It turns out that video games can be a great way to teach the skills to artificial intelligence assistants. This is the theory of a group of researchers working for Facebook, who focused on Minecraft as a potential pedagogical tool to build the generalist IA – a so-called "Virtual assistant". The research team does not try to build an artificial intelligence that is super-good at classifying images or any other content – it wants to create a generalist AI that can do a lot more tasks.
This is currently an under-researched area of research. The writers' writings:
Measured progress was also recorded in this context, with the integration of virtual personal assistants. These are able to accomplish thousands of tasks communicated via natural language, using a multi-turn dialog to clarify or clarify. Wizards can interact with other applications to obtain data or perform actions.
However, many difficult issues remain unresolved. Automatic language comprehension (NLU) is always rigid and limited to limited scenarios. The methods of using the dialog or other natural language for enhanced supervision remain primitive. In addition, since they must be able to reliably and predictably solve many simple tasks, their multimodal inputs and the constraints of their maintenance and deployment, the wizards are modular systems, as opposed to monolithic ML models. Modular ML systems able to improve from data while maintaining well-defined interfaces are not yet well studied
According to the team, Minecraft was chosen because it offered a regular distribution of tasks with "manual manipulations for NLU research", as well as pleasant opportunities for human-intelligence interaction, as well as many opportunities for human research. . Minecraft, for those who have never played there or have never heard of it, is a block-based game of exploration and exploration in which players explore a universe. 3D voxel grid populated with various types of materials, neutral characters and enemies. The goal of the team is to create a virtual assistant in artificial intelligence who can receive instructions in natural language from a Minecraft player, and who can reliably perform some of the main tasks that the player could engage, including the collection of materials, the construction of structures, handicrafts.
The authors of the paper aim at three specific achievements: creating a synergy between the components of machine learning and not machine learning, allowing them to work together; create an "ingrained" natural language simulation that allows the AI to understand what the players want, and communicate its success or failure to the end user; and create an AI that should not just be able to do what the player wants, but that his performance should also improve based on the observation of the human player.
We want the player to be able to specify tasks by dialog (rather than by simple command) so that the agent can request the missing information or the player can interrupt the agent's actions to clarify. In addition, we hope the dialogue will be useful for providing rich supervision. The player can label attributes related to the environment, for example "this house is too big", relationships between objects of the environment (or other concepts understood by the bot), by example "the window is in the middle of the wall" or rules. about these relationships or attributes. We expect the player to be able to question the mental state of the agent to give the appropriate feedback, and we expect the bot to ask for confirmation and use active learning strategies.
The machine learning code used for the Facebook-Minecraft bot is available on GitHub. Better artificial intelligence tools could be useful in many games, although they can also raise serious questions about what constitutes a multiplayer cheat.
Humans tend to bury objects that they consider most important for their lives. It is not surprising that some people have chosen to buy games, including board games. Some of these artifacts have survived the centuries, leaving modern archaeologists a puzzle. How do you understand a game when you do not know how it was played? The ancient humans, it seems, have been terribly terrible to lose the textbooks for their canned titles.
Tracking the evolution of games and games could teach us a lot about cultural exchange and evolution between two companies or within a company over time. In a few cases, archaeologists have been lucky and have discovered the rules of the game. More often, they are stuck trying to figure out how a game might have worked by comparing it to other games we have access to. make understand or analyze how a title is described in the art. The only clues to playing the old Egyptian game Senet, for example, are in the tomb of Queen Nefertari.
Now, historians are using new tools to recover old game rules. Cameron Browne runs the Digital Ludeme Project. This is a research project aimed at using computational techniques, including AI, to recreate the rules of the board games. To do this, Browne and his team break down a game into building blocks, coding them into units called "ludemas". A ludem is any game element or rule known to researchers. Cultural information is also incorporated to help assess the plausibility of various rule sets evaluated. The system they built to play games is called Ludii.
Modern AI techniques are used to create potential rule sets for games, which can then be evaluated by calculation. Artificial intelligence agents are assigned to games using the proposed rules and setting up movement lists. The data collected by the artificial intelligence agents during several phases of play can then be evaluated to determine if the final result is a playable game.
