Tag Archives: error
You don’t have to dig too deeply into the archive of dystopian science fiction to uncover the horror that intelligent machines might unleash. The Matrix and The Terminator are probably the most well-known examples of self-replicating, intelligent machines attempting to enslave or destroy humanity in the process of building a brave new digital world.
The prospect of artificially intelligent machines creating other artificially intelligent machines took a big step forward in 2017. However, we’re far from the runaway technological singularity futurists are predicting by mid-century or earlier, let alone murderous cyborgs or AI avatar assassins.
The first big boost this year came from Google. The tech giant announced it was developing automated machine learning (AutoML), writing algorithms that can do some of the heavy lifting by identifying the right neural networks for a specific job. Now researchers at the Department of Energy’s Oak Ridge National Laboratory (ORNL), using the most powerful supercomputer in the US, have developed an AI system that can generate neural networks as good if not better than any developed by a human in less than a day.
It can take months for the brainiest, best-paid data scientists to develop deep learning software, which sends data through a complex web of mathematical algorithms. The system is modeled after the human brain and known as an artificial neural network. Even Google’s AutoML took weeks to design a superior image recognition system, one of the more standard operations for AI systems today.
Of course, Google Brain project engineers only had access to 800 graphic processing units (GPUs), a type of computer hardware that works especially well for deep learning. Nvidia, which pioneered the development of GPUs, is considered the gold standard in today’s AI hardware architecture. Titan, the supercomputer at ORNL, boasts more than 18,000 GPUs.
The ORNL research team’s algorithm, called MENNDL for Multinode Evolutionary Neural Networks for Deep Learning, isn’t designed to create AI systems that cull cute cat photos from the internet. Instead, MENNDL is a tool for testing and training thousands of potential neural networks to work on unique science problems.
That requires a different approach from the Google and Facebook AI platforms of the world, notes Steven Young, a postdoctoral research associate at ORNL who is on the team that designed MENNDL.
“We’ve discovered that those [neural networks] are very often not the optimal network for a lot of our problems, because our data, while it can be thought of as images, is different,” he explains to Singularity Hub. “These images, and the problems, have very different characteristics from object detection.”
AI for Science
One application of the technology involved a particle physics experiment at the Fermi National Accelerator Laboratory. Fermilab researchers are interested in understanding neutrinos, high-energy subatomic particles that rarely interact with normal matter but could be a key to understanding the early formation of the universe. One Fermilab experiment involves taking a sort of “snapshot” of neutrino interactions.
The team wanted the help of an AI system that could analyze and classify Fermilab’s detector data. MENNDL evaluated 500,000 neural networks in 24 hours. Its final solution proved superior to custom models developed by human scientists.
In another case involving a collaboration with St. Jude Children’s Research Hospital in Memphis, MENNDL improved the error rate of a human-designed algorithm for identifying mitochondria inside 3D electron microscopy images of brain tissue by 30 percent.
“We are able to do better than humans in a fraction of the time at designing networks for these sort of very different datasets that we’re interested in,” Young says.
What makes MENNDL particularly adept is its ability to define the best or most optimal hyperparameters—the key variables—to tackle a particular dataset.
“You don’t always need a big, huge deep network. Sometimes you just need a small network with the right hyperparameters,” Young says.
A Virtual Data Scientist
That’s not dissimilar to the approach of a company called H20.ai, a startup out of Silicon Valley that uses open source machine learning platforms to “democratize” AI. It applies machine learning to create business solutions for Fortune 500 companies, including some of the world’s biggest banks and healthcare companies.
“Our software is more [about] pattern detection, let’s say anti-money laundering or fraud detection or which customer is most likely to churn,” Dr. Arno Candel, chief technology officer at H2O.ai, tells Singularity Hub. “And that kind of insight-generating software is what we call AI here.”
The company’s latest product, Driverless AI, promises to deliver the data scientist equivalent of a chessmaster to its customers (the company claims several such grandmasters in its employ and advisory board). In other words, the system can analyze a raw dataset and, like MENNDL, automatically identify what features should be included in the computer model to make the most of the data based on the best “chess moves” of its grandmasters.
“So we’re using those algorithms, but we’re giving them the human insights from those data scientists, and we automate their thinking,” he explains. “So we created a virtual data scientist that is relentless at trying these ideas.”
