Google’s artificial intelligence (AI) has allegedly flagged some parents’ accounts for possible abuse over naked pictures of their sick kids.
According to a father, the tech giant flagged the images as child sexual abuse material (CSAM) after he used his Android smartphone to take photos of an infection on the groin of his toddler. Google closed his accounts and reported him to the National Center for Missing and Exploited Children (NCMEC), thus spurring a police investigation.
This incident highlights the complications involved in identifying the difference between an innocent photo and potential abuse once it becomes a part of the user’s digital library, whether on their cloud storage or personal device. The incident occurred in 2021 when some hospitals were closed due to the pandemic.
As per the report, the father (whose name was not revealed) noticed swelling in his child’s groin and sent images of the issue at a nurse’s request before a video consultation. The doctor ended up prescribing antibiotics that cured the infection.
The father received a notification from Google two days after taking the photos. It stated that his accounts were locked due to harmful content that was a severe violation of policies of Google and might even be illegal.
Like several internet companies, including Twitter, Reddit, and Facebook, Google uses hash matching with Microsoft’s PhotoDNA to scan uploaded images to identify matches with known CSAM. It led to the arrest of a man in 2012 who was a registered sex offender and had used Gmail to send images of a minor girl.
Due to regulatory scrutiny, Chinese social media giant Tencent Holdings suspended issuing its non-fungible tokens (NFT) on the Huanhe platform last Tuesday. A year ago, the Chinese internet behemoth entered the non-fungible token market, sparking considerable interest in the art world. But Tencent is currently obliged to abandon its NFT ambitions due to the murky legal landscape in China. To comply with Beijing’s rules, it will cease releasing “digital collectibles” and would repay customers upon request, according to a statement on the company’s Huanhe app.
The year began with momentous interest and growth in NFTs for companies like Tencent and Alibaba. However, the popularity started to fizzle, with Tencent suffering huge losses in May and June. The Chinese government’s harsh regulations, which restrict the markets for digital collectibles, continue to be the major cause of the sharp decline in sales. Users who have obtained NFTs are prohibited from engaging in private transactions per government regulations. Since the NFTs become less lucrative as a result of this action, buyers are mainly demotivated due to the absence of a secondary market. Additionally, the legislation fosters a situation in which it is almost hard for purchasers to benefit from NFTs.
Additionally, individuals who register on the marketplace sites must be older than 18 and successfully complete an identity verification process. While this is not a major restraint, it still restricts the growth of the NFT industry.
Tech companies like Tencent and Ant Group reached an agreement in June to halt the secondary trade of digital collectibles and “self-regulate” their market operations after state media repeatedly raised problems surrounding NFT speculation in the nation. Later, during the first week of July, Tencent decided to shut down one of its NFT platforms. They took remedial action, such as removing the virtual collectible section from their news application and transferring the executives in charge of the NFT platform. At the same time, Tencent’s other NFT platform is having a difficult time surviving in the hostile market. Experts advise that companies in the Web3 domain must safeguard themselves while volatile market circumstances persist.
Huanhe never promoted its digital collectibles as investments in order to adhere to stringent Chinese regulations that prevent trading in digital assets, including cryptocurrency. Further, Huanhe customers had no way to resale their items, so the digital collectibles were inextricably linked to the buyer’s account after they have been purchased. In order to avoid regulatory issues, Chinese NFT platforms changed their names to digital collectible platforms in late 2021, giving rise to the phrase “digital collectible.”
Digital collectibles marketplaces have flourished locally despite strict oversight. According to research by China’s National Press and Publication Administration, there were only around 100 of these platforms in China in February 2022, but there were about 700 by early July.
Recently, KFC (Kentucky Fried Chicken) India has created something truly innovative in the NFT domain. On Thursday, the iconic red and white striped bucket from KFC India debuted as an NFT in the digital realm. The “KFC BuckETH,” as KFC calls it, is a unique single collectible that will be hosted on the OpenSea NFT marketplace. It was minted on the Ethereum blockchain in collaboration with Blink Digital. The NFT was announced during an Instagram Live hosted by comedian, author, and actor Danish Sait and influencer Sharan Hegde.
