Mediaboss Marketing

We are excited to bring Transform 2022 back in-person July 19 and virtually July 20 - 28. Join AI and data leaders for insightful talks and exciting networking opportunities. Register today!

2016 saw the completion of a significant milestone for humanity: artificial intelligence (AI) beat the world champion in the Go game. For context, Go is a board game previously thought to require too much human intuition for a computer to succeed in, and as a result, it was a North Star for AI.

For years, researchers tried and failed to create an AI system that could beat humans in the game. Until AlphaGo.

In 2016, AlphaGo, an AI system created by Googles DeepMind, not only beat its champion human counterpart (Lee Sedol); it demonstrated that machines could find playing strategies that no human would come up with. AlphaGo shocked the world when it performed its unimaginable move #37. It was a move so counterintuitive and strange to human experts that after AlphaGo played it, it stunned and perplexed Lee and all the onlookers and world experts. It ultimately led to the technologys triumph during that game.

Beyond exemplifying AIs potential in this context, the Go game demonstrated that AI could and should help humanity come up with the Move 37 for significant, real-world problems. Among these include fighting cancer.

Like board games, there is a particular element of a game in the proverbial contest between the human immune system and cancer. If the immune system is the policeman guarding the health of the body, cancer is like a mobster that is trying to elude capture. While the immune system police search for harmful cancer cells, viruses, infections and any disorders, cancer is busy coming up with various tactics of subversion, deceit and destruction.

Centuries ago, scientists and doctors operated largely in the dark when attempting to cure diseases and had to rely solely on their intuition. Today, however, humanity is uniquely positioned to fully utilize available resources with advancements in high throughput and measurement of biological data. We can now create AI models and use every bit of available data to allow these AIs to augment our innate intuition.

To illustrate this concept more clearly, consider the case of CAR-T cells edited with CRISPR (a genetic editing technology) to create a promising therapeutic option in treating cancer. Many current and past approaches in the field relied on a single researcher or academic groups intuition for prioritizing which genes to test edit. For example, some of the worlds experts in genetically engineered T cells came up with the idea of trying to knock out the PD1, which did not play out to improve patient outcomes. In this case, genes were not compared head-to-head, and a lot of human intuition was required to decide how to best proceed.

Recently, with advances in high-throughput single-cell CRISPR sequencing methods, we are nearing the possibility of simply testing all genes simultaneously on equal footing and in various experimental scenarios. This makes the data a better fit for AI and, in this case, we have the opportunity of letting AI help us decide on which genes look most promising to modify in patients to fight their cancer.

The ability to run extensive AI experiments and generate data for fighting cancer is a game-changer. Biology and disease are so complex that it is improbable that current and past strategies, driven largely by human intuition, are the best approaches. In fact, we predict that in the next 10 years, we will have an equivalent of a Move 37 against cancer: a therapy that at first may seem counterintuitive (and at which human intuition alone would not arrive) but that in the end, shocks us all and wins the game for patients.

Luis Voloch is CTO and cofounder of Immunai.

Welcome to the VentureBeat community!

DataDecisionMakers is where experts, including the technical people doing data work, can share data-related insights and innovation.

If you want to read about cutting-edge ideas and up-to-date information, best practices, and the future of data and data tech, join us at DataDecisionMakers.

You might even considercontributing an articleof your own!

Read More From DataDecisionMakers

More:
How AI and human intelligence will beat cancer - VentureBeat

Odds and Evens: Split.

Hard to go past progressive three-year-old 3. Mojo Classic who roared home from the back to claim his maiden fourth-up and four weeks between runs. Races like he will eat up the extra trip, especially sticking to a big track, and hes bred to thrive on rain-affected ground.Dangers: Stablemate filly 4. Stella Glow is also on the rise having notched her maiden win in similar ground third-up as a well backed favourite, and comes through some handy form lines. Keep safe 2. Duble Memory who surged home from a wide draw to win a class 1 in heavy ground third-up, while 1. Leica Bita Fun fourth-up and honest 5. Thailand, who draws inside, are both capable of running into the minor end of the money.How to play it: Mojo Classic win; quinella 3 and 4.Odds and Evens: Split.

