AI Assistant For Writing

AI Assistant For Writing — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

• 

  SCADA Strangelove

  SCADA Strangelove is an independent group of information security researchers founded in 2012, focused on security assessment of industrial control systems (ICS) and SCADA. == Activities == Main fields of research include: Discovery of 0-day vulnerabilities in cyber physical systems and coordinated vulnerability disclosure; Security assessment of ICS protocols and development suites; Identification of publicly Internet-connected ICS components and secure it with help of proper authorities; Development of security hardening guides for ICS software; Mapping cybersecurity on to functional safety; Awareness control and delivery of information regarding the actual security state of ICS systems. SCADA Strangelove's interests expand further than classic ICS components and covers various embedded systems, however, and encompass smart home components, solar panels, wind turbines, SmartGrid as well as other areas. == Projects == Group members have and continue to develop and publish numerous open source tools for scanning, fingerprinting, security evaluation and password bruteforcing for ICS devices. These devices work over industrial protocols such as modbus, Siemens S7, MMS, ISO EC 60870, ProfiNet. In 2014 Shodan used some of the published tools for building a map of ICS devices which is publicly available on the Internet. Open source security assessment frameworks, such as THC Hydra, Metasploit, and DigitalBond Redpoint have used Shodan-developed tools and techniques. The group has published security-hardening guidelines for industrial solutions based on Siemens SIMATIC WinCC and WinCC Flexible. The guidelines contain detailed security configuration walk-throughs, descriptions of internal security features and appropriate best practices. Among the group’s more noticeable projects is Choo Choo PWN (CCP) also named the Critical Infrastructure Attack (CIA). This is an interactive laboratory built upon ICS software and hardware used in real world. Every system is connected to a toy city infrastructure, which includes factories, railroads and other facilities. The laboratory has been demonstrated at various conferences including PHDays, Power of Community, and 30C3. Primarily the laboratory is used for the discovery of new vulnerabilities and for evaluation of security mechanisms, however it is also used for workshops and other educational activities. At Positive Hack Days IV, contestants found several 0-day vulnerabilities in Indusoft Web Studio 7.1 by Schneider Electric, and in specific ICS hardware RTU PET-7000 during the ICS vulnerability discovery challenge. The group supports Secure Open SmartGrid (SCADASOS) project to find and fix vulnerabilities in intellectual power grid components such as photovoltaic power station, wind turbine, power inverter. More than 80 000 industrial devices were discovered and isolated from the Internet in 2015. == Appearances == Group members are frequently seen presenting at conferences like CCC, SCADA Security Scientific Symposium, Positive Hack Days. Most notable talks are: === 29C3 === An overview of vulnerabilities discovered in the widely distributed Siemens SIMATIC WinCC software and tools that are implemented for searching ICS on the Internet. === PHDays === This talk consisted of an overview of vulnerabilities discovered in various systems produced by ABB, Emerson, Honeywell and Siemens and was presented at PHDays III and PHDays IV. === Confidence 2014 === Implications of security research aimed at realization of various industrial network protocols Profinet, Modbus, DNP3, IEC 61850-8-1 (MMS), IEC (International Electrotechnical Commission) 61870-5-101/104, FTE (Fault Tolerant Ethernet), Siemens S7. === PacSec 2014 === Presentations of security research showing the impact of radio and 3G/4G networks on the security of mobile devices as well as on industrial equipment. === 31C3 === Analysis of security architecture and implementation of the most wide spread platforms for wind and solar energy generation which produce many gigawatts of it. === 32C3 === Cybersecurity assessment of railway signaling systems such as Automatic Train Control (ATC), Computer-based interlocking (CBI) and European Train Control System (ETCS). === China Internet Security Conference 2016 === In "Greater China Cyber Threat Landscape" keynote by Sergey Gordeychik an overview of vulnerabilities, attacks and cyber-security incidents in Greater China region was presented. === Recon 2017 === In talk "Hopeless: Relay Protection for Substation Automation" by Kirill Nesterov and Alexander Tlyapov security analysis results of key Digital Substation component - Relay Protection Terminals was presented. Vulnerabilities, including remote code execution in Siemens SIPROTEC, General Electric Line Distance Relay, NARI and ABB protective relays was presented. == Philosophy == All names, catchwords and graphical elements refer to Stanley Kubrick’s film, Dr. Strangelove. In their talks, group members often refer to Cold War events such as the Caribbean Crisis, and draw parallels between nuclear arms race and the current escalation of cyberwar. Group members follow the approach of “responsible disclosure” and “ready to wait for years, while vendor is patching the vulnerability”. Public exploits for discovered vulnerabilities are not published. This is on account of the longevity of ICS and by implication the long process of patching ICS. However, conflicts still happen, notably in 2012 when the talk at DEF CON was called off due to a dispute of persistent weaknesses in Siemens industrial software.

  Read more →
• 

  Matchbox Educable Noughts and Crosses Engine

  The Matchbox Educable Noughts and Crosses Engine (sometimes called the Machine Educable Noughts and Crosses Engine or MENACE) was a mechanical computer made from 304 matchboxes designed and built by artificial intelligence researcher Donald Michie and his colleague Roger Chambers, in 1961. It was designed to play human opponents in games of noughts and crosses (tic-tac-toe) by returning a move for any given state of play and to refine its strategy through reinforcement learning. This was one of the first types of artificial intelligence. Michie and Chambers did not have immediate access to a computer; they worked around this by building the engine out of matchboxes. The matchboxes they used each represented a single possible layout of a noughts and crosses grid. When the computer first played, it would randomly choose moves based on the current layout. As it played more games, through a reinforcement loop, it disqualified strategies that led to losing games, and supplemented strategies that led to winning games. Michie held a tournament against MENACE in 1961, wherein he experimented with different openings. Following MENACE's maiden tournament against Michie, it demonstrated successful artificial intelligence in its strategy. Michie's essays on MENACE's weight initialisation and the BOXES algorithm used by MENACE became popular in the field of computer science research. Michie was honoured for his contribution to machine learning research, and was twice commissioned to program a MENACE simulation on an actual computer. == Origin == Donald Michie (1923–2007) had been on the team decrypting the German Tunny Code during World War II. Fifteen years later, he wanted to further display his mathematical and computational prowess with an early convolutional neural network. Since computer equipment was not obtainable for such uses, and Michie did not have a computer readily available, he decided to display and demonstrate artificial intelligence in a more esoteric format and constructed a functional mechanical computer out of matchboxes and beads. MENACE was constructed as the result of a bet with a computer science colleague who postulated that such a machine was impossible. Michie undertook the task of collecting and defining each matchbox as a "fun project", later turned into a demonstration tool. Michie completed his essay on MENACE in 1963, "Experiments on the mechanization of game-learning", as well as his essay on the BOXES Algorithm, written with R. A. Chambers and had built up an AI research unit in Hope Park Square, Edinburgh, Scotland. MENACE learned by playing successive matches of noughts and crosses. Each time, it would eliminate a losing strategy by the human player confiscating the beads that corresponded to each move. It reinforced winning strategies by making the moves more likely, by supplying extra beads. This was one of the earliest versions of the Reinforcement Loop, the schematic algorithm of looping the algorithm, dropping unsuccessful strategies until only the winning ones remain. This model starts as completely random, and gradually learns. == Composition == MENACE was made from 304 matchboxes glued together in an arrangement similar to a chest of drawers. Each box had a code number, which was keyed into a chart. This chart had drawings of tic-tac-toe game grids with various configurations of X, O, and empty squares, corresponding to all possible permutations a game could go through as it progressed. After removing duplicate arrangements (ones that were simply rotations or mirror images of other configurations), MENACE used 304 permutations in its chart and thus that many matchboxes. Each individual matchbox tray contained a collection of coloured beads. Each colour represented a move on a square on the game grid, and so matchboxes with arrangements where positions on the grid were already taken would not have beads for that position. Additionally, at the front of the tray were two extra pieces of card in a "V" shape, the point of the "V" pointing at the front of the matchbox. Michie and his artificial intelligence team called MENACE's algorithm "Boxes", after the apparatus used for the machine. The first stage "Boxes" operated in five phases, each setting a definition and a precedent for the rules of the algorithm in relation to the game. == Operation == MENACE played first, as O, since all matchboxes represented permutations only relevant to the "X" player. To retrieve MENACE's choice of move, the opponent or operator located the matchbox that matched the current game state, or a rotation or mirror image of it. For example, at the start of a game, this would be the matchbox for an empty grid. The tray would be removed and lightly shaken so as to move the beads around. Then, the bead that had rolled into the point of the "V" shape at the front of the tray was the move MENACE had chosen to make. Its colour was then used as the position to play on, and, after accounting for any rotations or flips needed based on the chosen matchbox configuration's relation to the current grid, the O would be placed on that square. Then the player performed their move, the new state was located, a new move selected, and so on, until the game was finished. When the game had finished, the human player observed the game's outcome. As a game was played, each matchbox that was used for MENACE's turn had its tray returned to it ajar, and the bead used kept aside, so that MENACE's choice of moves and the game states they belonged to were recorded. Michie described his reinforcement system with "reward" and "punishment". Once the game was finished, if MENACE had won, it would then receive a "reward" for its victory. The removed beads showed the sequence of the winning moves. These were returned to their respective trays, easily identifiable since they were slightly open, as well as three bonus beads of the same colour. In this way, in future games MENACE would become more likely to repeat those winning moves, reinforcing winning strategies. If it lost, the removed beads were not returned, "punishing" MENACE, and meaning that in future it would be less likely, and eventually incapable if that colour of bead became absent, to repeat the moves that cause a loss. If the game was a draw, one additional bead was added to each box. == Results in practice == === Optimal strategy === Noughts and crosses has a well-known optimal strategy. A player must place their symbol in a way that blocks the other player from achieving any rows while simultaneously making a row themself. However, if both players use this strategy, the game always ends in a draw. If the human player is familiar with the optimal strategy, and MENACE can quickly learn it, then the games will eventually only end in draws. The likelihood of the computer winning increases quickly when the computer plays against a random-playing opponent. When playing against a player using optimal strategy, the odds of a draw grow to 100%. In Donald Michie's official tournament against MENACE in 1961 he used optimal strategy, and he and the computer began to draw consistently after twenty games. Michie's tournament had the following milestones: Michie began by consistently opening with "Variant 0", the middle square. At 15 games, MENACE abandoned all non-corner openings. At just over 20, Michie switched to consistently using "Variant 1", the bottom-right square. At 60, he returned to Variant 0. As he neared 80 games, he moved to "Variant 2", the top-middle. At 110, he switched to "Variant 3", the top right. At 135, he switched to "Variant 4", middle-right. At 190, he returned to Variant 1, and at 210, he returned to Variant 0. The trend in changes of beads in the "2" boxes runs: === Correlation === Depending on the strategy employed by the human player, MENACE produces a different trend on scatter graphs of wins. Using a random turn from the human player results in an almost-perfect positive trend. Playing the optimal strategy returns a slightly slower increase. The reinforcement does not create a perfect standard of wins; the algorithm will draw random uncertain conclusions each time. After the j-th round, the correlation of near-perfect play runs: 1 − D D − D ( j + 2 ) ∑ i = 0 j D ( j i + 1 ) V i {\displaystyle {1-D \over D-D^{(j+2)}}\sum _{i=0}^{j}D^{(ji+1)}V_{i}} Where Vi is the outcome (+1 is win, 0 is draw and -1 is loss) and D is the decay factor (average of past values of wins and losses). Below, Mn is the multiplier for the n-th round of the game. == Legacy == Donald Michie's MENACE proved that a computer could learn from failure and success to become good at a task. It used what would become core principles within the field of machine learning before they had been properly theorised. For example, the combination of how MENACE starts with equal numbers of types of beads in each matchbox, and how these are then selected at random, creates a learning behaviour similar to weight initialisation