Inside the Black Box
Not unlike how the human brain reaches a conclusion, it’s not always possible to understand how a machine, despite being designed by humans, reaches its own solutions. The lack of transparency is often referred to as the AI “black box.” Experts like Young say we can learn something about the evolutionary process of machine learning by generating millions of neural networks and seeing what works well and what doesn’t.
“You’re never going to be able to completely explain what happened, but maybe we can better explain it than we currently can today,” Young says.
Transparency is built into the “thought process” of each particular model generated by Driverless AI, according to Candel.
The computer even explains itself to the user in plain English at each decision point. There is also real-time feedback that allows users to prioritize features, or parameters, to see how the changes improve the accuracy of the model. For example, the system may include data from people in the same zip code as it creates a model to describe customer turnover.
“That’s one of the advantages of our automatic feature engineering: it’s basically mimicking human thinking,” Candel says. “It’s not just neural nets that magically come up with some kind of number, but we’re trying to make it statistically significant.”
Much digital ink has been spilled over the dearth of skilled data scientists, so automating certain design aspects for developing artificial neural networks makes sense. Experts agree that automation alone won’t solve that particular problem. However, it will free computer scientists to tackle more difficult issues, such as parsing the inherent biases that exist within the data used by machine learning today.
“I think the world has an opportunity to focus more on the meaning of things and not on the laborious tasks of just fitting a model and finding the best features to make that model,” Candel notes. “By automating, we are pushing the burden back for the data scientists to actually do something more meaningful, which is think about the problem and see how you can address it differently to make an even bigger impact.”
The team at ORNL expects it can also make bigger impacts beginning next year when the lab’s next supercomputer, Summit, comes online. While Summit will boast only 4,600 nodes, it will sport the latest and greatest GPU technology from Nvidia and CPUs from IBM. That means it will deliver more than five times the computational performance of Titan, the world’s fifth-most powerful supercomputer today.
“We’ll be able to look at much larger problems on Summit than we were able to with Titan and hopefully get to a solution much faster,” Young says.
It’s all in a day’s work.
Image Credit: Gennady Danilkin / Shutterstock.com Continue reading
The first time Dr. Blake Richards heard about deep learning, he was convinced that he wasn’t just looking at a technique that would revolutionize artificial intelligence. He also knew he was looking at something fundamental about the human brain.
That was the early 2000s, and Richards was taking a course with Dr. Geoff Hinton at the University of Toronto. Hinton, a pioneer architect of the algorithm that would later take the world by storm, was offering an introductory course on his learning method inspired by the human brain.
The key words here are “inspired by.” Despite Richards’ conviction, the odds were stacked against him. The human brain, as it happens, seems to lack a critical function that’s programmed into deep learning algorithms. On the surface, the algorithms were violating basic biological facts already proven by neuroscientists.
But what if, superficial differences aside, deep learning and the brain are actually compatible?
Now, in a new study published in eLife, Richards, working with DeepMind, proposed a new algorithm based on the biological structure of neurons in the neocortex. Also known as the cortex, this outermost region of the brain is home to higher cognitive functions such as reasoning, prediction, and flexible thought.
The team networked their artificial neurons together into a multi-layered network and challenged it with a classic computer vision task—identifying hand-written numbers.
The new algorithm performed well. But the kicker is that it analyzed the learning examples in a way that’s characteristic of deep learning algorithms, even though it was completely based on the brain’s fundamental biology.
“Deep learning is possible in a biological framework,” concludes the team.
Because the model is only a computer simulation at this point, Richards hopes to pass the baton to experimental neuroscientists, who could actively test whether the algorithm operates in an actual brain.
If so, the data could then be passed back to computer scientists to work out the next generation of massively parallel and low-energy algorithms to power our machines.
It’s a first step towards merging the two fields back into a “virtuous circle” of discovery and innovation.
The blame game
While you’ve probably heard of deep learning’s recent wins against humans in the game of Go, you might not know the nitty-gritty behind the algorithm’s operations.
In a nutshell, deep learning relies on an artificial neural network with virtual “neurons.” Like a towering skyscraper, the network is structured into hierarchies: lower-level neurons process aspects of an input—for example, a horizontal or vertical stroke that eventually forms the number four—whereas higher-level neurons extract more abstract aspects of the number four.
To teach the network, you give it examples of what you’re looking for. The signal propagates forward in the network (like climbing up a building), where each neuron works to fish out something fundamental about the number four.