The girls lock horns in a tricky sprint, with several first-starters who are likely to have a big impact on the market. One of them, home-track Teofilo three-year-old 4. Mirrie Dancer, can make an instant statement. Liked the way she worked home strong from a mile back in heavy ground in the latest of two trials, and she looks well prepared for this distance. Drawn wide, but that might be too her pattern advantage.Dangers: Provincial three-year-old 5. Oakfield Redgum returns for only a second start behind a steady trial, and draws to get cover. Big watch on debutant Nicconi three-year-old 3. Golden Gate who has been taken along slowly at the trials. Liked the way hes performed in two hit-outs, the latest slow to begin and not settling early before working home well in open company, and is bred to handle the conditions.

How to play it: Mirrie Dancer each way.

Odds and Evens: Split.

Loading

Like provincial seven-year-old 4. Emperor Harada on suitable heavy ground.Dangers: Metro six-year-old 1. Skyray with multiple gear changes third-up is the clear threat in what is now a very thin affair.How to play it: Emperor Harada win; quinella 1 and 4.

Odds and Evens: Split.

Lonhro three-year-old 3. Hotstep debuts behind two forward trials on rain-affected ground. Trial jockey sticks for the real thing, and significantly he wears blinkers in a race thats down to four runners.Dangers: Another first starter in blinkers, 2. Beer Palace has had three recent trials and is the clear threat.

How to play it: Hotstep win; quinella 2 and 3.

Best Bets: Race 4 (3) Mojo Classic, Race 7 (3) Hotstep.Best Value: Race 5 (4) Mirrie Dancer.

Tips supplied by Racing NSWFull form and race replays available at racingnsw.com.au.

News, results and expert analysis from the weekend of sport sent every Monday. Sign up for our Sport newsletter.

Original post:
Race-by-race tips and preview for Newcastle on Monday - Sydney Morning Herald

Image credit: 123RF (with modifications)

Reinforcement learning is one of the exciting branches of artificial intelligence. It plays an important role in game-playing AI systems, modern robots, chip-design systems, and other applications.

There are many different types of reinforcement learning algorithms, but two main categories are model-based and model-free RL. They are both inspired by our understanding of learning in humans and animals.

Nearly every book on reinforcement learning contains a chapter that explains the differences between model-free and model-based reinforcement learning. But seldom are the biological and evolutionary precedents discussed in books about reinforcement learning algorithms for computers.

I found a very interesting explanation of model-free and model-based RL in The Birth of Intelligence, a book that explores the evolution of intelligence. In a conversation with TechTalks, Daeyeol Lee, neuroscientist and author of The Birth of Intelligence, discussed different modes of reinforcement learning in humans and animals, AI and natural intelligence, and future directions of research.

In the late nineteenth century, psychologist Edward Thorndike proposed the law of effect, which states that actions with positive effects in a particular situation become more likely to occur again in that situation, and responses that produce negative effects become less likely to occur in the future.

Thorndike explored the law of effect with an experiment in which he placed a cat inside a puzzle box and measured the time it took for the cat to escape it. To escape, the cat had to manipulate a series of gadgets such as strings and levers. Thorndike observed that as the cat interacted with the puzzle box, it learned the behavioral responses that could help it escape. Over time, the cat became faster and faster at escaping the box. Thorndike concluded that the cat learned from the reward and punishments that its actions provided.

The law of effect later paved the way for behaviorism, a branch of psychology that tries to explain human and animal behavior in terms of stimuli and responses.

The law of effect is also the basis for model-free reinforcement learning. In model-free reinforcement learning, an agent perceives the world, takes an action, and measures the reward. The agent usually starts by taking random actions and gradually repeats those that are associated with more rewards.

You basically look at the state of the world, a snapshot of what the world looks like, and then you take an action. Afterward, you increase or decrease the probability of taking the same action in the given situation depending on its outcome, Lee said. Thats basically what model-free reinforcement learning is. The simplest thing you can imagine.

In model-free reinforcement learning, theres no direct knowledge or model of the world. The RL agent must directly experience every outcome of each action through trial and error.

Thorndikes law of effect was prevalent until the 1930s, when Edward Tolman, another psychologist, discovered an important insight while exploring how fast rats could learn to navigate mazes. During his experiments, Tolman realized that animals could learn things about their environment without reinforcement.

For example, when a rat is let loose in a maze, it will freely explore the tunnels and gradually learn the structure of the environment. If the same rat is later reintroduced to the same environment and is provided with a reinforcement signal, such as finding food or searching for the exit, it can reach its goal much quicker than animals who did not have the opportunity to explore the maze. Tolman called this latent learning.