In order to commemorate the brand’s achievement of expanding to 600 restaurants across 150 Indian locations, emerging artists from all over India collaborated to design the artwork for this NFT. As a digital collectible, the KFC BuckETH showcases a combination of their distinctive styles. The artwork is considered an “ode to the unique melting pot culture that is India.”
A KFC spokeswoman provided further information regarding the KFC BuckETH, stating, “The Bucket is as iconic to KFC as the signature taste of our chicken. The Bucket is testimony to the brand’s heritage and has been an integral part of many celebratory moments for our customers.” With its first NFT, KFC BuckETH, which was chosen from a variety of vivid designs created by up-and-coming artists, the company is thrilled to take the Bucket in a new direction in the modern digital world. The KFC BuckETH offers customers a chance to interact with the company in a completely unique manner.
KFC India released a photo of its newly made NFT on Twitter and Instagram, calling it the “crunchiest bucket on the block (chain).” The KFC BuckETH will be given away as a prize in a social media contest along with a year’s supply of KFC!
To enter the contest for the NFT, you should visit KFC India’s official Instagram account @kfcindia_official and screenshot the “Ultimate Chicken Lover Checklist,” which will be displayed on the account’s Stories. After that, you must add GIFs, photos, or text to the checklist and share it to your Instagram story while tagging KFC. The restaurant chain will choose one winner, and they will get a year’s worth of KFC as well as the KFC BuckETH NFT.
KFC is not the first food chain to venture into the world of NFTs. Taco Bell was the first fast food company to introduce an NFT collection in March last year. On the NFTMarketplace Rarible, the company displayed five variations of the “NFTacoBells,” and all of them quickly sold out.
KFC Malaysia had introduced an NFT collection in May that featured illustrations of iconic KFC experiences. Three Malaysian artists created a total of 11 NFTs, which are obviously a tribute to the branch’s 11 top-secret herbs and spices.
The collection includes three Generations artwork by Arif Rafhan, four Together Editions artwork by Wilson Ng, and four Moving Fried Chicken artwork by Book of Lai.
The World Robot Conference 2022, also called China Robot Expo, is being held in Beijing from 18 to 21 August. Around 130 Chinese and foreign exhibitors have been showing off the latest innovations in robot technology, including drones, robot servers, exoskeletons for the health care sector, and humanoid robots used in search-and-rescue missions,
The robot makers at the China robot expo said they are expanding fast beyond the industrial droids with innovations that can care for the elderly in a rapidly aging society or deliver hotel room services.
Industrial robots that can help lift boxes or assemble cars account for the bulk of sales in the world’s biggest market, i.e., China. Now, Chinese robot makers are increasingly eyeing the health care and service sectors due to the reformed government policies that aim to turn China into an innovation hub by 2025.
Shuai Mei, chairwoman of Beijing AI-robotics Technology, said they have the first-move advantage and market advantage in terms of technology. According to the National Bureau of Statistics, last year, China’s production of industrial robots topped 330,000 units from January to November, surging 49% from the previous year.
In 2020, China was ranked first globally for sales of industrial robots by the Frankfurt-based International Federation of Robotics. The industry data showed that the operating income in 2020 topped 100 billion yuan for the first time. According to Macquarie Research, the sector is expected to post an annual growth rate of 20% through 2025.
A US appeals court has ruled that an artificial intelligence system cannot be an inventor under the United States (US) patent law. The US Court of Appeals for the Federal Circuit said the Patent Act needs the inventor to be a natural person. The ruling rejected computer scientist Stephen Thaler’s plea for patents on two inventions created by his DABUS system.
Thaler said that DABUS, i.e., Device for the Autonomous Bootstrapping of Unified Sentience, is sentient and natural. His attorney Ryan Abbott said the decision undermines the purpose of the Patent Act and has tangible negative social consequences.