"With our system, we can install the equipment, we can search for rules from the date (of the creation of the council), rules of the zone, rules of the cultural context", Browne said Vice "Then we will be able to gather the sets of rules that are likely to occur."
The team has already been successful. This article describe a game called latrunculi or Ludus latrunculorum, which scholars have struggled to decipher. Other specialists have come up with their own rules, but Browne thinks that Ludii has defined a more likely set of rules for the game – and you can actually l & # 39; try, if you use the beta version of the application and select the appropriate game.
Train the AI dataset on the rules of thousands of known games and explain to AI how the rules of play have evolved and have been transmitted over time in situations that we can already trace could help to teach AI how to trace the most likely routes of cultural transmission in the past Some games, like latrunculi, have been played for centuries throughout the Roman Empire.
Our difficulty in understanding the games of the past sheds an interesting light on the battle to preserve games. in the present. Efforts to preserve game data in machine-readable form and to ensure that unique gaming experiences are not lost in the face of the ravages of time are not simply motivated by hacking or foolish self-sacrifice. nerd-dom. Individuals who have preserved the games of humanity centuries ago were keeping a precious relic of their time. Knowing how a company chooses to spend their free time and the types of games they find fun tells us about the company, regardless of the type of entertainment.
Featured Image courtesy of Wikipedia
Intel may have launched Cascade Lake relatively recently, but another refresh of the 14-nm server is already on the horizon. Intel has lifted the veil on Cooper Lake today, giving new details on how the processor integrates into its product line with the 10-nm Ice Lake server chips supposed to be queuing for the deployment in 2020.
Cooper Lake features include support for Google's bfloat16 format. It will also support up to 56 processor cores in a snap-in format, unlike Cascade Lake-AP, which can scale up to 56 cores but only in a welded BGA configuration. The new take would be known as LGA4189. There is reports that these chips could offer up to 16 channels of memory (since Cascade Lake-AP and Cooper Lake use multiple chips on the same chip, Intel could run up to 16 channels of memory per socket with version double chip).
Bfloat16 support is a major addition to Intel's artificial intelligence efforts. While 16-bit semi-precision floating point numbers have been defined in the IEEE 754 standard for over 30 years, bfloat16 changes the balance between the format used for significant digits and that used for exponents. The original IEEE 754 standard is designed to give priority to precision, with only five bits of exponent. The new format allows a much larger range of values but with less precision. This is particularly useful for artificial intelligence and deep learning calculations, and is a major step on Intel's path to improving the performance of artificial intelligence and deep processor learning computations. Intel has released a White Book on bfloat16 if you are looking for more information on the subject. Google says that using bfloat16 instead of the conventional semi-precision floating point can generate significant performance benefits. The society written"Some operations are related to the memory bandwidth, which means that the memory bandwidth determines the time spent in such operations. Storing the inputs and outputs of memory bandwidth-related operations in bfloat16 format reduces the amount of data to be transferred, improving the speed of operations. "
The other benefit of Cooper Lake is that the CPU would share a socket with the upcoming Ice Lake servers in 2020. A theoretically important distinction between the two families is that Ice Lake servers at 10 nm can not support bfloat16, while 14nm Cooper Lake servers will. This could be the result of increased differentiation of Intel's product lines, although it is also possible that this reflects the difficult development of 10 nm.
The introduction of 56 cores as a base indicates that Intel expects Cooper Lake to expand to more customers than the Cascade Lake / Cascade Lake-AP target number. It also raises questions about the type of Ice Lake servers that Intel is going to market and the possibility of seeing 56-core versions of these chips as well. To date, all of Intel's 10-nm Ice Lake messaging has focused on servers or mobile devices. This may reflect the strategy used by Intel for Broadwell, where desktop versions of the processor were scarce, and where server and server components dominated this family – but Intel says later the fact of not publishing Broadwell desktop was a mistake and that the company had gaffed by skipping the market. Does this mean that Intel keeps launching an Ice Lake desktop or if the company has decided to no longer use its desktop computer? made understand that this time is not yet clear.