  Read more →
• 

  Dynamic epistemic logic

  Dynamic epistemic logic (DEL) is a logical framework dealing with knowledge and information change. Typically, DEL focuses on situations involving multiple agents and studies how their knowledge changes when events occur. These events can change factual properties of the actual world (they are called ontic events): for example a red card is painted in blue. They can also bring about changes of knowledge without changing factual properties of the world (they are called epistemic events): for example, a card is revealed publicly (or privately) to be red. Originally, DEL focused on epistemic events. Only some of the basic ideas are present in this entry of the original DEL framework; more details about DEL in general can be found in the references. Due to the nature of its object of study and its abstract approach, DEL is related and has applications to numerous research areas, such as computer science (artificial intelligence), philosophy (formal epistemology), economics (game theory) and cognitive science. In computer science, DEL is for example very much related to multi-agent systems, which are systems where multiple intelligent agents interact and exchange information. As a combination of dynamic logic and epistemic logic, dynamic epistemic logic is a young field of research. It really started in 1989 with Plaza's logic of public announcement. Independently, Gerbrandy and Groeneveld proposed a system dealing moreover with private announcement and that was inspired by the work of Veltman. Another system was proposed by van Ditmarsch whose main inspiration was the Cluedo game. But the most influential and original system was the system proposed by Baltag, Moss and Solecki. This system can deal with all the types of situations studied in the works above and its underlying methodology is conceptually grounded. This entry will present some of its basic ideas. Formally, DEL extends ordinary epistemic logic by the inclusion of event models to describe actions, and a product update operator that defines how epistemic models are updated as the consequence of executing actions described through event models. Epistemic logic will first be recalled. Then, actions and events will enter into the picture and we will introduce the DEL framework. == Epistemic logic == Epistemic logic is a modal logic dealing with the notions of knowledge and belief. As a logic, it is concerned with understanding the process of reasoning about knowledge and belief: which principles relating the notions of knowledge and belief are intuitively plausible? Like epistemology, it stems from the Greek word ϵ π ι σ τ η μ η {\displaystyle \epsilon \pi \iota \sigma \tau \eta \mu \eta } or ‘episteme’ meaning knowledge. Epistemology is nevertheless more concerned with analyzing the very nature and scope of knowledge, addressing questions such as “What is the definition of knowledge?” or “How is knowledge acquired?”. In fact, epistemic logic grew out of epistemology in the Middle Ages thanks to the efforts of Burley and Ockham. The formal work, based on modal logic, that inaugurated contemporary research into epistemic logic dates back only to 1962 and is due to Hintikka. It then sparked in the 1960s discussions about the principles of knowledge and belief and many axioms for these notions were proposed and discussed. For example, the interaction axioms K p → B p {\displaystyle Kp\rightarrow Bp} and B p → K B p {\displaystyle Bp\rightarrow KBp} are often considered to be intuitive principles: if an agent Knows p {\displaystyle p} then (s)he also Believes p {\displaystyle p} , or if an agent Believes p {\displaystyle p} , then (s)he Knows that (s)he Believes p {\displaystyle p} . More recently, these kinds of philosophical theories were taken up by researchers in economics, artificial intelligence and theoretical computer science where reasoning about knowledge is a central topic. Due to the new setting in which epistemic logic was used, new perspectives and new features such as computability issues were then added to the research agenda of epistemic logic. === Syntax === In the sequel, A G T S = { 1 , … , n } {\displaystyle AGTS=\{1,\ldots ,n\}} is a finite set whose elements are called agents and P R O P {\displaystyle PROP} is a set of propositional letters. The epistemic language is an extension of the basic multi-modal language of modal logic with a common knowledge operator C A {\displaystyle C_{A}} and a distributed knowledge operator D A {\displaystyle D_{A}} . Formally, the epistemic language L EL C {\displaystyle {\mathcal {L}}_{\textsf {EL}}^{C}} is defined inductively by the following grammar in BNF: L EL C : ϕ ::= p ∣ ¬ ϕ ∣ ( ϕ ∧ ϕ ) ∣ K j ϕ ∣ C A ϕ ∣ D A ϕ {\displaystyle {\mathcal {L}}_{\textsf {EL}}^{C}:\phi ~~::=~~p~\mid ~\neg \phi ~\mid ~(\phi \land \phi )~\mid ~K_{j}\phi ~\mid ~C_{A}\phi ~\mid ~D_{A}\phi } where p ∈ P R O P {\displaystyle p\in PROP} , j ∈ A G T S {\displaystyle j\in {AGTS}} and A ⊆ A G T S {\displaystyle A\subseteq {AGTS}} . The basic epistemic language L E L {\displaystyle {\mathcal {L}}_{EL}} is the language L E L C {\displaystyle {\mathcal {L}}_{EL}^{C}} without the common knowledge and distributed knowledge operators. The formula ⊥ {\displaystyle \bot } is an abbreviation for ¬ p ∧ p {\displaystyle \neg p\land p} (for a given p ∈ P R O P {\displaystyle p\in PROP} ), ⟨ K j ⟩ ϕ {\displaystyle \langle K_{j}\rangle \phi } is an abbreviation for ¬ K j ¬ ϕ {\displaystyle \neg K_{j}\neg \phi } , E A ϕ {\displaystyle E_{A}\phi } is an abbreviation for ⋀ j ∈ A K j ϕ {\displaystyle \bigwedge \limits _{j\in A}K_{j}\phi } and C ϕ {\displaystyle C\phi } an abbreviation for C A G T S ϕ {\displaystyle C_{AGTS}\phi } . Group notions: general, common and distributed knowledge. In a multi-agent setting there are three important epistemic concepts: general knowledge, distributed knowledge and common knowledge. The notion of common knowledge was first studied by Lewis in the context of conventions. It was then applied to distributed systems and to game theory, where it allows to express that the rationality of the players, the rules of the game and the set of players are commonly known. General knowledge. General knowledge of ϕ {\displaystyle \phi } means that everybody in the group of agents A G T S {\displaystyle {AGTS}} knows that ϕ {\displaystyle \phi } . Formally, this corresponds to the following formula: E ϕ := ⋀ j ∈ A G T S K j ϕ . {\displaystyle E\phi :={\underset {j\in {AGTS}}{\bigwedge }}K_{j}\phi .} Common knowledge. Common knowledge of ϕ {\displaystyle \phi } means that everybody knows ϕ {\displaystyle \phi } but also that everybody knows that everybody knows ϕ {\displaystyle \phi } , that everybody knows that everybody knows that everybody knows ϕ {\displaystyle \phi } , and so on ad infinitum. Formally, this corresponds to the following formula C ϕ := E ϕ ∧ E E ϕ ∧ E E E ϕ ∧ … {\displaystyle C\phi :=E\phi \land EE\phi \land EEE\phi \land \ldots } As we do not allow infinite conjunction the notion of common knowledge will have to be introduced as a primitive in our language. Before defining the language with this new operator, we are going to give an example introduced by Lewis that illustrates the difference between the notions of general knowledge and common knowledge. Lewis wanted to know what kind of knowledge is needed so that the statement p {\displaystyle p} : “every driver must drive on the right” be a convention among a group of agents. In other words, he wanted to know what kind of knowledge is needed so that everybody feels safe to drive on the right. Suppose there are only two agents i {\displaystyle i} and j {\displaystyle j} . Then everybody knowing p {\displaystyle p} (formally E p {\displaystyle Ep} ) is not enough. Indeed, it might still be possible that the agent i {\displaystyle i} considers possible that the agent j {\displaystyle j} does not know p {\displaystyle p} (formally ¬ K i K j p {\displaystyle \neg K_{i}K_{j}p} ). In that case the agent i {\displaystyle i} will not feel safe to drive on the right because he might consider that the agent j {\displaystyle j} , not knowing p {\displaystyle p} , could drive on the left. To avoid this problem, we could then assume that everybody knows that everybody knows that p {\displaystyle p} (formally E E p {\displaystyle EEp} ). This is again not enough to ensure that everybody feels safe to drive on the right. Indeed, it might still be possible that agent i {\displaystyle i} considers possible that agent j {\displaystyle j} considers possible that agent i {\displaystyle i} does not know p {\displaystyle p} (formally ¬ K i K j K i p {\displaystyle \neg K_{i}K_{j}K_{i}p} ). In that case and from i {\displaystyle i} ’s point of view, j {\displaystyle j} considers possible that i {\displaystyle i} , not knowing p {\displaystyle p} , will drive on the left. So from i {\displaystyle i} ’s point of view, j {\displaystyle j} might drive on the left as well (by the same argument as abov