The US Patent and Trademark Office (PTO) refused to comment on the decision. Thaler has lost other bids for patents that state DABUS as their inventor in Australia and the European Union. A Virginia court and PTO, both rejected Thaler’s applications for DABUS patents because the system is not a human being.
Thaler challenged the Virginia court decision before the Federal Circuit, which deals with patent appeals. Abbott told the Federal Circuit that the ruling was at odds with the plain language and purpose of the Patent Act. He added that the act is meant to promote innovation and does not explicitly mention that an inventor must be a natural person.
However, in a unanimous three-judge panel, Circuit Judge Leonard Stark said there is no ambiguity that the Patent Act requires inventors to be natural persons, i.e., human beings.
Stark said that Thaler’s argument that awarding patents to artificial intelligence systems would encourage innovation was speculative. He also dismissed Thaler’s concerns that denying AI patents would diminish the purpose of patents outlined in the US Constitution.
There have been instances where AI has been recognised as a patent inventor. In the month of February, artificial intelligence-powered facial recognition system developing company Clearview Ai received a US patent for its revolutionary face recognition platform.
Tesla, a multinational automotive company, has been facing severe scrutiny over its automobiles’ controversial autopilot and full self-driving features that have allegedly killed many. Several state and federal regulators are seeking action against the electric car maker.
The National Highway Traffic Safety Administration (NHTSA) has upgraded its investigation from a mere preliminary analysis to an extensive engineering evaluation. NHTSA has asked Tesla to give clarification on its cabin digital camera. It is a part of the probe into 830,000 Tesla automobiles that have autopilot.
The US regulator, in a letter, asked Tesla to describe the function that the cabin digital camera performs in enforcing driver engagement and attentiveness. It also asked for clarification on how the camera’s inputs are factored into the topic operation of the system.
The letter also requested Tesla to share inputs on:
The impression on driver engagement alert timing and sorts and its integration with the present engagement technique.
Recoverable information components pointing to its effect both through the automobile’s onboard storage or telemetry.
Impression on driver recoverable and alerting information if the motive force chooses not to share information from the digital camera with Tesla.
The Division of Motor Automobiles (DMV) in California accused Tesla earlier this month of operating pretend claims about its autopilot and full self-driving options. The company stated that Tesla wrongly implies that automobiles outfitted with autopilot can function autonomously.
Tesla has now responded, asking the California DMV for a hearing to present an opposition to the claims that it has misled potential prospects. The California DMV filed two separate complaints, alleging Tesla made deceptive claims about the autonomous driving capabilities of its automobiles.
Roskomnadzor, Russia’s internet watchdog, is creating a neural network called Oculus that will use artificial intelligence (AI) to scan websites for banned information.
The automatic scanner will analyze images, videos, chats, and URLs on forums, social media, websites, and even messenger channels to detect material that should be taken down or redacted.
Information targeted by Oculus includes misinformation that discredits official state and army sources, homosexuality propaganda and instructions on manufacturing weapons or drugs. The system will also look for expressions of disrespect for the state, calls for mass protests, and signs of terrorism and extremism.
Oculus’ real-time scanning capacity will be about 2.3 images per second or 200,000 images per day. For this, Eksikyushn RDC LLC will utilize 48 servers with solid GPUs. Oculus will be integrated onto a network of monitoring systems, the Unified Analysis Module, which is currently under development. The aim is to give the government a grip on controlling information flow.
According to Kommersant, Oculus will cost $965,000 (Russia 57.7 million rubles) and must be completed by December 2022. However, experts in the field suggest that the amount is unlikely to cover the cost of achieving such an aspiring project. Roskomnadzor will likely have to allocate more funds along the way.
Introducing Oculus will result in users either giving up on taking part in online discussions or using anonymization tools like the Tor network, a VPN. People may also resort to chatting apps that do not require PII when registering or logging user data.