Cooper Lake's attention to AI treatment means that it is not necessarily meant to go with AMD's next 7 nm Epyc. AMD has not talked much about AI or machine learning on its processors and, although its 7nm chips add support for 256-bit AVX2 operations, the company's CPU division does not tell us has not yet hinted that a particular goal is the AI market. AMD's efforts in this area are still based on a graphics processor and, although its processors will certainly work with AI code, it does not seem that the market is at the same level as that of Intel. Between the addition of a new support for AI to existing Xeons, its products Movidius and Nervana, projects like Loihiand plans the data center market with Xe, Intel is trying to build a market to protect its high-performance computing and high-end server operations, and to address Nvidia's current dominance of the industry.
In recent years, Intel has been talking about its Cascade Lake servers with DL Boost (also called VNNI, Vector Neural Net Instructions). These new features are a subset of the AVX-512 and are intended to specifically accelerate processor performance in artificial intelligence applications. Historically, many AI applications have favored GPUs over processors. The GPU architecture is much better suited to graphics processors than processors. Processors offer many more thread-based execution resources, but even today's multicore processors are overwhelmed by the parallelism available in a high-end graphics processor core.
Anandtech did you compare the performance of Cascade Lake, the Epyc 7601 (soon outperformed by the AMD Rome 7nm processors, but still today AMD's main server core), and a RTX Titan. The article, written by the excellent Johan De Gelas, discusses different types of neural networks beyond CNN networks (convolutional neural networks), which are generally compared, and explains how a key element of the strategy Intel is competing with Nvidia in workloads where GPUs are not as powerful. or can not yet meet the emerging market needs due to memory capacity constraints (GPUs still can not match CPUs here), the use of 'light' artificial intelligence models does not not requiring long periods of workout, or artificial intelligence models that rely on statistical models of non-neural networks.
Growing revenue from data centers is a critical part of Intel's global strategy for artificial intelligence and machine learning. Nvidia, meanwhile, is keen to protect a market in which it currently has virtually no competition. Intel's Artificial Intelligence strategy is broad and encompasses many products, from Movidius and Nervana to DL Boost on Xeon, to the next GPU Xe range. Nvidia seeks to show that GPUs can be used to handle artificial intelligence calculations in a wider range of workloads. Intel incorporates new artificial intelligence features into its existing products, uses new hardware that, it is hoped, will impact the market, and attempts to create its first serious GPU to challenge the work done by AMD and Nvidia in the consumer market.
Anandtech's benchmarks show, overall, that the gap between Intel and Nvidia remains wide, even with DL Boost. This graph of a recurrent neural network test used a "Long Short Term Memory (LSTM)" network as a neural network. A type of RNN, LSTM "selectively remembers" patterns over a period of time. "Anandtech also used three different configurations to test it – Tensorflow ready to use with conda, a Tensorflow optimized for Intel with PyPi, and optimized version of Tensorflow from the source using Bazel, using the latest version of Tensorflow.
This pair of images captures the relative scale between the processors as well as the comparison with the RTX Titan. Ready-to-use performance was quite poor under AMD, although it improved with the optimized code. Intel's performance skyrocketed like a rocket when the source-optimized version was tested, but even the source-optimized version did not fit the performance of Titan RTX very well. De Gelas notes: "Secondly, we were pretty surprised that our Titan RTX is less than three times faster than our dual Xeon setup," which explains how these comparisons are done in the larger article.
DL Boost is not enough to narrow the gap between Intel and Nvidia, but in all fairness, this has probably never been supposed to be. Intel's goal is to improve AI performance enough on Xeon to make the execution of these plausible workloads on servers that will be mainly used for other purposes, or when creating models of artificial intelligence that do not meet the constraints of D & C A modern graphics processor. The long-term goal of the company is to compete in the AI market with a range of equipment, not just Xeons. Since Xe is not quite ready yet, competing in the HPC space means competing with Xeon for the moment.
