In logic, statistical inference, and supervised learning, transduction or transductive inference is reasoning from observed, specific (training) cases to specific (test) cases. In contrast, induction is reasoning from observed training cases to general rules, which are then applied to the test cases. The distinction is most interesting in cases where the predictions of the transductive model are not achievable by any inductive model. Note that this is caused by transductive inference on different test sets producing mutually inconsistent predictions. Transduction was introduced in a computer science context by Vladimir Vapnik in the 1990s, motivated by his view that transduction is preferable to induction since, according to him, induction requires solving a more general problem (inferring a function) before solving a more specific problem (computing outputs for new cases): "When solving a problem of interest, do not solve a more general problem as an intermediate step. Try to get the answer that you really need but not a more general one.". An example of learning which is not inductive would be in the case of binary classification, where the inputs tend to cluster in two groups. A large set of test inputs may help in finding the clusters, thus providing useful information about the classification labels. The same predictions would not be obtainable from a model which induces a function based only on the training cases. Some people may call this an example of the closely related semi-supervised learning, since Vapnik's motivation is quite different. The most well-known example of a case-bases learning algorithm is the k-nearest neighbor algorithm, which is related to transductive learning algorithms. Another example of an algorithm in this category is the Transductive Support Vector Machine (TSVM). A third possible motivation of transduction arises through the need to approximate. If exact inference is computationally prohibitive, one may at least try to make sure that the approximations are good at the test inputs. In this case, the test inputs could come from an arbitrary distribution (not necessarily related to the distribution of the training inputs), which wouldn't be allowed in semi-supervised learning. An example of an algorithm falling in this category is the Bayesian Committee Machine (BCM). == Historical context == The mode of inference from particulars to particulars, which Vapnik came to call transduction, was already distinguished from the mode of inference from particulars to generalizations in part III of the Cambridge philosopher and logician W.E. Johnson's 1924 textbook, Logic. In Johnson's work, the former mode was called 'eduction' and the latter was called 'induction'. Bruno de Finetti developed a purely subjective form of Bayesianism in which claims about objective chances could be translated into empirically respectable claims about subjective credences with respect to observables through exchangeability properties. An early statement of this view can be found in his 1937 La Prévision: ses Lois Logiques, ses Sources Subjectives and a mature statement in his 1970 Theory of Probability. Within de Finetti's subjective Bayesian framework, all inductive inference is ultimately inference from particulars to particulars. == Example problem == The following example problem contrasts some of the unique properties of transduction against induction. A collection of points is given, such that some of the points are labeled (A, B, or C), but most of the points are unlabeled (?). The goal is to predict appropriate labels for all of the unlabeled points. The inductive approach to solving this problem is to use the labeled points to train a supervised learning algorithm, and then have it predict labels for all of the unlabeled points. With this problem, however, the supervised learning algorithm will only have five labeled points to use as a basis for building a predictive model. It will certainly struggle to build a model that captures the structure of this data. For example, if a nearest-neighbor algorithm is used, then the points near the middle will be labeled "A" or "C", even though it is apparent that they belong to the same cluster as the point labeled "B", compared to semi-supervised learning. Transduction has the advantage of being able to consider all of the points, not just the labeled points, while performing the labeling task. In this case, transductive algorithms would label the unlabeled points according to the clusters to which they naturally belong. The points in the middle, therefore, would most likely be labeled "B", because they are packed very close to that cluster. An advantage of transduction is that it may be able to make better predictions with fewer labeled points, because it uses the natural breaks found in the unlabeled points. One disadvantage of transduction is that it builds no predictive model. If a previously unknown point is added to the set, the entire transductive algorithm would need to be repeated with all of the points in order to predict a label. This can be computationally expensive if the data is made available incrementally in a stream. Further, this might cause the predictions of some of the old points to change (which may be good or bad, depending on the application). A supervised learning algorithm, on the other hand, can label new points instantly, with very little computational cost. == Transduction algorithms == Transduction