Deep Learning (DL) and Artificial Intelligence (AI) have made the future of self-driving automobiles and virtual assistants a reality. The innovations of DL can be found everywhere, on our smartphones, streaming services like Netflix, in virtual reality games, and more. The power of deep learning to make computers think, act, and behave like humans is remarkable. Given the rapid growth of computers and technology, newcomers and old professionals seek to learn this new domain of deep learning. As a result, many people opt for this field and try to contribute to the future. A career in deep learning will benefit young innovative minds to grow personally and professionally. Now, let’s learn what deep learning is and some of the best books for deep learning.
What is Deep Learning?
Deep learning is a subset of machine learning and artificial intelligence. This domain allows computers to process classification tasks directly from data like texts, images, and sounds. It is based on artificial neural networks in which multiple layers are processed, thus, called deep learning to extract higher-level features from data. Deep learning is the process of leveraging data analytics and the latest gains in computing power to work even faster than human minds. Studying deep learning can be hectic if you are not on the right track and don’t have the right resources. Many books have focused on deep learning in the last few years, but which one to pick? Here is a list of top deep learning books that may help you start with deep learning.
List of top Deep Learning books
The Hundred-Page Machine Learning Book by Andriy Burkov
To get into deep learning, you need to know about machine learning. And the best way to learn machine learning is by reading & understanding the algorithms and implementing them. Now, several books for deep learning & machine learning are out in the market, as the field of AI is vast, and so is the variety of books. Also, many things overlap in ML & DL. Thus, you want to grasp a good understanding from the beginning. The book ‘The Hundred-Page Machine Learning Book’, written by Andriy Burkov, an ML expert, is a practical guide to getting started with ML. The first few chapters focus on ML formulation, notations, and key terminologies. Thus, beginners and newcomers in the field can opt for this book. Then the coming chapters analyze the most important algorithms in ML and more advanced topics. Though this book contains only one chapter about neural networks, it indeed serves as a building block for DL.
Written by François Chollet, the creator of Keras and a Google AI researcher, ‘Deep Learning with Python’ explains the concepts of DL using the Python language and Keras library. It is one of the best deep learning books that provide a good understanding of the concepts through intuitive explanations and practical examples. This book encourages beginners and intermediate programmers to understand DL in-depth through extensive descriptions of implementing convolutional neural networks (CNNs). In overview, this book is divided into two parts, first, the fundamentals of DL, and two, DL in practice. The fundamentals cover high-level crucial concepts in DL, and practice mostly covers applications such as DL for computer vision, text & sequences, advanced DL practice, and generative DL. By finishing this book, you’ll have the hands-on skills to apply deep learning models in your projects. You can buy this neural network bookonline.
Hands-on Machine Learning with Scikit-Learn, Keras, and TensorFlow by Aurélien Géron
‘Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow’ will guide you through acquiring basic concepts of DL so that anyone can use simple and efficient tools to implement programs capable of learning from data. Written by Aurélien Géron, a machine learning consultant, this deep learning book comprises concrete examples with minimal theory and two production-ready Python frameworks, Sklearn and TensorFlow, to master the use of DL. This book provides an intuitive understanding of the concepts & tools for building intelligent systems using Scikit & Tensorflow.
You need prior programming knowledge to apply what you learn from this book. The exercises range from simple linear regression to processing deep neural networks, including CNN and transfer learning. This book on deep learning helps you to explore ML, particularly neural networks, and other training models like support vector machines (SVM), decision trees, and ensemble methods. Also, you learn the neural network architectures of CNN, recurrent neural network (RNN), and deep reinforcement learning. Then, you can use Sklearn to track end-to-end ML projects and TensorFlow to build & train the neural networks. The book retails at ₹2,600 for the second updated edition.