For those of you who are wondering about AMD, AMD does not really talk about performing artificial intelligence workloads on Epyc processors, but is focused on its RocM initiative to run CUDA code under OpenCL. AMD does not talk much about this side of its business, but Nvidia dominates the GPU application market for AI and HPC. AMD and Intel both want a piece of space. At the moment, both seem to be fighting for one.
In recent weeks, FaceApp – the photo enhancement tool for smartphone-based AI – has become the source of a major controversy over data privacy that appears to have been largely overestimated. However, this highlights a clear and common problem regarding the rights that we could give up with potentially any application we allow on our devices.
On July 14, developer Joshua Nozzi tweeted a charge (since removed) indicating that FaceApp seemed to download all the photos from a user's library and not just the photos selected by a given user for use with the application's services. He also pointed to Russia's involvement in the company, reinforcing common concerns about the illicit Russian involvement in US data-related cases. In a few days, a pseudonym security researcher Elliot Alderson responded at 9t05 cover of Nozzi's charge by Mac with contrary evidence. FaceApp too replied with a statement to 9t05Mac with similar intent. Here is the abridged version:
We could store a downloaded photo in the cloud. The main reason is the performance and the traffic: we want to make sure that the user does not download the photo several times for each editing operation. Most images are removed from our servers within 48 hours of the download date.
FaceApp performs the essential of processing photos in the cloud. We only upload a photo selected by a user for editing. We never transfer other images from the phone to the cloud.
Although the main R & D team is in Russia, user data is not transferred to Russia.
Although 9t05mac has taken the plunge by publishing Nozzi's accusation, his claims have been proven false, Chance Miller – the author of the article – raises an important point:
It's always wise to take a step back when applications like FaceApp become viral. Although they are often popular and can provide humorous content, they can have unintended consequences and privacy issues.
The false accusation of Nozzi seems more to be an honest mistake than a malicious act. Miller's argument shows why we are more prone to panic when independent circumstances give us a picture of danger. Although we should always take the time to find evidence of our claims before publishing them, in order to avoid widespread panic unnecessarily, it is not difficult to see how a person could commit this mistake while people are in alert status for this type of activity.
Although FaceApp has not prompted anyone to own their photo library to build a massive database of US citizens for the Russian government – or the conspiracy theory that you prefer – this incident highlights the ease with which we provide wide permissions once we download an application.
When an application requests access to your smartphone data, it generates a large network by necessity. Photo apps do not require the right to save photos or access only photos that you explicitly show, but to your entire photo library. You can not provide access to the microphone and camera, or anything else, with granular permissions that let you control what the application can do. In addition, smartphones do not provide a simple way to see what applications are doing. Newspapers of any kind, or a way to monitor network activity, are not made available to the average user.
For this reason, most users do not have the opportunity to know if an application has made them lose confidence or not. Until we have better control over the applications our apps can and can not access on our devices, we need to consider the worst case scenario every time we download. Unless a person has the knowledge and willingness to regularly monitor the activity of applications, as well as to read (and understand) the terms of service of each application in their entirety, that person can not exclude the possibility of the use of the application. malicious use of their data. After all Facebook has just been fined $ 5 billion for allowing the very consensual leak of user data. (not that it mattered) and much of this has happened through the association of a person with a user who has downloaded the problematic application.
Although the most commonly used apps do not end up in such controversial situations, data leaks happen quite frequently so we have to remember what we risk with every contribution of our personal information. Each granted access, each uploaded photo and each piece of information provided to an application, that it identifies us directly or indirectly-Provides a company with new information about us which it often claims ownership through its terms of service. They may or may not use the data collected for unpleasant purposes, but they allow themselves this right through a process they know that almost everyone will ignore. Businesses need a broad language in their legal agreements to protect themselves. Unfortunately, this legal requirement also creates a framework for leveraging users when a company publishes an application for data collection purposes.