Like children trying to learn a skill the first time, initially the network doesn’t do so well. It spits out what it thinks a universal number four should look like—think a Picasso-esque rendition.
But here’s where the learning occurs: the algorithm compares the output with the ideal output, and computes the difference between the two (dubbed “error”). This error is then “backpropagated” throughout the entire network, telling each neuron: hey, this is how far off you were, so try adjusting your computation closer to the ideal.
Millions of examples and tweakings later, the network inches closer to the desired output and becomes highly proficient at the trained task.
This error signal is crucial for learning. Without efficient “backprop,” the network doesn’t know which of its neurons are off kilter. By assigning blame, the AI can better itself.
The brain does this too. How? We have no clue.
What’s clear, though, is that the deep learning solution doesn’t work.
Backprop is a pretty needy function. It requires a very specific infrastructure for it to work as expected.
For one, each neuron in the network has to receive the error feedback. But in the brain, neurons are only connected to a few downstream partners (if that). For backprop to work in the brain, early-level neurons need to be able to receive information from billions of connections in their downstream circuits—a biological impossibility.
And while certain deep learning algorithms adapt a more local form of backprop— essentially between neurons—it requires their connection forwards and backwards to be symmetric. This hardly ever occurs in the brain’s synapses.
More recent algorithms adapt a slightly different strategy, in that they implement a separate feedback pathway that helps the neurons to figure out errors locally. While it’s more biologically plausible, the brain doesn’t have a separate computational network dedicated to the blame game.
What it does have are neurons with intricate structures, unlike the uniform “balls” that are currently applied in deep learning.
The team took inspiration from pyramidal cells that populate the human cortex.
“Most of these neurons are shaped like trees, with ‘roots’ deep in the brain and ‘branches’ close to the surface,” says Richards. “What’s interesting is that these roots receive a different set of inputs than the branches that are way up at the top of the tree.”
This is an illustration of a multi-compartment neural network model for deep learning. Left: Reconstruction of pyramidal neurons from mouse primary visual cortex. Right: Illustration of simplified pyramidal neuron models. Image Credit: CIFAR
Curiously, the structure of neurons often turn out be “just right” for efficiently cracking a computational problem. Take the processing of sensations: the bottoms of pyramidal neurons are right smack where they need to be to receive sensory input, whereas the tops are conveniently placed to transmit feedback errors.
Could this intricate structure be evolution’s solution to channeling the error signal?
The team set up a multi-layered neural network based on previous algorithms. But rather than having uniform neurons, they gave those in middle layers—sandwiched between the input and output—compartments, just like real neurons.
When trained with hand-written digits, the algorithm performed much better than a single-layered network, despite lacking a way to perform classical backprop. The cell-like structure itself was sufficient to assign error: the error signals at one end of the neuron are naturally kept separate from input at the other end.
Then, at the right moment, the neuron brings both sources of information together to find the best solution.
There’s some biological evidence for this: neuroscientists have long known that the neuron’s input branches perform local computations, which can be integrated with signals that propagate backwards from the so-called output branch.
However, we don’t yet know if this is the brain’s way of dealing blame—a question that Richards urges neuroscientists to test out.
What’s more, the network parsed the problem in a way eerily similar to traditional deep learning algorithms: it took advantage of its multi-layered structure to extract progressively more abstract “ideas” about each number.
“[This is] the hallmark of deep learning,” the authors explain.
The Deep Learning Brain
Without doubt, there will be more twists and turns to the story as computer scientists incorporate more biological details into AI algorithms.
One aspect that Richards and team are already eyeing is a top-down predictive function, in which signals from higher levels directly influence how lower levels respond to input.
Feedback from upper levels doesn’t just provide error signals; it could also be nudging lower processing neurons towards a “better” activity pattern in real-time, says Richards.
The network doesn’t yet outperform other non-biologically derived (but “brain-inspired”) deep networks. But that’s not the point.
“Deep learning has had a huge impact on AI, but, to date, its impact on neuroscience has been limited,” the authors say.
Now neuroscientists have a lead they could experimentally test: that the structure of neurons underlie nature’s own deep learning algorithm.
“What we might see in the next decade or so is a real virtuous cycle of research between neuroscience and AI, where neuroscience discoveries help us to develop new AI and AI can help us interpret and understand our experimental data in neuroscience,” says Richards.