Deep Learning from Scratch: Building with Python from First Principles by Seth Weidman
‘Deep Learning from Scratch: building with Python from First Principles’ is a handbook to build your foundation of deep learning. The author Seth Weidman is a data scientist who has a unique way of explaining the concepts with a visual representation of the working of the algorithm, a mathematical explanation of why the algorithm works, and a pseudocode implementation of the algorithm. It is one of the best books on deep learning that teaches how to apply multiplayer neural networks and convolutional networking. Also, it provides a comprehensive introduction to DL for data scientists & software engineers. It focuses on how neural networks work using the first principles hence, the name. The book starts with DL basis and then moves to extensive details of important advanced networks of CNN & RNN. It has a dedicated chapter on extensions and PyTorch, explaining loss function, momentum & weight initialization, etc, and how to implement DL models with PyTorch & unsupervised learning, respectively.
Deep Learning (Adaptive Computation and Machine Learning series) by Ian Goodfellow, Yoshua Bengio, Aaron Courville, Francis Bach
‘Deep Learning (Adaptive Computation and Machine Learning series)’ is on the list of top books for deep learning that presents an in-depth understanding of deep learning, written together by four computer scientists and deep learning enthusiasts. Ian J. Goodfellow, a research scientist in DeepMind, who invented generative adversarial networks (GANs). Yoshua Bengio is one of the leading experts in AI, a professor at the Université de Montréal & head of the Montreal institute for learning algorithms. Aaron Courville is an Associate professor at the Université de Montréal & member of the Mila-Quebec Artificial Intelligence Institute. Francis Bach is a world-renowned ML expert and researcher at the National institute for research in digital science and technology (INRIA). The books combine a wide range of concepts and topics in deep learning. It is divided into three parts, first, applied math & ML basics; second, modern practices in DL, and third, DL research. The first part has a firm mathematical foundation and covers linear algebra, probability theory, information theory, and numerical computation. In the second part, the book explains deep feedforward networks, regularization, optimization, CNN, sequence modeling, and applications. In the third and final part, the book offers insight into linear factor models, autoencoders, representation learning, Monte Carlo methods, structured probabilistic models, confronting partition function, and deep generative models. This book is an excellent addition to deep learning books, which is available online.
‘Grokking Deep Learning’ talks about the science behind DL by explaining the building and training of neural networks. The author Andrew W. Trask, a PhD student at Oxford University and a research scientist at DeepMind, focused on unveiling the science under the hood so that you understand every detail of training a neural network. This book emphasizes using Python and NumPy to train neural networks to see & understand images, translate text into different languages, etc, to master the working of DL frameworks. Beginners can see this neural network and deep learning textbook as a mentor, as it walks through every aspect of the why, what, and how of deep learning models. In the end, you get a chapter, ‘Where to go from here’ in which the factors of DL are explained and how DL will be a promising career for you.
Deep Learning with PyTorch by Eli Stevens, Luca Antiga, Thomas Viehmann
‘Deep Learning with PyTorch’ is among the most popular machine learning and deep learning books. This practical book dynamically gets you to build real-world projects from scratch. The authors are Eli Stevens, a software engineer & CTO of a startup company building software for radiology, Luca Antiga, the co-founder & CEO of an AI engineering company and a constant contributor to PyTorch, and Thomas Viehmann, a core PyTorch core developer and an ML & PyTorch specialist trainer & consultant. The book teaches you how to create neural networks & DL systems with PyTorch. It covers some of the best practices for DL pipeline and basics and takes you to larger projects. The highlight of this book is an elaborated neural network designed for cancer detection. This is a whole package for deep learning books where you discover ways to train networks with limited inputs and then focus on the diagnosis to fix problems in the network. Eventually, you will learn ways to improve the network & architecture, perform fine-tuning, and the results with augmented data.
Researchers at Stanford University, MIT’s Computer and AI lab, and the Autodesk AI Lab have created a deep learning framework to construct 3D objects. The Manual-to-Executable-Plan Network (MEPNet) framework was tested on Lego sets generated by computers. The training included genuine Lego set instructions and Minecraft-style voxel building plans.