Granular permissions on smartphones are a step forward in addressing this issue, but it will not prevent applications from continuing to request extended permissions and requiring access as an admission price. At this point, most of us know that we pay with our data when we do not pay with our dollars, but the problematic difference lies in the exact cost. Most people would probably not be afraid that FaceApp would use their selfies to improve the quality of service, but they might feel different if these data were used for some other reason. Even if we do not provide all of our photo libraries, and even if FaceApp removes images 48 hours later, they still have enough time to take advantage of the data voluntarily provided by users. Although it seems that they have no malicious intent, we do not know exactly what our data is costing us because we do not know how they use it.
The same applies to almost all the applications we download. Without transparency, we pay a fixed cost in secret. With repeated actions on many applications, it becomes very difficult to determine the source of the potential problems. FaceApp seems to work like any other application: asking for extended data permissions by necessity and reducing liability through a service terms contract. With each application, we must ask ourselves if the service provided is worth the cost of an unknown cost.
We are not the day when artificial intelligence will provide us with a brush for reality. As foundations on which we rely for their integrity, many people are frightened by what will happen. But we have always lived in a world where our senses do not represent reality. New technologies will help us get closer to the truth by showing where we can not find it.
D & # 39; a historical point of viewWe have never successfully halted the progress of a technology and we owe the level of security and safety that we appreciate this continued progression. While normal accidents occur and the inconveniences of progress will probably never cease to exist, we aggravate the problem when we try to fight the inevitable. In addition, the reality has never been so clear and precise as we want to believe. We are fighting against new technologies because we think this creates uncertainty when, more precisely, it only highlights the uncertainty that has always existed and that we have preferred to ignore.
The dissolution of our reality – a fear provoked by artificial intelligence – is a mirage. For a long time, we trusted what we see and hear throughout our lives, whether in the media or with people we know. But no reality is reality because the reality has never been absolute. Our reality is a relative construction. That's what we agree on based on information from our experience. By observing and sharing our observations, we can try to construct an image of objective reality. Of course, this goal becomes much harder to reach when people are lying or using technology that makes convincing lies more possible. This seems to threaten the very stability of reality as we know it.
But our idea of reality is imperfect. It includes human observation and conjecture. It is limited by the way our body perceives the world around us and by our brain that processes the information acquired. Although we can capture a lot, we can only detect a fragment of the electromagnetic spectrum and even that is too much for our brain to deal with immediately. As the healing brush in Photoshop, our brains fill the gaps in our vision with his best guess what it belongs to. You can test your blind spots to get a better idea of how it works or just watch it in action by looking at an optical illusion like this:
This, among other cognitive processesyou produce submitted versions of reality. You can not feel every aspect of a moment and you will certainly remember every detail. But on top of that, you do not even see everything you see. Your brain builds the missing parts, hide visual information (mostly when we move), makes you hear the bad soundsand can confuse the rubber members with yours. When you have a limited view of a given moment and the information you get is not completely accurate, you start with a subjective version of the reality that you are able to gauge. Trusting collective human observations led us to believe geese grew up on trees for about 700 years. Human observations, conclusions and beliefs are not an objective reality. Even in the best case, we will sometimes come up with some extraordinary mistakes.
Everything you know and understand goes through your brain and brain does not give an accurate picture of reality. To make matters worse, we often miss our memories in many ways. Our vision of the world is neither true nor distant. So for a long time, we relied on others to help us understand what is true. This can work very well in many situations, but sometimes people will have very different versions of the same situation because of past experiences. In both cases, problems arise when subjective observations contradict one another and people can not agree on what really happened. Technology has helped us to improve this technological problem that we had feared during its initial introduction.
Over time, we have created tools to help us survive as a species. By developing new tools, we were able to disseminate information more easily and create a climate of trust. Video and audio recordings allowed us to bypass brain processes and record an unenriched record of an event – at least from a singular point of view. A video camera still fails to capture all the reality of a given moment.
For example, imagine that someone pulls a knife in a fight and pretends to sweep to try to scare off his attacker without any intention of causing real harm. Video surveillance draws a different picture without this context. For an officer of justice, the images of security will show assaults with a deadly weapon. In the absence of other evidence, the officer must exercise caution and make an arrest.