Image Credit: christitzeimaging.com / Shutterstock.com Continue reading
I don’t have to open the doors of AImotive’s white 2015 Prius to see that it’s not your average car. This particular Prius has been christened El Capitan, the name written below the rear doors, and two small cameras are mounted on top of the car. Bundles of wire snake out from them, as well as from the two additional cameras on the car’s hood and trunk.
Inside is where things really get interesting, though. The trunk holds a computer the size of a microwave, and a large monitor covers the passenger glove compartment and dashboard. The center console has three switches labeled “Allowed,” “Error,” and “Active.”
Budapest-based AImotive is working to provide scalable self-driving technology alongside big players like Waymo and Uber in the autonomous vehicle world. On a highway test ride with CEO Laszlo Kishonti near the company’s office in Mountain View, California, I got a glimpse of just how complex that world is.
Camera-Based Feedback System
AImotive’s approach to autonomous driving is a little different from that of some of the best-known systems. For starters, they’re using cameras, not lidar, as primary sensors. “The traffic system is visual and the cost of cameras is low,” Kishonti said. “A lidar can recognize when there are people near the car, but a camera can differentiate between, say, an elderly person and a child. Lidar’s resolution isn’t high enough to recognize the subtle differences of urban driving.”
Image Credit: AImotive
The company’s aiDrive software uses data from the camera sensors to feed information to its algorithms for hierarchical decision-making, grouped under four concurrent activities: recognition, location, motion, and control.
Kishonti pointed out that lidar has already gotten more cost-efficient, and will only continue to do so.
“Ten years ago, lidar was best because there wasn’t enough processing power to do all the calculations by AI. But the cost of running AI is decreasing,” he said. “In our approach, computer vision and AI processing are key, and for safety, we’ll have fallback sensors like radar or lidar.”
aiDrive currently runs on Nvidia chips, which Kishonti noted were originally designed for graphics, and are not terribly efficient given how power-hungry they are. “We’re planning to substitute lower-cost, lower-energy chips in the next six months,” he said.
Testing in Virtual Reality
Waymo recently announced its fleet has now driven four million miles autonomously. That’s a lot of miles, and hard to compete with. But AImotive isn’t trying to compete, at least not by logging more real-life test miles. Instead, the company is doing 90 percent of its testing in virtual reality. “This is what truly differentiates us from competitors,” Kishonti said.
He outlined the three main benefits of VR testing: it can simulate scenarios too dangerous for the real world (such as hitting something), too costly (not every company has Waymo’s funds to run hundreds of cars on real roads), or too time-consuming (like waiting for rain, snow, or other weather conditions to occur naturally and repeatedly).
“Real-world traffic testing is very skewed towards the boring miles,” he said. “What we want to do is test all the cases that are hard to solve.”
On a screen that looked not unlike multiple games of Mario Kart, he showed me the simulator. Cartoon cars cruised down winding streets, outfitted with all the real-world surroundings: people, trees, signs, other cars. As I watched, a furry kangaroo suddenly hopped across one screen. “Volvo had an issue in Australia,” Kishonti explained. “A kangaroo’s movement is different than other animals since it hops instead of running.” Talk about cases that are hard to solve.
AImotive is currently testing around 1,000 simulated scenarios every night, with a steadily-rising curve of successful tests. These scenarios are broken down into features, and the car’s behavior around those features fed into a neural network. As the algorithms learn more features, the level of complexity the vehicles can handle goes up.
On the Road
After Kishonti and his colleagues filled me in on the details of their product, it was time to test it out. A safety driver sat in the driver’s seat, a computer operator in the passenger seat, and Kishonti and I in back. The driver maintained full control of the car until we merged onto the highway. Then he flicked the “Allowed” switch, his copilot pressed the “Active” switch, and he took his hands off the wheel.
What happened next, you ask?
A few things. El Capitan was going exactly the speed limit—65 miles per hour—which meant all the other cars were passing us. When a car merged in front of us or cut us off, El Cap braked accordingly (if a little abruptly). The monitor displayed the feed from each of the car’s cameras, plus multiple data fields and a simulation where a blue line marked the center of the lane, measured by the cameras tracking the lane markings on either side.