Latent learning enables animals and humans to develop a mental representation of their world and simulate hypothetical scenarios in their minds and predict the outcome. This is also the basis of model-based reinforcement learning.

In model-based reinforcement learning, you develop a model of the world. In terms of computer science, its a transition probability, how the world goes from one state to another state depending on what kind of action you produce in it, Lee said. When youre in a given situation where youve already learned the model of the environment previously, youll do a mental simulation. Youll basically search through the model youve acquired in your brain and try to see what kind of outcome would occur if you take a particular series of actions. And when you find the path of actions that will get you to the goal that you want, youll start taking those actions physically.

The main benefit of model-based reinforcement learning is that it obviates the need for the agent to undergo trial-and-error in its environment. For example, if you hear about an accident that has blocked the road you usually take to work, model-based RL will allow you to do a mental simulation of alternative routes and change your path. With model-free reinforcement learning, the new information would not be of any use to you. You would proceed as usual until you reached the accident scene, and then you would start updating your value function and start exploring other actions.

Model-based reinforcement learning has especially been successful in developing AI systems that can master board games such as chess and Go, where the environment is deterministic.

In some cases, creating a decent model of the environment is either not possible or too difficult. And model-based reinforcement learning can potentially be very time-consuming, which can prove to be dangerous or even fatal in time-sensitive situations.

Computationally, model-based reinforcement learning is a lot more elaborate. You have to acquire the model, do the mental simulation, and you have to find the trajectory in your neural processes and then take the action, Lee said.

Lee added, however, that model-based reinforcement learning does not necessarily have to be more complicated than model-free RL.

What determines the complexity of model-free RL is all the possible combinations of stimulus set and action set, he said. As you have more and more states of the world or sensor representation, the pairs that youre going to have to learn between states and actions are going to increase. Therefore, even though the idea is simple, if there are many states and those states are mapped to different actions, youll need a lot of memory.

On the contrary, in model-based reinforcement learning, the complexity will depend on the model you build. If the environment is really complicated but can be modeled with a relatively simple model that can be acquired quickly, then the simulation would be much simpler and cost-efficient.

And if the environment tends to change relatively frequently, then rather than trying to relearn the stimulus-action pair associations whenever the world changes, you can have a much more efficient outcome if youre using model-based reinforcement learning, Lee said.

Basically, neither model-based nor model-free reinforcement learning is a perfect solution. And wherever you see a reinforcement learning system tackling a complicated problem, theres a likely chance that it is using both model-based and model-free RLand possibly more forms of learning.

Research in neuroscience shows that humans and animals have multiple forms of learning, and the brain constantly switches between these modes depending on the certainty it has on them at any given moment.

If the model-free RL is working really well and it is accurately predicting the reward all the time, that means theres less uncertainty with model-free and youre going to use it more, Lee said. And on the contrary, if you have a really accurate model of the world and you can do the mental simulations of whats going to happen every moment of time, then youre more likely to use model-based RL.

In recent years, there has been growing interest in creating AI systems that combine multiple modes of reinforcement learning. Recent research by scientists at UC San Diego shows that combining model-free and model-based reinforcement learning achieves superior performance in control tasks.

If you look at a complicated algorithm like AlphaGo, it has elements of both model-free and model-based RL, Lee said. It learns the state values based on board configurations, and that is basically model-free RL, because youre trying values depending on where all the stones are. But it also does forward search, which is model-based.

But despite remarkable achievements, progress in reinforcement learning is still slow. As soon as RL models are faced with complex and unpredictable environments, their performance starts to degrade. For example, creating a reinforcement learning system that played Dota 2 at championship level required tens of thousands of hours of training, a feat that is physically impossible for humans. Other tasks such as robotic hand manipulation also require huge amounts of training and trial-and-error.

Part of the reason reinforcement learning still struggles with efficiency is the gap remaining in our knowledge of learning in humans and animals. And we have much more than just model-free and model-based reinforcement learning, Lee believes.

I think our brain is a pandemonium of learning algorithms that have evolved to handle many different situations, he said.

In addition to constantly switching between these modes of learning, the brain manages to maintain and update them all the time, even when they are not actively involved in decision-making.

When you have multiple learning algorithms, they become useless if you turn some of them off. Even if youre relying on one algorithmsay model-free RLthe other algorithms must continue to run. I still have to update my world model rather than keep it frozen because if I dont, several hours later, when I realize that I need to switch to the model-based RL, it will be obsolete, Lee said.