That such assumptions lead to fewer crimes or more questionable arrests does not change the fact that an objective record of the reality lacks information. We trust recordings as truth when they offer only part of the truth. When we trust video, audio or anything that can not tell the whole story, we rely on a support that lies by omission by design, as any observer of reality.
Technology has flaws, but that does not give it away. Overall, we have benefited from advances that have resulted in objective recordings of the world around us. All records do not require additional context. A video of a cute puppy might not please everyone, but most people will agree that they will see a puppy. In the meantime, we called the green sky and can not agree on the color of a dress in a bad photo. As technology progresses and becomes accessible to more and more people, we all begin to learn when and how we can paint reality with a less precise brush than we would have thought.
This awareness causes fear because of our system of understanding, the world begins to collapse. We can not rely on tools that we could not understand our world. We need to question the reliability of the things we have recorded and it goes against much of what we have learned, experienced and integrated with our identities. As new technologies emerge, they further weaken our ability to trust what is familiar to us, they create this fear that we tend to attribute to technology rather than ourselves. Phone calls are part of normal life, but they were, in the beginning, considered an instrument of the devil.
Today, AI is experiencing similar problems. Deepfakes panic when people began to understand how to easily exchange faces with stunning accuracy – with many quality videos and photos meeting specific requirements. Although these deepfakes have rarely deceived anyone, we have all glimpsed the near future in which artificial intelligence would progress to the point of not knowing the difference. That day came last month at Stanford University, Princeton University, the Max Planck Institute for Computing, and Adobe. published a paper this demonstrated an incredibly simple way to edit a recorded video to change the spoken dialogue visually and audibly, which fooled the majority of people who saw the results. Take a look:
Visit the paper summary and you will find most of the text devoted to ethical considerations – a common practice these days. Researchers in artificial intelligence can not get a good job without considering the possible applications of their work. This includes discussion of cases of malicious use so that people understand how to use it for useful purposes and allow them to also prepare themselves for problems that may arise.
Ethical statements can fuel public panic because they act indirectly as a kind of vague science fiction in which our fearful imaginary must fill the gaps. When experts present the problem, it's easy to think only about the worst-case scenarios. Even if the benefits are taken into account, faster video editing and error correction seem to be a slight advantage when the negatives include false information that people will have trouble identifying.
However, this technology will emerge regardless of the efforts made to stop it. Our own history demonstrates time and time again that Any effort to stop the progress of science will, at best, result in a short delay. We should not want to prevent people who understand and care about ethics from what they are, because it leaves others to create the same technology in the shadows. What we can not see seems less frightening for a moment, but we have no way to prepare, understand, or guide these efforts when they are invisible.
Although technologies such as the aforementioned textual video editor inevitably lead to both uses and artificial intelligence more efficient in the future, we are already victims of similar manipulations in everyday life. The doctored photos are nothing new and manipulative editing shows how the context can determine the meaning.a technique taught at the film school. AI adds another tool to the box and increases mistrust in media that has always been easily manipulated. It's unpleasant to live, but ultimately a good thing.
We trust our senses and the recordings we watch too much. Reminders of this help prevent us from doing so. When Apple adds a correction of attention to video chats and Google offers a voice assistant that can make phone calls for you. we will have to remember that we see and hear may not accurately represent reality. Life does not require precision to progress and prosper. To pretend that we can observe objective reality does more harm than to accept that we can not. We do not know everything, our goal remains a mystery to science and we will always make mistakes. Our problem is not with artificial intelligence, but rather that we believe we know the full story when we only know a few details.
As we enter this new era, we should not be fighting against the inevitable technology that continues to shine to highlight our misplaced trust. AI continues to demonstrate the fragility of how we view reality as a species at a very fast pace. This kind of change hurts. We realized that we had only imagined the stable ground on which we had walked all our lives. We are looking for a new place of stability in the face of uncertainties because we consider that the solution is the problem. We may not be ready for this change, but if we fight the inevitable, we will never be.
Artificial intelligence will continue to erode the false comfort we enjoy, which can be scary, but it is also an opportunity. This gives us a choice: to oppose something that scares us or to try to understand it and use it for the benefit of humanity.
Top image credit: Getty Images