algorithms can be broadly divided into two categories: those that seek to assign discrete labels to unlabeled points, and those that seek to regress continuous labels for unlabeled points. Algorithms that seek to predict discrete labels tend to be derived by adding partial supervision to a clustering algorithm. Two classes of algorithms can be used: flat clustering and hierarchical clustering. The latter can be further subdivided into two categories: those that cluster by partitioning, and those that cluster by agglomerating. Algorithms that seek to predict continuous labels tend to be derived by adding partial supervision to a manifold learning algorithm. === Partitioning transduction === Partitioning transduction can be thought of as top-down transduction. It is a semi-supervised extension of partition-based clustering. It is typically performed as follows: Consider the set of all points to be one large partition. While any partition P contains two points with conflicting labels: Partition P into smaller partitions. For each partition P: Assign the same label to all of the points in P. Of course, any reasonable partitioning technique could be used with this algorithm. Max flow min cut partitioning schemes are very popular for this purpose. === Agglomerative transduction === Agglomerative transduction can be thought of as bottom-up transduction. It is a semi-supervised extension of agglomerative clustering. It is typically performed as follows: Compute the pair-wise distances, D, between all the points. Sort D in ascending order. Consider each point to be a cluster of size 1. For each pair of points {a,b} in D: If (a is unlabeled) or (b is unlabeled) or (a and b have the same label) Merge the two clusters that contain a and b. Label all points in the merged cluster with the same label. === Continuous Label Transduction === These methods seek to regress continuous labels, often via manifold learning techniques. The idea is to learn a low-dimensional representation of the data and infer values smoothly across the manifold. == Applications and related concepts == Transduction is closely related to: Semi-supervised learning – uses both labeled and unlabeled data but typically induces a model. Case-based reasoning – such as the k-nearest neighbor (k-NN) algorithm, often considered a transductive method. Transductive Support Vector Machines (TSVM) – extend standard SVMs to incorporate unlabeled test data during training. Bayesian Committee Machine (BCM) – an approximation method that makes transductive predictions when exact inference is too costly.
Quantification (machine learning)
In machine learning, quantification (variously called learning to quantify, or supervised prevalence estimation, or class prior estimation) is the task of using supervised learning in order to train models (quantifiers) that estimate the relative frequencies (also known as prevalence values) of the classes of interest in a sample of unlabelled data items. For instance, in a sample of 100,000 unlabelled tweets known to express opinions about a certain political candidate, a quantifier may be used to estimate the percentage of these tweets which belong to class `Positive' (i.e., which manifest a positive stance towards this candidate), and to do the same for classes `Neutral' and `Negative'. Quantification may also be viewed as the task of training predictors that estimate a (discrete) probability distribution, i.e., that generate a predicted distribution that approximates the unknown true distribution of the items across the classes of interest. Quantification is different from classification, since the goal of classification is to predict the class labels of individual data items, while the goal of quantification it to predict the class prevalence values of sets of data items. Quantification is also different from regression, since in regression the training data items have real-valued labels, while in quantification the training data items have class labels. It has been shown in multiple research works that performing quantification by classifying all unlabelled instances and then counting the instances that have been attributed to each class (the 'classify and count' method) usually leads to suboptimal quantification accuracy. This suboptimality may be seen as a direct consequence of 'Vapnik's principle', which states: If you possess a restricted amount of information for solving some problem, try to solve the problem directly and never solve a more general problem as an intermediate step. It is possible that the available information is sufficient for a direct solution but is insufficient for solving a more general intermediate problem. In our case, the problem to be solved directly is quantification, while the more general intermediate problem is classification. As a result of the suboptimality of the 'classify and count' method, quantification has evolved as a task in its own right, different (in goals, methods, techniques, and evaluation measures) from classification. == Quantification tasks == === Quantification tasks according to the set of classes === The main variants of quantification, according to the characteristics of the set of classes used, are: Binary quantification, corresponding to the case in which there are only n = 2 {\displaystyle n=2} classes and each data item belongs to exactly one of them; Single-label multiclass quantification, corresponding to the case in which there are n > 2 {\displaystyle n>2} classes and each data item belongs to exactly one of them; Multi-label multiclass quantification, corresponding to the case in which there are n ≥ 2 {\displaystyle n\geq 2} classes and each data item can belong to zero, one, or several classes at the same time; Ordinal quantification, corresponding to the single-label multiclass case in which a total order is defined on the set of classes. Regression quantification, a task which stands to 'standard' quantification as regression stands to classification. Strictly speaking, this task is not a quantification task as defined above (since the individual items do not have class labels but are labelled by real values), but has enough commonalities with other quantification tasks to be considered one of them. Most known quantification methods address the binary case or the single-label multiclass case, and only few of them address the multi-label, ordinal, and regression cases. Binary-only methods include the Mixture Model (MM) method, the HDy method, SVM(KLD), and SVM(Q). Methods that can deal with both the binary case and the single-label multiclass case include probabilistic classify and count (PCC), adjusted classify and count (ACC), probabilistic adjusted classify and count (PACC), the Saerens-Latinne-Decaestecker EM-based method (SLD), and KDEy. Methods for multi-label quantification include regression-based quantification (RQ) and label powerset-based quantification (LPQ). Methods for the ordinal case include ordinal versions of the above-mentioned ACC, PACC, and SLD methods, and ordinal versions of the above-mentioned HDy method. Methods for the regression case include Regress and splice and Adjusted regress and sum. === Quantification tasks according to the type of data === Several subtasks of quantification may be identified according to the type of data involved. Example such tasks are: Quantification of networked data. This task consists of performing quantification when the datapoints are members of a relation, i.e., are interlinked. As such, this task is a strict relative of collective classification. Quantification over time. This task consists of performing quantification on sets that become available in a temporal sequence, i.e., as a data stream, and finds application in contexts in which class prevalence values must be monitored over time. == Evaluation measures for quantification == Several evaluation measures can be used for evaluating the error of a quantification method. Since quantification consists of generating a predicted probability distribution that estimates a true probability distribution, these evaluation measures are ones that compare two probability distributions. Most evaluation measures for quantification belong to the class of divergences. Evaluation measures for binary quantification, single-label multiclass quantification, and multi-label quantification, are Absolute Error Squared Error Relative Absolute Error Kullback–Leibler divergence Pearson Divergence Evaluation measures for ordinal quantification are Normalized Match Distance (a particular case of the Earth Mover's Distance) Root Normalized Order-Aware Distance == Applications == Quantification is of special interest in fields such as the social sciences, epidemiology, market research, allocating resources, and ecological modelling, since these fields are inherently concerned with aggregate data. However, quantification is also useful as a building block for solving other downstream tasks, such as improving the accuracy of classifiers on out-of-distribution data, measuring classifier bias and ranker bias, and estimating the accuracy of classifiers on out-of-distribution data. == Resources == LQ 2021: the 1st International Workshop on Learning to Quantify LQ 2022: the 2nd International Workshop on Learning to Quantify LQ 2023: the 3rd International Workshop on Learning to Quantify LQ 2024: the 4th International Workshop on Learning to Quantify LQ 2025: the 5th International Workshop on Learning to Quantify LeQua 2022: the 1st Data Challenge on Learning to Quantify LeQua 2024: the 2nd Data Challenge on Learning to Quantify QuaPy: An open-source Python-based software library for quantification QuantificationLib: A Python library for quantification and prevalence estimation
Noise-based logic
Noise-based logic (NBL) is a class of multivalued deterministic logic schemes, developed in the twenty-first century, where the logic values and bits are represented by different realizations of a stochastic process. The concept of noise-based logic and its name was created by Laszlo B. Kish. In its foundation paper it is noted that the idea was inspired by the stochasticity of brain signals and by the unconventional noise-based communication schemes, such as the Kish cypher. == The noise-based logic space and hyperspace == The logic values are represented by multi-dimensional "vectors" (orthogonal functions) and their superposition, where the orthogonal basis vectors are independent noises. By the proper combination (products or set-theoretical products) of basis-noises, which are called noise-bit, a logic hyperspace can be constructed with D(N) = 2N number of dimensions, where N is the number of noise-bits. Thus N noise-bits in a single wire correspond to a system of 2N classical bits that can express 22N different logic values. Independent realizations of a stochastic process of zero mean have zero cross-correlation with each other and with