Existing methods of rendering 3D objects are simple but computationally expensive and not very good at handling unseen shapes. Additionally, a few problems surface when existing AI techniques interpret 2D instructions to transform them into 3D. Visual instructions like Lego sets consist entirely of images; hence, identifying differences between 2D and 3D can become complex because they are usually assembled.
The researchers said, “This increases the difficulty for machines to interpret Lego manuals: it requires inferring 3D poses of unseen objects composed of seen primitives.”
MEPNet combines the existing upsides and new 3D rendering techniques by starting with a 3D model of components, Lego set, and 2D manual images. It predicts a set of 2D keypoints and masks each component.
After masking, 2D keypoints are “back-projected to 3D by finding possible connections between the base shape and the new components.” The team said the combination “maintains the efficiency of learning-based models.”
All you have to do is interpret MEPNet’s 3D renderings, which would hopefully be easier than flat-pack furniture instructions. You can test MEPNet here if you are familiar with PyTorch.
Google's PaLM-SayCan robots use AI language models to understand that picking up a sponge is useful someone who needs help with a spilled drink.
Credit: Stephen Shankland
Imagine you have come home after hitting the gym, you want to have a smoothie, but you are too tired to make one. You suddenly remember you have to wash utensils from the previous meal, vacuum the floor and cook for dinner–but you are still tired and sore from your intensive workout routine at the gym. Well, Google can help you with it! Google has revealed that it is working on an artificial intelligence (AI) system that can pick up on human communication styles and innately carry out human wishes. Google has also published a robot that is in development that is outfitted with this AI as per its paper ‘Do As I Can, Not As I Say: Grounding Language in Robotic Affordances.’
You may hire a butler like Batman’s Alfred to help you with daily chores (if you can afford one) or ask your robot butler. The problem with robots is that though they are adept at carrying out short hard-coded instructions systematically, they may fail in comprehending ambiguous requests. For instance, if you mention that you are hungry to a robot, it may acknowledge that yet do not know what to do next. However, a robot from Everyday Robots Project, a group under its experimental X labs, can offer you a bag of Doritos from your Kitchen counter without receiving instructions from you about the same. Through the use of millions of web-scraped text pages, the robot’s control software has developed the ability to convert spoken words into a series of physical movements.
With the use of the technology and Google’s AI language model, a robot can now decipher ambiguous human commands and put together a series of responses. That contrasts sharply with the carefully programmed tasks that the majority of robots carry out under strictly regulated conditions, such as fixing windshields on a vehicle manufacturing line. This proves that we are closer to witnessing robots straight out of science fiction.
Google reveals that, unlike virtual assistants like Alexa or Siri, a person doesn’t need to deliver orders using a certain set of previously approved wake-up words for this AI robot. The robot would try to fetch you something to drink if you say “I’m thirsty,” and it should return with a sponge if you say, “Whoops, I just spilled my drink.” This technological feat has been made possible with the use of the most powerful large language model developed by Google. Dubbed the Pathways Language Model (PaLM), this large language model is a dense decoder-only Transformer model with 540 billion parameters that was trained using the Pathways technology, allowing Google to effectively train a single model across several TPU v4 Pods.
PaLM was trained using a combination of English and multilingual datasets, including GitHub code, high-quality web publications, articles from Wikipedia, and chats. Additionally, Google had developed a “lossless” vocabulary that breaks numbers into separate tokens, one for each digit, splits non-vocabulary Unicode characters into bytes, and maintains all whitespace (which is crucial for coding). At the time of its announcement, Google claimed that PaLM performs impressively on a variety of BIG-bench tests for natural language processing and creation. The model, for instance, can recognize cause and effect, comprehend conceptual combinations in certain settings, and even identify a movie from an emoji.