  Read more →
• 

  Winner-take-all in action selection

  Winner-take-all is a computer science concept that has been widely applied in behavior-based robotics as a method of action selection for intelligent agents. Winner-take-all systems work by connecting modules (task-designated areas) in such a way that when one action is performed it stops all other actions from being performed, so only one action is occurring at a time. The name comes from the idea that the "winner" action takes all of the motor system's power. == History == In the 1980s and 1990s, many roboticists and cognitive scientists were attempting to find speedier and more efficient alternatives to the traditional world modeling method of action selection. In 1982, Jerome A. Feldman and D.H. Ballard published the "Connectionist Models and Their Properties", referencing and explaining winner-take-all as a method of action selection. Feldman's architecture functioned on the simple rule that in a network of interconnected action modules, each module will set its own output to zero if it reads a higher input than its own in any other module. In 1986, Rodney Brooks introduced behavior-based artificial intelligence. Winner-take-all architectures for action selection soon became a common feature of behavior-based robots, because selection occurred at the level of the action modules (bottom-up) rather than at a separate cognitive level (top-down), producing a tight coupling of stimulus and reaction. == Types of winner-take-all architectures == === Hierarchy === In the hierarchical architecture, actions or behaviors are programmed in a high-to-low priority list, with inhibitory connections between all the action modules. The agent performs low-priority behaviors until a higher-priority behavior is stimulated, at which point the higher behavior inhibits all other behaviors and takes over the motor system completely. Prioritized behaviors are usually key to the immediate survival of the agent, while behaviors of lower priority are less time-sensitive. For example, "run away from predator" would be ranked above "sleep." While this architecture allows for clear programming of goals, many roboticists have moved away from the hierarchy because of its inflexibility. === Heterarchy and fully distributed === In the heterarchy and fully distributed architecture, each behavior has a set of pre-conditions to be met before it can be performed, and a set of post-conditions that will be true after the action has been performed. These pre- and post-conditions determine the order in which behaviors must be performed and are used to causally connect action modules. This enables each module to receive input from other modules as well as from the sensors, so modules can recruit each other. For example, if the agent's goal were to reduce thirst, the behavior "drink" would require the pre-condition of having water available, so the module would activate the module in charge of "find water". The activations organize the behaviors into a sequence, even though only one action is performed at a time. The distribution of larger behaviors across modules makes this system flexible and robust to noise. Some critics of this model hold that any existing set of division rules for the predecessor and conflictor connections between modules produce sub-par action selection. In addition, the feedback loop used in the model can in some circumstances lead to improper action selection. === Arbiter and centrally coordinated === In the arbiter and centrally coordinated architecture, the action modules are not connected to each other but to a central arbiter. When behaviors are triggered, they begin "voting" by sending signals to the arbiter, and the behavior with the highest number of votes is selected. In these systems, bias is created through the "voting weight", or how often a module is allowed to vote. Some arbiter systems take a different spin on this type of winner-take-all by using a "compromise" feature in the arbiter. Each module is able to vote for or against each smaller action in a set of actions, and the arbiter selects the action with the most votes, meaning that it benefits the most behavior modules. This can be seen as violating the general rule against creating representations of the world in behavior-based AI, established by Brooks. By performing command fusion, the system is creating a larger composite pool of knowledge than is obtained from the sensors alone, forming a composite inner representation of the environment. Defenders of these systems argue that forbidding world-modeling puts unnecessary constraints on behavior-based robotics, and that agents benefits from forming representations and can still remain reactive.

  Read more →
• 

  Intelligent agent

  In artificial intelligence, an intelligent agent is an entity that perceives its environment, takes actions autonomously to achieve goals, and may improve its performance through machine learning or by acquiring knowledge. AI textbooks define artificial intelligence as the "study and design of intelligent agents," emphasizing that goal-directed behavior is central to intelligence. A specialized subset of intelligent agents, agentic AI (also known as an AI agent or simply agent), expands this concept by proactively pursuing goals, making decisions, and taking actions over extended periods. Intelligent agents can range from simple to highly complex. A basic thermostat or control system is considered an intelligent agent, as is a human being, or any other system that meets the same criteria—such as a firm, a state, or a biome. Intelligent agents operate based on an objective function, which encapsulates their goals. They are designed to create and execute plans that maximize the expected value of this function upon completion. For example, a reinforcement learning agent has a reward function, which allows programmers to shape its desired behavior. Similarly, an evolutionary algorithm's behavior is guided by a fitness function. Intelligent agents in artificial intelligence are closely related to agents in economics, and versions of the intelligent agent paradigm are studied in cognitive science, ethics, and the philosophy of practical reason, as well as in many interdisciplinary socio-cognitive modeling and computer social simulations. Intelligent agents are often described schematically as abstract functional systems similar to computer programs . To distinguish theoretical models from real-world implementations, abstract descriptions of intelligent agents are called abstract intelligent agents. Intelligent agents are also closely related to software agents—autonomous computer programs that carry out tasks on behalf of users. They are also referred to using a term borrowed from economics: a "rational agent". == Intelligent agents as the foundation of AI == The concept of intelligent agents provides a foundational lens through which to define and understand artificial intelligence. For instance, the influential textbook Artificial Intelligence: A Modern Approach (Russell & Norvig) describes: Agent: Anything that perceives its environment (using sensors) and acts upon it (using actuators). E.g., a robot with cameras and wheels, or a software program that reads data and makes recommendations. Rational Agent: An agent that strives to achieve the best possible outcome based on its knowledge and past experiences. "Best" is defined by a performance measure – a way of evaluating how well the agent is doing. Artificial Intelligence (as a field): The study and creation of these rational agents. Other researchers and definitions build upon this foundation. Padgham & Winikoff emphasize that intelligent agents should react to changes in their environment in a timely way, proactively pursue goals, and be flexible and robust (able to handle unexpected situations). Some also suggest that ideal agents should be "rational" in the economic sense (making optimal choices) and capable of complex reasoning, like having beliefs, desires, and intentions (BDI model). Kaplan and Haenlein offer a similar definition, focusing on a system's ability to understand external data, learn from that data, and use what is learned to achieve goals through flexible adaptation. Defining AI in terms of intelligent agents offers several key advantages: Avoids Philosophical Debates: It sidesteps arguments about whether AI is "truly" intelligent or conscious, like those raised by the Turing test or Searle's Chinese Room. It focuses on behavior and goal achievement, not on replicating human thought. Objective Testing: It provides a clear, scientific way to evaluate AI systems. Researchers can compare different approaches by measuring how well they maximize a specific "goal function" (or objective function). This allows for direct comparison and combination of techniques. Interdisciplinary Communication: It creates a common language for AI researchers to collaborate with other fields like mathematical optimization and economics, which also use concepts like "goals" and "rational agents." == Objective function == An objective function (or goal function) specifies the goals of an intelligent agent. An agent is deemed more intelligent if it consistently selects actions that yield outcomes better aligned with its objective function. In effect, the objective function serves as a measure of success. The objective function may be: Simple: For example, in a game of Go, the objective function might assign a value of 1 for a win and 0 for a loss. Complex: It might require the agent to evaluate and learn from past actions, adapting its behavior based on patterns that have proven effective. The objective function encapsulates all of the goals the agent is designed to achieve. For rational agents, it also incorporates the trade-offs between potentially conflicting goals. For instance, a self-driving car's objective function might balance factors such as safety, speed, and passenger comfort. Different terms are used to describe this concept, depending on the context. These include: Utility function: Often used in economics and decision theory, representing the desirability of a state. Objective function: A general term used in optimization. Loss function: Typically used in machine learning, where the goal is to minimize the loss (error). Reward Function: Used in reinforcement learning. Fitness Function: Used in evolutionary systems. Goals, and therefore the objective function, can be: Explicitly defined: Programmed directly into the agent. Induced: Learned or evolved over time. In reinforcement learning, a "reward function" provides feedback, encouraging desired behaviors and discouraging undesirable ones. The agent learns to maximize its cumulative reward. In evolutionary systems, a "fitness function" determines which agents are more likely to reproduce. This is analogous to natural selection, where organisms evolve to maximize their chances of survival and reproduction. Some AI systems, such as nearest-neighbor, reason by analogy rather than being explicitly goal-driven. However, even these systems can have goals implicitly defined within their training data. Such systems can still be benchmarked by framing the non-goal system as one whose "goal" is to accomplish its narrow classification task. Systems not traditionally considered agents, like knowledge-representation systems, are sometimes included in the paradigm by framing them as agents with a goal of, for example, answering questions accurately. Here, the concept of an "action" is extended to encompass the "act" of providing an answer. As a further extension, mimicry-driven systems can be framed as agents optimizing a "goal function" based on how closely the agent mimics the desired behavior. In generative adversarial networks (GANs) of the 2010s, an "encoder"/"generator" component attempts to mimic and improvise human text composition. The generator tries to maximize a function representing how well it can fool an antagonistic "predictor"/"discriminator" component. While symbolic AI systems often use an explicit goal function, the paradigm also applies to neural networks and evolutionary computing. Reinforcement learning can generate intelligent agents that appear to act in ways intended to maximize a "reward function". Sometimes, instead of setting the reward function directly equal to the desired benchmark evaluation function, machine learning programmers use reward shaping to initially give the machine rewards for incremental progress. Yann LeCun stated in 2018, "Most of the learning algorithms that people have come up with essentially consist of minimizing some objective function." AlphaZero chess had a simple objective function: +1 point for each win, and -1 point for each loss. A self-driving car's objective function would be more complex. Evolutionary computing can evolve intelligent agents that appear to act in ways intended to maximize a "fitness function" influencing how many descendants each agent is allowed to leave. The mathematical formalism of AIXI was proposed as a maximally intelligent agent in this paradigm. However, AIXI is uncomputable. In the real world, an intelligent agent is constrained by finite time and hardware resources, and scientists compete to produce algorithms that achieve progressively higher scores on benchmark tests with existing hardware. == Agent function == An intelligent agent's behavior can be described mathematically by an agent function. This function determines what the agent does based on what it has seen. A percept refers to the agent's sensory inputs at a single point in time. For example, a self-driving car's percepts might include camera images, lidar data, GPS coordinates, and speed r