Some interesting work in AI research shows how this might work. A recent technique inspired by psychologist Daniel Kahnemans System 1 and System 2 thinking shows that maintaining different learning modules and updating them in parallel helps improve the efficiency and accuracy of AI systems.

Another thing that we still have to figure out is how to apply the right inductive biases in our AI systems to make sure they learn the right things in a cost-efficient way. Billions of years of evolution have provided humans and animals with the inductive biases needed to learn efficiently and with as little data as possible.

The information that we get from the environment is very sparse. And using that information, we have to generalize. The reason is that the brain has inductive biases and has biases that can generalize from a small set of examples. That is the product of evolution, and a lot of neuroscientists are getting more interested in this, Lee said.

However, while inductive biases might be easy to understand for an object recognition task, they become a lot more complicated for abstract problems such as building social relationships.

The idea of inductive bias is quite universal and applies not just to perception and object recognition but to all kinds of problems that an intelligent being has to deal with, Lee said. And I think that is in a way orthogonal to the model-based and model-free distinction because its about how to build an efficient model of the complex structure based on a few observations. Theres a lot more that we need to understand.

Go here to read the rest:
A gentle introduction to model-free and model-based reinforcement learning - TechTalks

We are survival machines robot vehicles blindly programmed to preserve the selfish molecules known as genes. This is a truth which still fills me with astonishment.

Ancient Greeks imagined their gods to be capable of building robots. Hundreds of years have passed since the collapse of the greatest civilization, yet the human pursuit of developing something in their own image continues unabated. Thanks to the advances in AI, we now have the means to create human-like robots.

In his book, Human Natures: Genes Cultures and the Human Prospect, Paul Ehrlich said the concept of religion first appeared when humans developed brains large enough for abstract thought. On the other hand, artificial intelligence is brand new but pervasive.

Science and religion rarely see eye to eye, except for occasional outliers like the artificial intelligence church. Even though the church is closed now, it does pose a critical question, how will religion react to a sentient machine with free will?

Interestingly, some cultures are more open to technology than others. For example, a 400-year-old Buddhist temple in Kyoto, Japan, called Kodaji caught the public imagination last year when it announced a new clergy member, a robot priest that performs sermons. Called Mindar, the USD 1 million robot was designed to look like Kannon, the Buddhist deity of mercy. The idea behind the priest robot was to rekindle peoples faith. Meanwhile, SoftBanks humanoid robot Pepper is available for hire as a Buddhist priest for funerals.

At the World Economic Forum in Davos in 2018, Pope Francis said AI, robotics, and other innovations should be used to serve humanity and protect our common home.

These are examples of religion leveraging modern-day technology. But, what happens when AI hits singularity, or we achieve AGI? What happens when we create something in our own image?

What happens when an AI robot with the same intellectual ability as a human makes its own decisions? Should this machine be considered a human? Does it have a soul?

Religion argues that God has a plan for everybody. Religion also tells people how they should lead their life. So, where does this AI robot fit in? Herein lies the rub: Technology has the power to shake the foundations of religion.

In his 1950 paper Computing Machinery and Intelligence, Alan Turing, the founding father of AI, said: Thinking is a function of mans immortal soul. God has given an immortal soul to every man and woman, but not to any other animal or machines. Hence no animal or machine can think. Plato believed that the soul was both the source of life and the mind.

Different faiths react differently to technology. For example, last year, a top religious body in Indonesia forbade cryptocurrencies under Islamic law. Also, it can be argued that faith is very personal, so individuals will respond differently to AI.

AGI is the north star of companies like OpenAI, DeepMind and AI2. While OpenAIs mission is to be the first to build a machine with human-like reasoning abilities, DeepMinds motto is to solve intelligence.

DeepMinds AlphaGo is one of the biggest success stories in AI. In a six-day challenge in 2016, the computer programme defeated the worlds greatest Go player Lee Sedol. DeepMinds latest model, Gato, is a multi-modal, multi-task, multi-embodiment generalist agent. Googles 2021 model, GLaM, can perform tasks like open domain question answering, common-sense reading, in-context reading comprehension, the SuperGLUE tasks and natural language inference.

OpenAIs DALL.E2 blew minds just a few months ago with imaginative renderings based on text inputs. Yet, all these achievements are pale compared to the intelligence of the human child.