I noticed El Cap wobbling out of our lane a bit, but it wasn’t until two things happened in a row that I felt a little nervous: first we went under a bridge, then a truck pulled up next to us, both bridge and truck casting a complete shadow over our car. At that point El Cap lost it, and we swerved haphazardly to the right, narrowly missing the truck’s rear wheels. The safety driver grabbed the steering wheel and took back control of the car.
What happened, Kishonti explained, was that the shadows made it hard for the car’s cameras to see the lane markings. This was a new scenario the algorithm hadn’t previously encountered. If we’d only gone under a bridge or only been next to the truck for a second, El Cap may not have had so much trouble, but the two events happening in a row really threw the car for a loop—almost literally.
“This is a new scenario we’ll add to our testing,” Kishonti said. He added that another way for the algorithm to handle this type of scenario, rather than basing its speed and positioning on the lane markings, is to mimic nearby cars. “The human eye would see that other cars are still moving at the same speed, even if it can’t see details of the road,” he said.
After another brief—and thankfully uneventful—hands-off cruise down the highway, the safety driver took over, exited the highway, and drove us back to the office.
Driving into the Future
I climbed out of the car feeling amazed not only that self-driving cars are possible, but that driving is possible at all. I squint when driving into a tunnel, swerve to avoid hitting a stray squirrel, and brake gradually at stop signs—all without consciously thinking to do so. On top of learning to steer, brake, and accelerate, self-driving software has to incorporate our brains’ and bodies’ unconscious (but crucial) reactions, like our pupils dilating to let in more light so we can see in a tunnel.
Despite all the progress of machine learning, artificial intelligence, and computing power, I have a wholly renewed appreciation for the thing that’s been in charge of driving up till now: the human brain.
Kishonti seemed to feel similarly. “I don’t think autonomous vehicles in the near future will be better than the best drivers,” he said. “But they’ll be better than the average driver. What we want to achieve is safe, good-quality driving for everyone, with scalability.”
AImotive is currently working with American tech firms and with car and truck manufacturers in Europe, China, and Japan.
Image Credit: Alex Oakenman / Shutterstock.com Continue reading
On a dark night, away from city lights, the stars of the Milky Way can seem uncountable. Yet from any given location no more than 4,500 are visible to the naked eye. Meanwhile, our galaxy has 100–400 billion stars, and there are even more galaxies in the universe.
The numbers of the night sky are humbling. And they give us a deep perspective…on drugs.
Yes, this includes wow-the-stars-are-freaking-amazing-tonight drugs, but also the kinds of drugs that make us well again when we’re sick. The number of possible organic compounds with “drug-like” properties dwarfs the number of stars in the universe by over 30 orders of magnitude.
Next to this multiverse of possibility, the chemical configurations scientists have made into actual medicines are like the smattering of stars you’d glimpse downtown.
But for good reason.
Exploring all that potential drug-space is as humanly impossible as exploring all of physical space, and even if we could, most of what we’d find wouldn’t fit our purposes. Still, the idea that wonder drugs must surely lurk amid the multitudes is too tantalizing to ignore.
Which is why, Alex Zhavoronkov said at Singularity University’s Exponential Medicine in San Diego last week, we should use artificial intelligence to do more of the legwork and speed discovery. This, he said, could be one of the next big medical applications for AI.
Dogs, Diagnosis, and Drugs
Zhavoronkov is CEO of Insilico Medicine and CSO of the Biogerontology Research Foundation. Insilico is one of a number of AI startups aiming to accelerate drug discovery with AI.
In recent years, Zhavoronkov said, the now-famous machine learning technique, deep learning, has made progress on a number of fronts. Algorithms that can teach themselves to play games—like DeepMind’s AlphaGo Zero or Carnegie Mellon’s poker playing AI—are perhaps the most headline-grabbing of the bunch. But pattern recognition was the thing that kicked deep learning into overdrive early on, when machine learning algorithms went from struggling to tell dogs and cats apart to outperforming their peers and then their makers in quick succession.
[Watch this video for an AI update from Neil Jacobstein, chair of Artificial Intelligence and Robotics at Singularity University.]
In medicine, deep learning algorithms trained on databases of medical images can spot life-threatening disease with equal or greater accuracy than human professionals. There’s even speculation that AI, if we learn to trust it, could be invaluable in diagnosing disease. And, as Zhavoronkov noted, with more applications and a longer track record that trust is coming.