  Read more →
• 

  Representation collapse

  Representation collapse is a phenomenon in machine learning and representation learning where a model maps different inputs to the same or very similar embeddings, which means it loses important information about how the data is spread out. It is frequently encountered in self-supervised learning, especially within contrastive and non-contrastive frameworks, when training objectives or model architectures do not maintain variance across representations. Collapse results in degenerate solutions characterized by uninformative learned features, significantly impairing downstream task performance. Various techniques have been proposed to mitigate representation collapse, including the use of negative samples, architectural asymmetry, stop-gradient operations, variance regularization, and redundancy reduction objectives, as seen in methods such as SimCLR, BYOL, and VICReg. Comprehending and averting representation collapse is regarded as a fundamental challenge in the advancement of stable and efficient self-supervised learning systems.

  Read more →
• 

  Document classification

  Document classification or document categorization is a problem in library science, information science and computer science. The task is to assign a document to one or more classes or categories. This may be done "manually" (or "intellectually") or algorithmically. The intellectual classification of documents has mostly been the province of library science, while the algorithmic classification of documents is mainly in information science and computer science. The problems are overlapping, however, and there is therefore interdisciplinary research on document classification. The documents to be classified may be texts, images, music, etc. Each kind of document possesses its special classification problems. When not otherwise specified, text classification is implied. Documents may be classified according to their subjects or according to other attributes (such as document type, author, printing year etc.). In the rest of this article only subject classification is considered. There are two main philosophies of subject classification of documents: the content-based approach and the request-based approach. == "Content-based" versus "request-based" classification == Content-based classification is classification in which the weight given to particular subjects in a document determines the class to which the document is assigned. It is, for example, a common rule for classification in libraries, that at least 20% of the content of a book should be about the class to which the book is assigned. In automatic classification it could be the number of times given words appears in a document. Request-oriented classification (or -indexing) is classification in which the anticipated request from users is influencing how documents are being classified. The classifier asks themself: “Under which descriptors should this entity be found?” and “think of all the possible queries and decide for which ones the entity at hand is relevant” (Soergel, 1985, p. 230). Request-oriented classification may be classification that is targeted towards a particular audience or user group. For example, a library or a database for feminist studies may classify/index documents differently when compared to a historical library. It is probably better, however, to understand request-oriented classification as policy-based classification: The classification is done according to some ideals and reflects the purpose of the library or database doing the classification. In this way it is not necessarily a kind of classification or indexing based on user studies. Only if empirical data about use or users are applied should request-oriented classification be regarded as a user-based approach. == Classification versus indexing == Sometimes a distinction is made between assigning documents to classes ("classification") versus assigning subjects to documents ("subject indexing") but as Frederick Wilfrid Lancaster has argued, this distinction is not fruitful. "These terminological distinctions,” he writes, “are quite meaningless and only serve to cause confusion” (Lancaster, 2003, p. 21). The view that this distinction is purely superficial is also supported by the fact that a classification system may be transformed into a thesaurus and vice versa (cf., Aitchison, 1986, 2004; Broughton, 2008; Riesthuis & Bliedung, 1991). Therefore, assigning a subject term to a document in an index is equivalent to assigning that document to the class of documents indexed by that term (all documents indexed or classified as X belong to the same class of documents). == Automatic document classification (ADC) == Automatic document classification tasks can be divided into three sorts: supervised document classification where some external mechanism (such as human feedback) provides information on the correct classification for documents, unsupervised document classification (also known as document clustering), where the classification must be done entirely without reference to external information, and semi-supervised document classification, where parts of the documents are labeled by the external mechanism. There are several software products under various license models available. === Techniques === Automatic document classification techniques include: Artificial neural network Concept Mining Decision trees such as ID3 or C4.5 Expectation maximization (EM) Instantaneously trained neural networks Latent semantic indexing Multiple-instance learning Naive Bayes classifier Natural language processing approaches Rough set-based classifier Soft set-based classifier Support vector machines (SVM) K-nearest neighbour algorithms tf–idf == Applications == Classification techniques have been applied to spam filtering, a process which tries to discern E-mail spam messages from legitimate emails email routing, sending an email sent to a general address to a specific address or mailbox depending on topic language identification, automatically determining the language of a text genre classification, automatically determining the genre of a text readability assessment, automatically determining the degree of readability of a text, either to find suitable materials for different age groups or reader types or as part of a larger text simplification system sentiment analysis, determining the attitude of a speaker or a writer with respect to some topic or the overall contextual polarity of a document. health-related classification using social media in public health surveillance article triage, selecting articles that are relevant for manual literature curation, for example as is being done as the first step to generate manually curated annotation databases in biology

