" rel="noopener">to pass at least five tests</a>, foremost of which is <a href="https://futurism.com/synthetic-cells-have-passed-a-turing-test/">the Turing Test</a>—where a machine and a human both converse with a second human being, who will determine which one is a machine. Barreau said that he&#8217;s confident that we will see in our lifetime an AI passing the Turing Test; i.e., that it would pass as a human being. However, he says this won&#8217;t necessarily be &#8220;AGI, but good enough to pass as AGI.&#8221;
<iframe width="500" height="281" src="https://www.youtube.com/embed/9snY7lhJA4c?feature=oembed" frameborder="0" gesture="media" allowfullscreen></iframe>
<h2>A Case for Augmented Intelligence</h2>
It goes without saying that an AGI is the prerequisite for the so-called singularity. If you aren&#8217;t familiar with the concept of &#8220;singularity,&#8221; it&#8217;s essentially that moment when intelligent machines surpass humankind&#8217;s levels of intelligence, spurring runaway and exponential technological growth that will transform the foundations of life as we know it. The term was <a href="https://www.frc.ri.cmu.edu/~hpm/book98/com.ch1/vinge.singularity.html">coined in 1993 by Vernor Vinge</a>, who wrote: &#8220;We will soon create intelligences greater than our own. When this happens, human history will have reached a kind of singularity, an intellectual transition as impenetrable as the knotted space-time at the center of a black hole, and the world will pass far beyond our understanding.&#8221;
While this is something SoftBank CEO Masayoshi Son and <a href="https://futurism.com/kurzweil-claims-that-the-singularity-will-happen-by-2045/">Google&#8217;s Ray Kurzweil</a> are excitedly looking forward to, other brilliant minds of today, such as Elon Musk, Stephen Hawking, and even Bill Gates, aren&#8217;t quite as keen for this moment. They assert that, in the same way that we don&#8217;t really understand what it means to have a super-intelligent AI, we&#8217;re also not prepared for whatever consequences the singularity would bring.
But what if we shift our perspective a bit? Instead of looking at AI as humanity&#8217;s downfall, why not see it as a partner? Musk seems to hint on this with his <a href="https://futurism.com/elon-musk-wants-meld-human-brain-computers/" target="_blank" rel="noopener">Neuralink project</a>, while Kurzweil mentioned this when he <a href="https://futurism.com/kurzweil-by-2030-nanobots-will-flow-throughout-our-bodies/" target="_blank" rel="noopener">talked about nanobots living inside us</a>, augmenting our capabilities. The key word here is augmenting, something <a href="https://futurism.com/googles-ai-is-a-new-paradigm-that-unites-humans-and-machines/" target="_blank" rel="noopener">Google&#8217;s current push for AI seems to be setting the ground for</a>.
&#8220;We should focus our efforts on an exciting outcome of AI: augmented intelligence (i.e. human intelligence being augmented by AI),&#8221; Barreau said. Like Aiva and Shelley, other <a href="https://futurism.com/the-worlds-first-album-composed-and-produced-by-an-ai-has-been-unveiled/" target="_blank" rel="noopener">AIs have done considerably well</a> when working side-by-side with human beings.
Still, with intelligent robots like Hanson Robotics&#8217; Sophia and SotfBank&#8217;s Pepper, it does not seem very far-fetched to imagine truly intelligent machines living among us. Could Masayoshi Son&#8217;s super-intelligent AI, <a href="https://futurism.com/softbank-ceo-promises-super-artificial-intelligences-with-iq-of-10000-in-30-years/" target="_blank" rel="noopener">with an IQ of 10,000,</a> be the cognitive machine intelligence we&#8217;re looking for? If that&#8217;s the case, we might have to wait for at least three more decades. &#8220;It’s probably only 30 to 50 years away,&#8221; Tang said. &#8220;So it is likely–it will just take some time to get there. But it also means many of us will have a chance to see that day come!&#8221;

According to experts, synthetically intelligent beings will soon be in our midst.

The rise of conscious AI is just decades away

Artificial intelligence, autonomous, technology

It's frequently said that conscious AI are on the horizon. Here, we speak with experts in order to determine how much truth is in these claims.

/Agi

Bureaucratic Takeover

There’s a Power Struggle Inside Google to Control Superhuman AI

CLEAR AIs, FULL HEARTS

Why Should We Bother Building Human-Level AI? Five Experts Weigh In

Artificial Stupidity

You Have No Idea What Artificial Intelligence Really Does

Artificial Intelligence

When Will We Have Artificial Intelligence As Smart as a Human? Here’s What Experts Think

Artificial Intelligence

Five Experts Share What Scares Them the Most About AI

Artificial Intelligence

New Artificial Intelligence Does Something Extraordinary — It Remembers

Artificial Intelligence

Advanced Artificial Intelligence Could Run The World Better Than Humans Ever Could

Artificial Intelligence

If We Ever Want Artificial General Intelligence, Governments Need To Invest In It

Artificial Intelligence

Should Evil AI Research Be Published? Five Experts Weigh In.

Artificial Intelligence

The Rise of Conscious AI is Just Decades Away

Load More