Ai Traffic Guide: Fill & Download for Free

Let me tell you some Amazing tech facts to make your day!1)About 51% of internet users are not human!Not humans? Are you kidding me? Yes, you heard me right. Ok, let’s get to the point. It’s said that about 51% of internet traffic is caused by bots. Remember that weird “I am not a robot” checkbox that we get on many websites? These bots are the primary reason for that.Bonus for the lazy readers (just watch the video):2) Robots started to get citizenship!On October 25th, 2017, Sophia the robot was granted Saudi Arabian citizenship. Humans are struggling to be accepted as nationals in some parts of the world, and robots are starting to get it already!3) Most AI assistants are femaleHave you noticed that most AI assistants are females? From Siri and Google's assistant to Bixby, Cortana, and Alexa, most of them are females. Ever wonder why? It’s because studies have shown that people prefer female voices over male voices (both men and women prefer female voices).4) Your smartphone is a million times more powerful than the computers NASA used for launching Apollo 11The computers used to guide humans through space, about 356,000km away from earth, were a million times less powerful than the smartphone in your hand right now. Just imagine the potential we have got.5) The worlds most expensive food bought was 2 Papa John’s pizzasDon't believe me? Let me explain:The first bitcoin ever spent was used to buy 2 pizzas by a man named Hanyecz from Florida. He bought the pizza for 10,000 bitcoins, which was just 30$ then, but would be worth millions of dollars now.If you want to know more: 49 Fun facts about technology to blow your mind (2020)Every upvote will motivate me to write more interesting answers!THANK YOU

Questions like this are always a challenge to answer, but definitely worth the effort in trying to synthesize what one has learned over a lifetime of doing research in a field. So, what is my choice of the top 5 most important concepts in AI?Intelligence as search in problem spaces: Alan Newell was one of the most forceful advocates of the concept of search as a fundamental category of problem. His landmark work on SOAR, a cognitive architecture, was based on modeling every AI problem as that of search within some problem space. If we think of the major successes of AI recently — whether it be Alpha Go Zero at Go or IBM’s Jeopardy player or even the success at building self-driving cars — it is clear that Newell’s observation regarding the primacy of search remains as relevant today as it was several decades ago. My former thesis advisor, Tom Mitchell, wrote an influential article called “Generalization as Search”, showing how much of machine learning is once again reducible to search in some hypothesis space (or parameter space). Indeed, it is hard to think of an AI problem that does not involve some search. So, search shows up as the first on my list of the most important concepts in AI. There are many ways to model search: search in continuous spaces is basically the work on optimization, either the traditional convex kind as in methods like support vector machines, or the non-convex kind in deep learning. Discrete search problems are like constraint satisfaction problems (think of Sudoku) or search in games (chess, Go) or map search (e.g. A* in Google Maps).Knowledge as an effective mechanism to simplify search: Given that search is the primal category in AI, the major problem that needs to be addressed is how to make search more efficient. It goes without saying that “blind” search methods are unlikely to scale in any real problem. Search is tractable when guided by relevant knowledge. For example, in convex optimization, one exploits the knowledge that the function being minimized is “bowl-shaped” and has a unique minimum. Knowledge is also extremely useful in simplifying the space of policies or mappings being searched over in reinforcement learning. Consider the problem of training a self-driving car. Here, knowing the traffic rules greatly simplifies the problem of what policies are “legal”, and entire classes of illegal actions can be eliminated. Newell also introduced a foundational concept that unfortunately has been forgotten and needs to be reintroduced to the modern student of AI. Just like computer systems can be described at various levels — e.g. the hardware level, the software level etc. — Newell introduced the foundational concept of the “knowledge level” — characterizing an AI system by what it “knows”. This characterization is remarkable useful in providing a high level characterization of a system, abstracted from the details of how the knowledge is represented, stored, accessed and used. Sadly, the concept of knowledge level is seldom used anymore, but in my mind, it continues to provide one of the most important ways to distinguish AI systems from other smart systems. AI systems are intelligent to the extent they “know” things about the world, and can act rationally given their knowledge. So, knowledge level characterization of an AI system is the second most important concept in AI.Representation and tractable inference: it has become evident over several decades of research in AI that the form of knowledge — the representation — plays a crucial role in determining how efficient inference will be in using the knowledge to guide decisions. Numerous results — from the theory of PAC learning in computational learning theory to work in graphical models for probabilistic inference and work on logical inference — show that there is a fundamental tradeoff between expressivity and tractability. The more expressive a knowledge representation scheme is, the less tractable it is, and this imposes a fundamental barrier in building efficient AI systems. Take the simple problem of learning boolean functions from examples. It turns out that it is extremely difficult, if not impossible, to efficiently learn any boolean function from a relatively small number of examples reliably, but if one limits the boolean function to some smaller class — say conjunctive boolean expressions where each “clause” is limited to a disjunction of at most “k” literals — efficient algorithms exist. Thus, representations play a foundational role in determining whether a given knowledge structure can be learned, and can be used efficiently to reduce search. Ultimately, many basic questions about representation in AI reduce to the ultimate P=NP? question that has bedeviled computer scientists for over four decades. In mathematics, representations play a key role in understanding fundamental structures, such as linear transformations or symmetries. To understand a linear transformation, one maps it to a matrix, which is a representation of that transformation in some basis. To understand rotations in six dimensions, one maps it to a group of a particular type, which has a matrix representation as well. The theory of representations thus forms the third most important concept in AI.Optimization vs. equilibration: in giving AI systems goals, a natural tendency is to expect them to be capable of finding an “optimal” solution with respect to some loss or utility function. We want our self-driving car to perform optimally according to some set of metrics. Herein lies the rub, as the saying goes. Most real-world problems involve trading off a set of mutually incompatible metrics. A self-driving car that optimizes safety might not optimize other metrics, such as getting passengers to their destinations on time. Looking at the process of natural selection, it is clear that biology favors the process of equilibration — finding equilibrium solutions — rather than optimization. If I am trying to decide what the best route is from say San Francisco to Palo Alto every morning, it is clear that I have control over only a small number of variables, such as whether I drive or take the Caltrain, and if I drive, which highway I choose — Route 1 or 280 — but I have no control over the tens of thousands of other drivers who are pursuing their own self-serving driving goals of getting to work. At best, I can try to equilibrate, and find a solution such that it forms a “local” Nash type of equilibrium, so that there is no local improvement to my policy as long as the other drivers don’t deviate from their choices. Nash equilibria seem fundamental to how to design multi agent AI systems, since optimization seems a hopelessly ideal metric that cannot be achieved in a real world problem. So, my fourth most important concept is that of finding equilibrium solutions.Distributivity vs. locality and fault tolerance: clearly, if we are to build robust fault tolerant AI systems, they must have some inherent capacity to withstand failure of individual components. The human brain is nonpareil in this regard, as even patients with severe brain injuries are able to compensate for their losses and recover almost full functionality. So, any knowledge representation system that is ultimately successful in AI must be similarly capable of smooth deterioration, where loss of individual components does not render the whole system inoperable. Unlike most modern computer systems, where the loss of a single sector can render a hard drive unreadable sometimes, AI systems need knowledge stored in a highly redundant manner, so that knowledge can be reconstructed in a fault tolerant way, much as human memory is able to reconstruct events. The requirement of fault tolerance leads inexorably to distributed representations and inference, and ultimately to neurally inspired models of AI, where many simple computing elements are combined to produce intelligent behavior. So, my fifth most important concept in AI would be the design of parallel distributed knowledge-based systems that can function in a failsafe fault tolerant manner, much like the human brain.OK, there’s my list of the five most important concepts in AI, and why they represent my top 5 choices. Hope you find it useful!