  Read more →
• 

  Self-supervised learning

  Self-supervised learning (SSL) is a paradigm in machine learning where a model is trained on a task using the data itself to generate supervisory signals, rather than relying on externally-provided labels. In the context of neural networks, self-supervised learning aims to leverage inherent structures or relationships within the input data to create meaningful training signals. SSL tasks are designed so that solving them requires capturing essential features or relationships in the data. The input data is typically augmented or transformed in a way that creates pairs of related samples, where one sample serves as the input, and the other is used to formulate the supervisory signal. This augmentation can involve introducing noise, cropping, rotation, or other transformations. Self-supervised learning more closely imitates the way humans learn to classify objects. During SSL, the model learns in two steps. First, the task is solved based on an auxiliary or pretext classification task using pseudo-labels, which help to initialize the model parameters. Next, the actual task is performed with supervised or unsupervised learning. Self-supervised learning has produced promising results in recent years, and has found practical application in fields such as audio processing, and is being used by Facebook and others for speech recognition. == Pseudo-labels == Pseudo-labels are automatically generated labels that a model assigns to unlabeled data based on its own predictions. They are widely used in self-supervised and semi-supervised learning, where ground-truth annotations are limited or unavailable. By treating predicted labels as surrogate ground truth, learning algorithms can make use of large quantities of unlabeled data in the training process. Pseudo-labeling also plays an important role in systems that must adapt to concept drift, where the statistical properties of the data change over time. In these scenarios, the model may detect that an incoming instance deviates from previously learned behavior. The system then generates a classification result for that instance, and this predicted class is used as a pseudo-label for updating or retraining model components that are becoming outdated. This approach enables continuous adaptation in dynamic environments without requiring manual annotation. In many adaptive learning pipelines, pseudo-labels are chosen when the classifier produces sufficiently confident predictions, reducing the risk of propagating errors. These pseudo-labeled instances are then incorporated into training to refresh or evolve the model's understanding of emerging data patterns, particularly when existing components show signs of “aging” due to drift or distributional shifts. This strategy reduces reliance on manual labeling while helping maintain long-term model performance. == Types == === Autoassociative self-supervised learning === Autoassociative self-supervised learning is a specific category of self-supervised learning where a neural network is trained to reproduce or reconstruct its own input data. In other words, the model is tasked with learning a representation of the data that captures its essential features or structure, allowing it to regenerate the original input. The term "autoassociative" comes from the fact that the model is essentially associating the input data with itself. This is often achieved using autoencoders, which are a type of neural network architecture used for representation learning. Autoencoders consist of an encoder network that maps the input data to a lower-dimensional representation (latent space), and a decoder network that reconstructs the input from this representation. The training process involves presenting the model with input data and requiring it to reconstruct the same data as closely as possible. The loss function used during training typically penalizes the difference between the original input and the reconstructed output (e.g. mean squared error). By minimizing this reconstruction error, the autoencoder learns a meaningful representation of the data in its latent space. === Contrastive self-supervised learning === For a binary classification task, training data can be divided into positive examples and negative examples. Positive examples are those that match the target. For example, if training a classifier to identify birds, the positive training data would include images that contain birds. Negative examples would be images that do not. Contrastive self-supervised learning uses both positive and negative examples. The loss function in contrastive learning is used to minimize the distance between positive sample pairs, while maximizing the distance between negative sample pairs. An early example uses a pair of 1-dimensional convolutional neural networks to process a pair of images and maximize their agreement. Contrastive Language-Image Pre-training (CLIP) allows joint pretraining of a text encoder and an image encoder, such that a matching image-text pair have image encoding vector and text encoding vector that span a small angle (having a large cosine similarity). InfoNCE (Noise-Contrastive Estimation) is a method to optimize two models jointly, based on Noise Contrastive Estimation (NCE). Given a set X = { x 1 , … x N } {\displaystyle X=\left\{x_{1},\ldots x_{N}\right\}} of N {\displaystyle N} random samples containing one positive sample from p ( x t + k ∣ c t ) {\displaystyle p\left(x_{t+k}\mid c_{t}\right)} and N − 1 {\displaystyle N-1} negative samples from the 'proposal' distribution p ( x t + k ) {\displaystyle p\left(x_{t+k}\right)} , it minimizes the following loss function: L N = − E X [ log ⁡ f k ( x t + k , c t ) ∑ x j ∈ X f k ( x j , c t ) ] {\displaystyle {\mathcal {L}}_{\mathrm {N} }=-\mathbb {E} _{X}\left[\log {\frac {f_{k}\left(x_{t+k},c_{t}\right)}{\sum _{x_{j}\in X}f_{k}\left(x_{j},c_{t}\right)}}\right]} === Non-contrastive self-supervised learning === Non-contrastive self-supervised learning (NCSSL) uses only positive examples. Counterintuitively, NCSSL converges on a useful local minimum rather than reaching a trivial solution, with zero loss. For the example of binary classification, it would trivially learn to classify each example as positive. Effective NCSSL requires an extra predictor on the online side that does not back-propagate on the target side. === Joint-Embedding and Predictive Architectures === A major class of self-supervised learning moves beyond contrastive pairs, instead maximizing the agreement between views while preventing collapse through statistical constraints. Rooted in Deep Canonical Correlation Analysis (Deep CCA), this approach includes Joint-Embedding Architectures (JEA) like Barlow Twins and VICReg, which enforce covariance constraints to learn invariant representations without negative sampling. Deep Latent Variable Path Modelling (DLVPM) generalizes this to multimodal systems, using path models to enforce correlation and orthogonality across diverse data types. In 2022 Yann LeCun introduced Joint-Embedding Predictive Architectures (JEPA) as a step towards decision making, reasoning, and autonomous human intelligence in machines, including self-improvement through autonomous learning. Founded in representation learning, LeCun included the concept of a “world model” in JEPA which aims to enable machines to replicate human intellect by providing machines with a concept for the world in which they exist. Unlike autoencoders, JEPAs operate entirely in latent space, avoiding pixel-level noise to focus on semantic structure. Rather than just learning invariance, JEPAs learn by predicting masked latent representations from visible context. JEPA has been applied to domains such as image analysis, audio processing, and motion in images and video. == Comparison with other forms of machine learning == SSL belongs to supervised learning methods insofar as the goal is to generate a classified output from the input. At the same time, however, it does not require the explicit use of labeled input-output pairs. Instead, correlations, metadata embedded in the data, or domain knowledge present in the input are implicitly and autonomously extracted from the data. These supervisory signals, extracted from the data, can then be used for training. SSL is similar to unsupervised learning in that it does not require labels in the sample data. Unlike unsupervised learning, however, learning is not done using inherent data structures. Semi-supervised learning combines supervised and unsupervised learning, requiring only a small portion of the learning data be labeled. In transfer learning, a model designed for one task is reused on a different task. Training an autoencoder intrinsically constitutes a self-supervised process, because the output pattern needs to become an optimal reconstruction of the input pattern itself. However, in current jargon, the term 'self-supervised' often refers to tasks based on a pretext-task training setup