“Tesla is already putting cars on the street,” Zhavoronkov said. “Three-year, four-year-old technology is already carrying passengers from point A to point B, at 100 miles an hour, and one mistake and you’re dead. But people are trusting their lives to this technology.”
“So, why don’t we do it in pharma?”
Trial and Error and Try Again
AI wouldn’t drive the car in pharmaceutical research. It’d be an assistant that, when paired with a chemist or two, could fast-track discovery by screening more possibilities for better candidates.
There’s plenty of room to make things more efficient, according to Zhavoronkov.
Drug discovery is arduous and expensive. Chemists sift tens of thousands of candidate compounds for the most promising to synthesize. Of these, a handful will go on to further research, fewer will make it to human clinical trials, and a fraction of those will be approved.
The whole process can take many years and cost hundreds of millions of dollars.
This is a big data problem if ever there was one, and deep learning thrives on big data. Early applications have shown their worth unearthing subtle patterns in huge training databases. Although drug-makers already use software to sift compounds, such software requires explicit rules written by chemists. AI’s allure is its ability to learn and improve on its own.
“There are two strategies for AI-driven innovation in pharma to ensure you get better molecules and much faster approvals,” Zhavoronkov said. “One is looking for the needle in the haystack, and another one is creating a new needle.”
To find the needle in the haystack, algorithms are trained on large databases of molecules. Then they go looking for molecules with attractive properties. But creating a new needle? That’s a possibility enabled by the generative adversarial networks Zhavoronkov specializes in.
Such algorithms pit two neural networks against each other. One generates meaningful output while the other judges whether this output is true or false, Zhavoronkov said. Together, the networks generate new objects like text, images, or in this case, molecular structures.
“We started employing this particular technology to make deep neural networks imagine new molecules, to make it perfect right from the start. So, to come up with really perfect needles,” Zhavoronkov said. “[You] can essentially go to this [generative adversarial network] and ask it to create molecules that inhibit protein X at concentration Y, with the highest viability, specific characteristics, and minimal side effects.”
Zhavoronkov believes AI can find or fabricate more needles from the array of molecular possibilities, freeing human chemists to focus on synthesizing only the most promising. If it works, he hopes we can increase hits, minimize misses, and generally speed the process up.
Proof’s in the Pudding
Insilico isn’t alone on its drug-discovery quest, nor is it a brand new area of interest.
Last year, a Harvard group published a paper on an AI that similarly suggests drug candidates. The software trained on 250,000 drug-like molecules and used its experience to generate new molecules that blended existing drugs and made suggestions based on desired properties.
An MIT Technology Review article on the subject highlighted a few of the challenges such systems may still face. The results returned aren’t always meaningful or easy to synthesize in the lab, and the quality of these results, as always, is only as good as the data dined upon.
Stanford chemistry professor and Andreesen Horowitz partner, Vijay Pande, said that images, speech, and text—three of the areas deep learning’s made quick strides in—have better, cleaner data. Chemical data, on the other hand, is still being optimized for deep learning. Also, while there are public databases, much data still lives behind closed doors at private companies.
To overcome the challenges and prove their worth, Zhavoronkov said, his company is very focused on validating the tech. But this year, skepticism in the pharmaceutical industry seems to be easing into interest and investment.
AI drug discovery startup Exscientia inked a deal with Sanofi for $280 million and GlaxoSmithKline for $42 million. Insilico is also partnering with GlaxoSmithKline, and Numerate is working with Takeda Pharmaceutical. Even Google may jump in. According to an article in Nature outlining the field, the firm’s deep learning project, Google Brain, is growing its biosciences team, and industry watchers wouldn’t be surprised to see them target drug discovery.
With AI and the hardware running it advancing rapidly, the greatest potential may yet be ahead. Perhaps, one day, all 1060 molecules in drug-space will be at our disposal. “You should take all the data you have, build n new models, and search as much of that 1060 as possible” before every decision you make, Brandon Allgood, CTO at Numerate, told Nature.
Today’s projects need to live up to their promises, of course, but Zhavoronkov believes AI will have a big impact in the coming years, and now’s the time to integrate it. “If you are working for a pharma company, and you’re still thinking, ‘Okay, where is the proof?’ Once there is a proof, and once you can see it to believe it—it’s going to be too late,” he said.
Image Credit: Klavdiya Krinichnaya / Shutterstock.com Continue reading