Ask yourself what free will is.OK, let me give you an example. Say, you come up on a set of traffic lights at an intersection. It is an old set of traffic lights, governed by a simple electromechanical timer. Does it have free will? Of course not. It’s just a glorified conventional alarm clock.So let’s say that at the next intersection, you find a newer set of traffic lights. It is controlled by more sophisticated electronics that takes into account the time of day and even sensors in the road, determining, e.g., whether to allow left turns during a cycle. Does this set of lights have free will? I’m guessing you would say no.But then, at the next intersection, you come across a still newer set of lights. It has sophisticated control electronics that is networked. It coordinates its behavior with other traffic lights in the neighborhood. It is connected to a network of sensors and cameras that are used to estimate traffic flow. It even recognizes emergency vehicles, adjusting its behavior. Moreover, let’s say, it has a simple learning capability, so that it can adaptively adjust its cycle to minimize wait times and maximize vehicle throughput, again in coordination with other traffic lights. Does this set of lights have free will? Again, I am guessing you would say no.So let me take it one step further. The next set of lights is straight from the future. It is governed by an artificial intelligence. The AI, annoyed and bored like Marvin the Paranoid Android from the Hitchhiker’s Guide to the Galaxy, nonetheless performs its job admirably. He may be using his spare brain capacity to study the limits of quantum theory or analyze 19th century French literature, but ultimately, he is a deterministic machine: every bit of his programming, every logic gate in its considerable brain follows a predetermined pattern, even if the overall complexity makes the machine’s behavior practically unpredictable. Does this machine have free will, in your opinion?Because if you say it doesn’t, I have to ask: What exactly does it take for an entity to have free will? Is it the lack of deterministic behavior?So then, going back to the first set of traffic lights, with its simple electromechanical timer: If I added to this setup a random number generator that causes the traffic lights to behave unpredictably, would you say that it suddenly acquired free will? Now that would be silly, wouldn’t it. But doesn’t it demonstrate that it is not the lack of determinism, not the absence of causality that is the secret of free will?In my mind, that AI machine that decided to spend its spare time reading Victor Hugo absolutely qualifies as an entity with free will. Sure, it is deterministic. Every one of its components behaves in a deterministic fashion. But the system as a whole is so complex, its behavior cannot be predicted: Not unless you build an identical copy or simulation, and subject it to the exact same set of stimuli, so that it forms the same memories and responds in the same manner as the original.But who says that if I had the means to build an identical copy or simulation of you and subjected it to the exact same stimuli that form your life experience, it would not develop the exact same identity that you call your own? Yet I presume you believe you have free will (I certainly believe that I have free will myself.)The bottom line: I think it is wrong to present free will as an opposite of determinism. The opposite of determinism is randomness. A sophisticated machine may, in turn, respond to stimuli in novel ways based on its internal state (accumulated experience), demonstrating every apparent aspect of what we call free will, despite the fact that ultimately, its behavior is fully deterministic.