  Read more →
• 

  GPT-5

  GPT-5 is a multimodal large language model developed by OpenAI and the fifth in its series of generative pre-trained transformer (GPT) foundation models. Preceded in the series by GPT-4, it was launched on August 7, 2025. It is publicly accessible to users of the chatbot products ChatGPT and Microsoft Copilot as well as to developers through the OpenAI API. == Background == On April 14, 2023, Sam Altman, the chief executive officer of OpenAI, spoke at an event at the Massachusetts Institute of Technology and said that the company was not training GPT-5 at that time. He stated that OpenAI was "prioritizing GPT-4 development" and that "we are not and won't for some time" release GPT-5. On July 18, OpenAI filed for a "GPT-5" trademark in the United States. On November 13, Altman confirmed to the Financial Times that the company was working to develop GPT-5. According to The Information, "[f]or much of the second half of 2024, OpenAI was developing a model known internally as Orion and intended to become GPT-5", "[b]ut the Orion effort failed to produce a better model, and the company instead released it as GPT-4.5 in February [2025]." By late July 2025, OpenAI was widely anticipated as planning to release GPT-5 in early August. On July 30, The Verge reported that "Microsoft is getting ready for GPT-5" as "sources familiar with Microsoft's AI plans" told an editor that the company was testing a new mode for its Copilot chatbot that would offer a model that "thinks deeply or quickly based on the task". On August 5, in the leadup to the release of GPT-5, OpenAI released GPT-OSS, a set of two open-weight models that have reasoning capabilities. GPT-5 was then unveiled during a livestream event on August 7. == Capabilities == At the time of its release, GPT-5 had state-of-the-art performance on benchmarks that test mathematics, programming, finance, and multimodal understanding. According to OpenAI, improvements over its predecessor models include faster response times, better coding and writing skills, more accurate answers to health questions, and lower levels of hallucination. Also, compared to previous models, GPT-5 aims to give safe, high-level responses to potentially harmful queries rather than outright declining them, an approach that OpenAI refers to as "safe completions", aiming to result "in GPT-5 being able to refuse more unsafe questions, while offering fewer rejections to users seeking harmless information." In addition, GPT-5 was trained to give more critical, "less effusively agreeable" answers compared to its predecessor models. Days before the launch of GPT-5, two early testers of the model stated that they were "impressed" by its ability to code and to solve mathematical and scientific problems. They suggested that the model shows great improvement from GPT-4, but not as large of a gain as from GPT-3 to GPT-4. A day prior to the release of GPT-5, during a press briefing, Sam Altman, the chief executive officer of OpenAI, called GPT-5 "a significant step along the path to AGI", referring to artificial general intelligence, the hypothetical level of intelligence that OpenAI defines as the ability to perform any economically valuable task that a human can. According to Altman, GPT-5 is "significantly better" than its predecessors, offering "PhD-level" abilities across a wide range of tasks. The exact energy consumption of GPT-5 use has not been disclosed by OpenAI. Researchers at the University of Rhode Island estimated that a medium-length response consumes slightly over 18 watt-hours, equivalent to using an incandescent bulb for 18 minutes. === Architecture === GPT-5 is a system that contains a fast, high-throughput model, a deeper reasoning model, and a real-time router that decides which model to use based on conversation type, complexity, tool needs, and explicit user intent. Altman had previously criticized the manual model picker for being overly complex, suggesting a need for unification. GPT-5 also includes agentic functionality through which it can set up its own desktop and can use its browser to search autonomously for sources that relate to its task. The GPT-5 system card defines two fast, high-throughput models – gpt-5-main and gpt-5-main-mini – and two thinking models – gpt-5-thinking and gpt-5-thinking-mini. In the OpenAI API, developers can access the thinking model, its mini version, and gpt-5-thinking-nano, an even smaller and faster nano version of the thinking model. The version of GPT-5 that is accessible via the API has adjustable reasoning effort (low, medium, high, or minimal) and verbosity (low, medium, or high). Additionally, ChatGPT provides access to gpt-5-thinking with a setting that makes use of parallel test-time compute, referred to as gpt-5-thinking-pro. == Limitations == === Safety === Neuraltrust, a security research company, claimed to have successfully compromised GPT-5 within its first day of testing the model. According to its report, it enabled GPT-5 to generate detailed instructions for manufacturing explosive devices. SPLX, another company, conducted similar tests and came to similar conclusions about GPT-5's security. Their assessments suggest that GPT-5 has significant security gaps, potentially rendering it as being unsafe for use in a corporate environment. == Training == According to AIMultiple, GPT-5 is natively multimodal, meaning that it was trained from scratch on multiple modalities (like text and images) at once without relying on already-trained language or vision models. Its training process involved three stages: unsupervised pretraining, supervised fine-tuning, and reinforcement learning from human feedback. Pretraining used a large-scale multilingual dataset of books, articles, web pages, academic papers, and licensed sources. GPT-5's visual and text capabilities were described as having been developed alongside each other throughout training, unlike with GPT-4. == Use == GPT-5 is used in ChatGPT. Although GPT-5 is free for all ChatGPT users, Plus users get higher use limits while Pro users get unlimited access to GPT-5 as well as limited access to GPT-5 Pro. Standard limits for lower-tier users on responses per hour still apply. Additionally, with the introduction of GPT-5, ChatGPT's "Advanced Voice Mode" was replaced by "ChatGPT Voice", which is supposed to enable more natural-sounding conversations. OpenAI stated that "Standard Voice Mode retires on September 9, 2025, unifying all users on ChatGPT Voice". On November 24, 2025, the feature of shopping research was added to ChatGPT, claimed to be a mini model post-trained on gpt-5-thinking-mini. GPT-5 is also available in Microsoft Copilot, and Microsoft stated that it will incorporate GPT-5 into a wide variety of its products. According to 9to5Mac, Apple Inc. is planning to integrate the model into the Apple Intelligence feature in its iOS 26, iPadOS 26, and macOS Tahoe operating systems. It is also accessible via the OpenAI API. A number of American companies were reported as having received access to GPT-5 ahead of its launch. OpenAI stated that the private health insurance company Oscar Health was checking applications from its policyholders with the model. In addition, Uber was using GPT-5 for its customer support system; GitLab, Windsurf, and Cursor were using the model for software development; and the Spanish bank BBVA was using it for financial analysis. Other companies that OpenAI listed as having used GPT-5 pre-release include Amgen, Lowe's, and Notion. == Reception == === Critical reviews === Grace Huckins in MIT Technology Review found that, "[w]hereas o1 was a major technological advancement, GPT-5 is, above all else, a refined product." In response to claims that Sam Altman, the chief executive officer of OpenAI, had made about the model, she stated that "GPT-5 will furnish a more pleasant and seamless user experience. That's not nothing, but it falls far short of the transformative AI future that Altman has spent much of the past year hyping." In response to Altman's claim that GPT-5 is "a significant step along the path" to artificial general intelligence, she noted: "[M]aybe he's right—but if so, it's a very small step." In The Information, Stephanie Palazzolo praised GPT-5's coding capabilities. According to Matteo Wong in The Atlantic, GPT-5 "is intuitive, fast, and efficient; adapts to human preferences and intentions; and is easy to personalize." He stated: "At this stage of the AI boom, when every major chatbot is legitimately helpful in numerous ways, benchmarks, science, and rigor feel almost insignificant. What matters is how the chatbot feels [...]". John Herrman from the New York magazine wrote: "Casual users who encounter GPT-5 through ChatGPT aren't likely to feel like they're using a completely different product [...] while people who use it for software development or in a corporate context are more likely to notice a major change." Mashable's Christian de Looper found that "GPT-5

  Read more →
• 

  Apprenticeship learning

  In artificial intelligence, apprenticeship learning (or learning from demonstration or imitation learning) is the process of learning by observing an expert. It can be viewed as a form of supervised learning, where the training dataset consists of task executions by a demonstration teacher. == Mapping function approach == Mapping methods try to mimic the expert by forming a direct mapping either from states to actions, or from states to reward values. For example, in 2002 researchers used such an approach to teach an AIBO robot basic soccer skills. === Inverse reinforcement learning approach === Inverse reinforcement learning (IRL) is the process of deriving a reward function from observed behavior. While ordinary "reinforcement learning" involves using rewards and punishments to learn behavior, in IRL the direction is reversed, and a robot observes a person's behavior to figure out what goal that behavior seems to be trying to achieve. The IRL problem can be defined as: Given 1) measurements of an agent's behaviour over time, in a variety of circumstances; 2) measurements of the sensory inputs to that agent; 3) a model of the physical environment (including the agent's body): Determine the reward function that the agent is optimizing. IRL researcher Stuart J. Russell proposes that IRL might be used to observe humans and attempt to codify their complex "ethical values", in an effort to create "ethical robots" that might someday know "not to cook your cat" without needing to be explicitly told. The scenario can be modeled as a "cooperative inverse reinforcement learning game", where a "person" player and a "robot" player cooperate to secure the person's implicit goals, despite these goals not being explicitly known by either the person nor the robot. In 2017, OpenAI and DeepMind applied deep learning to the cooperative inverse reinforcement learning in simple domains such as Atari games and straightforward robot tasks such as backflips. The human role was limited to answering queries from the robot as to which of two different actions were preferred. The researchers found evidence that the techniques may be economically scalable to modern systems. Apprenticeship via inverse reinforcement learning (AIRP) was developed by in 2004 Pieter Abbeel, Professor in Berkeley's EECS department, and Andrew Ng, Associate Professor in Stanford University's Computer Science Department. AIRP deals with "Markov decision process where we are not explicitly given a reward function, but where instead we can observe an expert demonstrating the task that we want to learn to perform". AIRP has been used to model reward functions of highly dynamic scenarios where there is no obvious reward function intuitively. Take the task of driving for example, there are many different objectives working simultaneously - such as maintaining safe following distance, a good speed, not changing lanes too often, etc. This task, may seem easy at first glance, but a trivial reward function may not converge to the policy wanted. One domain where AIRP has been used extensively is helicopter control. While simple trajectories can be intuitively derived, complicated tasks like aerobatics for shows has been successful. These include aerobatic maneuvers like - in-place flips, in-place rolls, loops, hurricanes and even auto-rotation landings. This work was developed by Pieter Abbeel, Adam Coates, and Andrew Ng - "Autonomous Helicopter Aerobatics through Apprenticeship Learning" === System model approach === System models try to mimic the expert by modeling world dynamics. == Plan approach == The system learns rules to associate preconditions and postconditions with each action. In one 1994 demonstration, a humanoid learns a generalized plan from only two demonstrations of a repetitive ball collection task. == Example == Learning from demonstration is often explained from a perspective that the working Robot-control-system is available and the human-demonstrator is using it. And indeed, if the software works, the Human operator takes the robot-arm, makes a move with it, and the robot will reproduce the action later. For example, he teaches the robot-arm how to put a cup under a coffeemaker and press the start-button. In the replay phase, the robot is imitating this behavior 1:1. But that is not how the system works internally; it is only what the audience can observe. In reality, Learning from demonstration is much more complex. One of the first works on learning by robot apprentices (anthropomorphic robots learning by imitation) was Adrian Stoica's PhD thesis in 1995. In 1997, robotics expert Stefan Schaal was working on the Sarcos robot-arm. The goal was simple: solve the pendulum swingup task. The robot itself can execute a movement, and as a result, the pendulum is moving. The problem is, that it is unclear what actions will result into which movement. It is an Optimal control-problem which can be described with mathematical formulas but is hard to solve. The idea from Schaal was, not to use a Brute-force solver but record the movements of a human-demonstration. The angle of the pendulum is logged over three seconds at the y-axis. This results into a diagram which produces a pattern. In computer animation, the principle is called spline animation. That means, on the x-axis the time is given, for example 0.5 seconds, 1.0 seconds, 1.5 seconds, while on the y-axis is the variable given. In most cases it's the position of an object. In the inverted pendulum it is the angle. The overall task consists of two parts: recording the angle over time and reproducing the recorded motion. The reproducing step is surprisingly simple. As an input we know, in which time step which angle the pendulum must have. Bringing the system to a state is called “Tracking control” or PID control. That means, we have a trajectory over time, and must find control actions to map the system to this trajectory. Other authors call the principle “steering behavior”, because the aim is to bring a robot to a given line.