The robot butler was developed by Google researchers using new software that takes advantage of PaLM’s text processing skills to transform a spoken command or phrase into a series of relevant actions that the robot may carry out, such as “open drawer” or “pick up chips.” Google has christened the resulting system PaLM-SayCan, a catchphrase that describes how the model blends the language comprehension skills of LLMs (“Say”) with the “dynamic capabilities grounding” of its robots (that’s “Can” – processing instructions via various actions).
The robot has a glowing green rim around their faces to signify when it is active. It changes color or switches off in other circumstances. Credit: Stephen Shankland
In order for the robot to independently explore a location and recognize objects and locations relevant to a command, it also has hearing and optical sensors.
Per the Everyday Robots, by incorporating a multitude of machine learning algorithms such as reinforcement learning, collaborative learning, and learning from demonstration, the robots have progressively improved their knowledge of their environment and their aptitude for doing common tasks.
An octet of Google PaLM-StayCan Robots practices manipulations like opening drawers and grabbing objects. Credit: Stephen Shankland
Through a separate training phase, where people remotely operated the PaLM-SayCan robot to demonstrate how to perform things like picking up objects, the robot learned its library of physical activities. It can only carry out a certain number of activities inside its surroundings, which helps avoid language model ambiguities from manifesting as wayward behavior. Google claims that this technology is ready to go mainstream as the company researchers have accomplished research undertaking. Instead of testing it in a more controlled lab setting, Google has been trialing it in the employee kitchen area so as to create robots that can be useful in the unexpected turmoil of our everyday lives. This exemplifies the potential for butler robots to adjust to the uncertainty of the real world. The ability of robots to browse the internet and fulfill purchases is already progressing as Google Research, and Everyday Robots collaborate together to integrate the finest language models with robot learning.
Google PaLM-SayCan robot retrieves a bag of chips from a drawer in an employee kitchen area. Credit: Stephen Shankland
However, due to the assistants’ limited ability to respond to orders contextually and the fact that the announcement merely served as a preview of possible capabilities, the robotic butlers are not yet suitable for commercial deployment.
Meanwhile, although Google claims to be pursuing research responsibly, fears about robots becoming surveillance machines or possessing technology that might respond in an inappropriate manner could ultimately cause adoption to stagnate. Google reassures individuals who worry that things can go wrong that they take a proactive approach to this research and adhere to Google’s AI Principles while building helper robots.
Google PaLM-SayCan Robot drops a Pepsi can into a recycle bin. Credit: Stephen Shankland
According to Google, by incorporating PaLM-SayCan into their robots, the robots were able to chart the right actions to 101 user instructions 84% of the time and carry them out 74% of the time. Despite the fact that these statistics are impressive, the data should be interpreted cautiously. Since we don’t have access to all 101 commands, it’s unclear how limited these directives were. Were these 101 instructions tailored to grasp the complexity of language that a true robot butler would be able to understand? Can they understand complex commands or wishes like, ‘I want an orange soda instead of lime,’ ‘Can you organize the closet,’ or ‘Would you julienne the tomatoes instead of dicing them.’ Can their actions align with human expectations every time? For instance, when asked to ‘put on the TV,’ would it switch on the TV (human intent) or put the TV at some place (machine logical reasoning)?
Some skeptics believe that once an AI system reaches a certain level of complexity and reacts to its surroundings in a manner resembling that of a human, we should consider it to be aware and, maybe, to have rights. The recent controversy around permitting AI-powered robots to be a part of daily human life was brought up during the Moscow Chess Open competition when a chess-playing robot went rogue and assaulted its 7-year-old opponent for not waiting for the robot to make its move.
From an architectural perspective, the majority of contemporary AI systems focus on one job or a narrow band of tasks at a time. In contrast, PaLM-SayCan will be expected to understand the human conversation and deliver expected results that span multiple tasks, including fetching you a bag of chips. Not only that, but PaLM-SayCan must also differentiate between logically carrying out human commands and cognitive-ethical reasoning. If asked to feed chocolate to a dog, would it promptly follow the instructions or use the fact that dogs are allergic to chocolate to refrain from following human commands?