However, people from the AI/ML community believe that AGI is achievable. Recently, Elon Musk, who has invested a lot in AI over the years, tweeted that he would be surprised if we do not achieve AGI by 2029.

Meanwhile, a popular critic of deep learning and AGI Gary Marcus said current AI is illiterate in an interview. It can fake its way through, but it doesnt understand what it reads. So the idea that all of those things will change on one day and on that magical day, machines will be smarter than peopleis a gross oversimplification, he said.

When Anthony Levandowski established the first church of artificial intelligence, called Way of the Future, it raised a few eyebrows.

If we achieve AGI in the coming years, it will change many things. When technology becomes far superior, and these artificial beings can do things beyond a human being, there are chances that people will associate them with a higher power. For example, AI in its current form is already aiding scientists in drug discovery. Humans are always on the lookout for God, even if on a personal level.

But one of the worst possible outcomes would be that AI emerges as a polarising factor. Hence, it warrants a greater discussion in this context. There is also an increasing need for the larger participation of religious groups in deliberating the ethical implications of AI development.

Read the original post:
The role of 'God' in the 'Matrix' - Analytics India Magazine

As an observer of artificial intelligence over the last few years at DSAITrends, it is fascinating to observe the dichotomy between the sheer amount of research and development in AI, and its glacial real-world impact.

No doubt, we do have plenty of jaw-dropping developments from AI-synthesized faces that are indistinguishable from real faces, AI models that can explain jokes, and the ability to create original, realistic images and art from text descriptions.

But this has not translated into business benefits for more than a handful of top tech firms. For the most part, businesses are still wrestling with their board about whether to implement AI or struggling to operationalize AI.

In the meantime, ethical quandaries are as yet unresolved, bias is rampant, and at least one regulator has warned banks about the use of AI.

One popular business quote comes to mind: We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.

So yes, while immediate AI gains seem lacking, the impact of AI in the long term might yet exceed our wildest expectations. And new, powerful AI hardware could well accelerate AI developments.

More powerful AI hardware

But why the fascination with more powerful hardware? In the groundbreaking Scaling laws for neural language models paper published in 2020, researchers from OpenAI concluded that larger AI models will continue to perform better and be much more sample efficient than previously appreciated.

While the researchers cautioned that more work is needed to test if the scaling holds, the current hypothesis is that more powerful AI hardware could train much larger models that will yield capabilities far beyond todays AI model.

Leading the charge on the hardware front would be data center-class GPUs from NVIDIA and AMD, as well as specialized AI processors from technology giants such as Google. For example:

Stepping outside the box

There are research fields that could impact the development of AI, too. For example, the Loihi 2 is a second-generation experimental neuromorphic chip by Intel. Announced last year, a neuromorphic processor mimics the human brain using programmable components to simulate neurons.

According to its technical brief (pdf), the Loihi 2 has 128 cores and has potentially more than a million digital neurons due to its asynchronous design. The human brain does have roughly 90 billion interconnected neurons, so there is still some way to go yet.

Chips like the Loihi 2 has another advantage though. As noted by a report on The Register, high-end AI systems such as DeepMinds AlphaGo require thousands of processing units running in parallel, with each consuming around 200 watts. Thats a lot of power and we havent even factored in the ancillary systems or cooling equipment yet.

On its part, neuromorphic hardware promises between four and 16 times better energy efficiency than other AI models running on conventional hardware.

Warp speed ahead with quantum computing

While the Loihi 2 is made of traditional transistors there are 2.3 billion of them in the Loihi 2 another race is underway to make a completely different type of computer known as quantum computers.

According to a report on AIMultiple, quantum computing can be used for the rapid training of machine learning models and to create optimized algorithms. Of course, it must be pointed out that quantum computers are far more complex to build due to the special materials and operating environments required to access the requisite quantum states.

Indeed, experts estimate that it could take another two decades to produce a general quantum computer, though working quantum computers of up to 127-qubit exists.

In Southeast Asia, Singapore is stepping up its investments in quantum computing with new initiatives to boost talent development and provide access to the technology. This includes a foundry to develop the components and materials needed to build quantum computers.

Whatever the future brings for AI in the decades ahead, it will not be for lack of computing prowess.

Paul Mah is the editor of DSAITrends. A former system administrator, programmer, and IT lecturer, he enjoys writing both code and prose. You can reach him at [emailprotected].

Image credit: iStockphoto/jiefeng jiang

See the original post:
The Powerful New AI Hardware of the Future - CDOTrends