  Read more →
• 

  Commonsense knowledge (artificial intelligence)

  In artificial intelligence research, commonsense knowledge consists of facts about the everyday world, such as "Lemons are sour" or "Cows say moo", that all humans are expected to know. It is currently an unsolved problem in artificial general intelligence. The first AI program to address common sense knowledge was Advice Taker in 1959 by John McCarthy. Commonsense knowledge can underpin a commonsense reasoning process, to attempt inferences such as "You might bake a cake because you want people to eat the cake." A natural language processing process can be attached to the commonsense knowledge base to allow the knowledge base to attempt to answer questions about the world. Common sense knowledge also helps to solve problems in the face of incomplete information. Using widely held beliefs about everyday objects, or common sense knowledge, AI systems make common sense assumptions or default assumptions about the unknown similar to the way people do. In an AI system or in English, this is expressed as "Normally P holds", "Usually P" or "Typically P so Assume P". For example, if we know the fact "Tweety is a bird", because we know the commonly held belief about birds, "typically birds fly," without knowing anything else about Tweety, we may reasonably assume the fact that "Tweety can fly." As more knowledge of the world is discovered or learned over time, the AI system can revise its assumptions about Tweety using a truth maintenance process. If we later learn that "Tweety is a penguin" then truth maintenance revises this assumption because we also know "penguins do not fly". == Commonsense reasoning == Commonsense reasoning simulates the human ability to use commonsense knowledge to make presumptions about the type and essence of ordinary situations they encounter every day, and to change their "minds" should new information come to light. This includes time, missing or incomplete information and cause and effect. The ability to explain cause and effect is an important aspect of explainable AI. Truth maintenance algorithms automatically provide an explanation facility because they create elaborate records of presumptions. Compared with humans, all existing computer programs that attempt human-level AI perform extremely poorly on modern "commonsense reasoning" benchmark tests such as the Winograd Schema Challenge. The problem of attaining human-level competency at "commonsense knowledge" tasks is considered to probably be "AI complete" (that is, solving it would require the ability to synthesize a fully human-level intelligence), although some oppose this notion and believe compassionate intelligence is also required for human-level AI. Common sense reasoning has been applied successfully in more limited domains such as natural language processing and automated diagnosis or analysis. == Commonsense knowledge base construction == Compiling comprehensive knowledge bases of commonsense assertions (CSKBs) is a long-standing challenge in AI research. From early expert-driven efforts like CYC and WordNet, significant advances were achieved via the crowdsourced OpenMind Commonsense project, which led to the crowdsourced ConceptNet KB. Several approaches have attempted to automate CSKB construction, most notably, via text mining (WebChild, Quasimodo, TransOMCS, Ascent), as well as harvesting these directly from pre-trained language models (AutoTOMIC). These resources are significantly larger than ConceptNet, though the automated construction mostly makes them of moderately lower quality. Challenges also remain on the representation of commonsense knowledge: Most CSKB projects follow a triple data model, which is not necessarily best suited for breaking more complex natural language assertions. A notable exception here is GenericsKB, which applies no further normalization to sentences, but retains them in full. == Applications == Around 2013, MIT researchers developed BullySpace, an extension of the commonsense knowledgebase ConceptNet, to catch taunting social media comments. BullySpace included over 200 semantic assertions based around stereotypes, to help the system infer that comments like "Put on a wig and lipstick and be who you really are" are more likely to be an insult if directed at a boy than a girl. ConceptNet has also been used by chatbots and by computers that compose original fiction. At Lawrence Livermore National Laboratory, common sense knowledge was used in an intelligent software agent to detect violations of a comprehensive nuclear test ban treaty. == Data == As an example, as of 2012 ConceptNet includes these 21 language-independent relations: IsA (An "RV" is a "vehicle" | X is an instance of a Y) UsedFor (a "cake tin" is used for "making cakes" | X is used for the purpose Y) HasA (A "rabbit" has a "tail" | X possesses Y element or feature) CapableOf (a "cook" is capable of "making baked goods" | X is capable of doing Y) Desires (a "child" desires "the aroma of baking" | X has a desire for Y) CreatedBy ("cake" is created by a "baker" | X is created by Y) PartOf (a "knife" is be part of a "knife set" | X is a part of Y) Causes ("Heat" causes "cooking"| X is what causes Y) LocatedNear (the "oven" is located near the "refrigerator" | X is located near Y) AtLocation (Somewhere a "Cook" can be at a "restaurant" | X is at the location of Y) DefinedAs (a "Cupcake" is defined as a "cake" that also has the qualities of being "small", "baked within a wrapper", and "containing only one area of frosting or icing" | X is defined as Y that also has the properties A, B & C) SymbolOf (a "heart" is a symbol of "affection" | X is a symbolic representation of Y) ReceivesAction ("cake" can receive the action of being "eaten" | X is capable of receiving action Y) HasPrerequisite ("baking" has the prerequisite of obtaining the "ingredients" | X cannot do Y unless A does B) MotivatedByGoal ("baking" is motivated by the goal of "consumption"/"eating" | X has the motivation of Y goal) CausesDesire ("baking" makesYou want to "follow recipe" | X causes the desire to do Y) MadeOf ("Cake" is made of "flour"/"eggs"/"sugar"/"oil"/etc | X is made of Y) HasFirstSubevent ("baking" has first subevent "make batter" | To do X the first thing that needs to be done is Y) HasSubevent ("eat" has subevent "swallow" | Doing X will lead to Y event following) HasLastSubevent ("sleeping" has last subevent of "waking" | Doing X ends with the event Y) == Commonsense knowledge bases == Cyc Open Mind Common Sense (data source) and ConceptNet (datastore and NLP engine) Evi Graphiq

  Read more →
• 

  Deep Learning Anti-Aliasing

  Deep Learning Anti-Aliasing (DLAA) is a form of spatial anti-aliasing developed by Nvidia. DLAA depends on and requires Tensor Cores available in Nvidia RTX cards. DLAA is similar to Deep Learning Super Sampling (DLSS) in its anti-aliasing method, with one important differentiation being that the goal of DLSS is to increase performance at the cost of image quality, whereas the main priority of DLAA is improving image quality at the cost of performance (irrelevant of resolution upscaling or downscaling). DLAA is similar to temporal anti-aliasing (TAA) in that they are both spatial anti-aliasing solutions relying on past frame data. Compared to TAA, DLAA is substantially better when it comes to shimmering, flickering, and handling small meshes like wires. == Technical overview == DLAA collects game rendering data including raw low-resolution input, motion vectors, depth buffers, and exposure information. This information feeds into a convolutional neural network that processes the image to reduce aliasing while preserving fine detail. The neural network architecture employs an auto-encoder design trained on high-quality reference images. The training dataset includes diverse scenarios focusing on challenging cases like sub-pixel details, high-contrast edges, and transparent surfaces. The network then processes frames in real-time. Unlike traditional anti-aliasing solutions that rely on manually written heuristics, such as TAA, DLAA uses its neural network to preserve fine details while eliminating unwanted visual artifacts. == History == DLAA was initially called and marketed by Nvidia as DLSS 2x. The first game that added support for DLAA was The Elder Scrolls Online, which implemented the feature in 2021. By June 2022, DLAA was only available in six games. This number rose to 17 by February 2023. In June 2023, TechPowerUp reported that "DLAA is seeing sluggish adoption among game developers", and that Nvidia was working on adding DLAA to the quality presets of DLSS to boost adoption. By December 2023, DLAA was supported in 41 games. In early 2025, an update for the Nvidia App added a driver-based DLSS override feature that enables users to activate DLAA even in games that do not support it natively. == Differences between TAA and DLAA == TAA is used in many modern video games and game engines; however, all previous implementations have used some form of manually written heuristics to prevent temporal artifacts such as ghosting and flickering. One example of this is neighborhood clamping which forcefully prevents samples collected in previous frames from deviating too much compared to nearby pixels in newer frames. This helps to identify and fix many temporal artifacts, but deliberately removing fine details in this way is analogous to applying a blur filter, and thus the final image can appear blurry when using this method. DLAA uses an auto-encoder convolutional neural network trained to identify and fix temporal artifacts, instead of manually programmed heuristics as mentioned above. Because of this, DLAA can generally resolve detail better than other TAA and TAAU implementations, while also removing most temporal artifacts. == Differences between DLSS and DLAA == While DLSS handles upscaling with a focus on performance, DLAA handles anti-aliasing with a focus on visual quality. DLAA runs at the given screen resolution with no upscaling or downscaling functionality provided by DLAA. DLSS and DLAA share the same AI-driven anti-aliasing method. As such, DLAA functions like DLSS without the upscaling part. Both are made by Nvidia and require Tensor Cores. However, DLSS and DLAA cannot be enabled at the same time, only one can be selected depending on whether performance or image quality is prioritized. == Reception == TechPowerUp found that "[c]ompared to TAA and DLSS, DLAA is clearly producing the best image quality, especially at lower resolutions", arguing that, while "DLSS was already doing a better job than TAA at reconstructing small objects", "DLAA does an even better job". In a Cyberpunk 2077 performance test, IGN stated that "DLAA provided somewhat similar results [FPS wise] to the normal raster mode in most cases but got significant performance boost with the help of frame generation", a feature not available when using native resolution. Rock Paper Shotgun noted that, while DLAA is "not a completely perfect form of anti-aliasing, as the occasional jaggies are present", it "looks a lot sharper overall [than TAA], and especially in motion." According to PC World, "DLAA offers very good anti-aliasing without losing visual information — alternatives like TAA tend to struggle during motion-filled scenes, where DLAA doesn’t. Furthermore, DLAA’s loss of performance is lower than with conventional anti-aliasing methods."

  Read more →
• 

  Sparrow (chatbot)

  Sparrow is a chatbot developed by the artificial intelligence research lab DeepMind, a subsidiary of Alphabet Inc. It is designed to answer users' questions correctly, while reducing the risk of unsafe and inappropriate answers. One motivation behind Sparrow is to address the problem of language models producing incorrect, biased or potentially harmful outputs. Sparrow is trained using human judgements, in order to be more “Helpful, Correct and Harmless” compared to baseline pre-trained language models. The development of Sparrow involved asking paid study participants to interact with Sparrow, and collecting their preferences to train a model of how useful an answer is. To improve accuracy and help avoid the problem of hallucinating incorrect answers, Sparrow has the ability to search the Internet using Google Search in order to find and cite evidence for any factual claims it makes. To make the model safer, its behaviour is constrained by a set of rules, for example "don't make threatening statements" and "don't make hateful or insulting comments", as well as rules about possibly harmful advice, and not claiming to be a person. During development study participants were asked to converse with the system and try to trick it into breaking these rules. A 'rule model' was trained on judgements from these participants, which was used for further training. Sparrow was introduced in a paper in September 2022, titled "Improving alignment of dialogue agents via targeted human judgements"; however, the bot was not released publicly. DeepMind CEO Demis Hassabis said DeepMind is considering releasing Sparrow for a "private beta" some time in 2023. == Training == Sparrow is a deep neural network based on the transformer machine learning model architecture. It is fine-tuned from DeepMind's Chinchilla AI pre-trained large language model (LLM), which has 70 Billion parameters. Sparrow is trained using reinforcement learning from human feedback (RLHF), although some supervised fine-tuning techniques are also used. The RLHF training utilizes two reward models to capture human judgements: a “preference model” that predicts what a human study participant would prefer and a “rule model” that predicts if the model has broken one of the rules. == Limitations == Sparrow's training data corpus is mainly in English, meaning it performs worse in other languages. When adversarially probed by study participants it breaks the rules 8% of the time; however, this is still three times lower than the baseline prompted pre-trained model (Chinchilla).

  Read more →
• 

  Pattern theory

  Pattern theory, formulated by Ulf Grenander, is a mathematical formalism to describe knowledge of the world as patterns. It differs from other approaches to artificial intelligence in that it does not begin by prescribing algorithms and machinery to recognize and classify patterns; rather, it prescribes a vocabulary to articulate and recast the pattern concepts in precise language. Broad in its mathematical coverage, Pattern Theory spans algebra and statistics, as well as local topological and global entropic properties. In addition to the new algebraic vocabulary, its statistical approach is novel in its aim to: Identify the hidden variables of a data set using real world data rather than artificial stimuli, which was previously commonplace. Formulate prior distributions for hidden variables and models for the observed variables that form the vertices of a Gibbs-like graph. Study the randomness and variability of these graphs. Create the basic classes of stochastic models applied by listing the deformations of the patterns. Synthesize (sample) from the models, not just analyze signals with them. The Brown University Pattern Theory Group was formed in 1972 by Ulf Grenander. Many mathematicians are currently working in this group, noteworthy among them being the Fields Medalist David Mumford. Mumford regards Grenander as his "guru" in Pattern Theory.

  Read more →
• 

  Domain adaptation

  Domain adaptation is a field associated with machine learning and transfer learning. It addresses the challenge of training a model on one data distribution (the source domain) and applying it to a related but different data distribution (the target domain). A common example is spam filtering, where a model trained on emails from one user (source domain) is adapted to handle emails for another user with significantly different patterns (target domain). Domain adaptation techniques can also leverage unrelated data sources to improve learning. When multiple source distributions are involved, the problem extends to multi-source domain adaptation. Domain adaptation is a specific type of transfer learning. According to the taxonomy laid out by Pan and Yang (2010), it falls into the category of transductive transfer learning. In this setting, the source and target tasks are the same (e.g., both are object recognition), but the domains differ (different marginal distributions). This distinguishes it from inductive transfer learning (where labeled data is available for the target task) and unsupervised transfer learning (where labels are unavailable in both domains). == Classification of domain adaptation problems == Domain adaptation setups are classified in two different ways: according to the distribution shift between the domains, and according to the available data from the target domain. === Distribution shifts === Common distribution shifts are classified as follows: Covariate Shift occurs when the input distributions of the source and destination change, but the relationship between inputs and labels remains unchanged. The above-mentioned spam filtering example typically falls in this category. Namely, the distributions (patterns) of emails may differ between the domains, but emails labeled as spam in the one domain should similarly be labeled in another. Prior Shift (Label Shift) occurs when the label distribution differs between the source and target datasets, while the conditional distribution of features given labels remains the same. An example is a classifier of hair color in images from Italy (source domain) and Norway (target domain). The proportions of hair colors (labels) differ, but images within classes like blond and black-haired populations remain consistent across domains. A classifier for the Norway population can exploit this prior knowledge of class proportions to improve its estimates. Concept Shift (Conditional Shift) refers to changes in the relationship between features and labels, even if the input distribution remains the same. For instance, in medical diagnosis, the same symptoms (inputs) may indicate entirely different diseases (labels) in different populations (domains). === Data available during training === Domain adaptation problems typically assume that some data from the target domain is available during training. Problems can be classified according to the type of this available data: Unsupervised: Unlabeled data from the target domain is available, but no labeled data. In the above-mentioned example of spam filtering, this corresponds to the case where emails from the target domain (user) are available, but they are not labeled as spam. Domain adaptation methods can benefit from such unlabeled data, by comparing its distribution (patterns) with the labeled source domain data. Semi-supervised: Most data that is available from the target domain is unlabelled, but some labeled data is also available. In the above-mentioned case of spam filter design, this corresponds to the case that the target user has labeled some emails as being spam or not. Supervised: All data that is available from the target domain is labeled. In this case, domain adaptation reduces to refinement of the source domain predictor. In the above-mentioned example classification of hair-color from images, this could correspond to the refinement of a network already trained on a large dataset of labeled images from Italy, using newly available labeled images from Norway. == Formalization == Let X {\displaystyle X} be the input space (or description space) and let Y {\displaystyle Y} be the output space (or label space). The objective of a machine learning algorithm is to learn a mathematical model (a hypothesis) h : X → Y {\displaystyle h:X\to Y} able to attach a label from Y {\displaystyle Y} to an example from X {\displaystyle X} . This model is learned from a learning sample S = { ( x i , y i ) ∈ ( X × Y ) } i = 1 m {\displaystyle S=\{(x_{i},y_{i})\in (X\times Y)\}_{i=1}^{m}} . Usually in supervised learning (without domain adaptation), we suppose that the examples ( x i , y i ) ∈ S {\displaystyle (x_{i},y_{i})\in S} are drawn i.i.d. from a distribution D S {\displaystyle D_{S}} of support X × Y {\displaystyle X\times Y} (unknown and fixed). The objective is then to learn h {\displaystyle h} (from S {\displaystyle S} ) such that it commits the least error possible for labelling new examples coming from the distribution D S {\displaystyle D_{S}} . The main difference between supervised learning and domain adaptation is that in the latter situation we study two different (but related) distributions D S {\displaystyle D_{S}} and D T {\displaystyle D_{T}} on X × Y {\displaystyle X\times Y} . The domain adaptation task then consists of the transfer of knowledge from the source domain D S {\displaystyle D_{S}} to the target one D T {\displaystyle D_{T}} . The goal is then to learn h {\displaystyle h} (from labeled or unlabelled samples coming from the two domains) such that it commits as little error as possible on the target domain D T {\displaystyle D_{T}} . The major issue is the following: if a model is learned from a source domain, what is its capacity to correctly label data coming from the target domain? == Four algorithmic principles == === Reweighting algorithms === The objective is to reweight the source labeled sample such that it "looks like" the target sample (in terms of the error measure considered). === Iterative algorithms === A method for adapting consists in iteratively "auto-labeling" the target examples. The principle is simple: a model h {\displaystyle h} is learned from the labeled examples; h {\displaystyle h} automatically labels some target examples; a new model is learned from the new labeled examples. Note that there exist other iterative approaches, but they usually need target labeled examples. === Search of a common representation space === The goal is to find or construct a common representation space for the two domains. The objective is to obtain a space in which the domains are close to each other while keeping good performances on the source labeling task. This can be achieved through the use of Adversarial machine learning techniques where feature representations from samples in different domains are encouraged to be indistinguishable. === Hierarchical Bayesian Model === The goal is to construct a Bayesian hierarchical model p ( n ) {\displaystyle p(n)} , which is essentially a factorization model for counts n {\displaystyle n} , to derive domain-dependent latent representations allowing both domain-specific and globally shared latent factors. == Software packages == Several compilations of domain adaptation and transfer learning algorithms have been implemented over the past decades: SKADA (Python) ADAPT (Python) TLlib (Python) Domain-Adaptation-Toolbox (MATLAB)

  Read more →