other stochastic processes of zero mean. Thus the basis noise vectors are orthogonal not only to each other but they and all the noise-based logic states (superpositions) are orthogonal also to any background noises in the hardware. Therefore, the noise-based logic concept is robust against background noises, which is a property that can potentially offer a high energy-efficiency. == The types of signals used in noise-based logic == In the paper, where noise-based logic was first introduced, generic stochastic-processes with zero mean were proposed and a system of orthogonal sinusoidal signals were also proposed as a deterministic-signal version of the logic system. The mathematical analysis about statistical errors and signal energy was limited to the cases of Gaussian noises and superpositions as logic signals in the basic logic space and their products and superpositions of their products in the logic hyperspace (see also. In the subsequent brain logic scheme, the logic signals were (similarly to neural signals) unipolar spike sequences generated by a Poisson process, and set-theoretical unifications (superpositions) and intersections (products) of different spike sequences. Later, in the instantaneous noise-based logic schemes and computation works, random telegraph waves (periodic time, bipolar, with fixed absolute value of amplitude) were also utilized as one of the simplest stochastic processes available for NBL. With choosing unit amplitude and symmetric probabilities, the resulting random-telegraph wave has 0.5 probability to be in the +1 or in the −1 state which is held over the whole clock period. == The noise-based logic gates == Noise-based logic gates can be classified according to the method the input identifies the logic value at the input. The first gates analyzed the statistical correlations between the input signal and the reference noises. The advantage of these is the robustness against background noise. The disadvantage is the slow speed and higher hardware complexity. The instantaneous logic gates are fast, they have low complexity but they are not robust against background noises. With either neural spike type signals or with bipolar random-telegraph waves of unity absolute amplitude, and randomness only in the sign of the amplitude offer very simple instantaneous logic gates. Then linear or analog devices unnecessary and the scheme can operate in the digital domain. However, whenever instantaneous logic must be interfaced with classical logic schemes, the interface must use correlator-based logic gates for an error-free signal. == Universality of noise-based logic == All the noise-based logic schemes listed above have been proven universal. The papers typically produce the NOT and the AND gates to prove universality, because having both of them is a satisfactory condition for the universality of a Boolean logic. == Computation by noise-based logic == The string verification work over a slow communication channel shows a powerful computing application where the methods is inherently based on calculating the hash function. The scheme is based on random telegraph waves and it is mentioned in the paper that the authors intuitively conclude that the intelligence of the brain is using similar operations to make a reasonably good decision based on a limited amount of information. The superposition of the first D(N) = 2N integer numbers can be produced with only 2N operations, which the authors call "Achilles ankle operation" in the paper. == Computer chip realization of noise-based logic == Preliminary schemes have already been published to utilize noise-based logic in practical computers. However, it is obvious from these papers that this young field has yet a long way to go before it will be seen in everyday applications.
International Conference on Autonomous Agents and Multiagent Systems
The International Conference on Autonomous Agents and Multiagent Systems or AAMAS is the leading scientific conference for research in the areas of artificial intelligence, autonomous agents, and multiagent systems. It is annually organized by a non-profit organization called the International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS). == History == The International Conference on Autonomous Agents and Multiagent Systems (AAMAS) is a highly respected joint conference that provides a quality forum for discussing research in intelligent computational agents and their interactions. It is a merger of three major international conferences/workshops, namely the International Conference on Autonomous Agents (AGENTS), International Conference on Multi-Agent Systems (ICMAS), and International Workshop on Agent Theories, Architectures, and Languages (ATAL). ICMAS is itself a merger of three formative workshops, each with an attendance of fewer than 50 researchers. At a meeting during IJCAI-93 held in Chambery, France in August 1993, the leaders of the European Workshops on Modelling Autonomous Agents in a Multi-Agent World, the Asian MAAC Workshops, and the North American Distributed Artificial Intelligence Workshops (Victor Lesser, Michael N. Huhns, Les Gasser, Barbara Grosz, Nicholas Jennings, Michael Wooldridge, Gerhard Weiss, Mario Tokoro, and Toru Ishida) began the planning for a combined conference, which resulted in the first ICMAS in San Francisco, CA, USA in 1995, attended by more than 500 researchers. The AAMAS Conference is under the guidance and management of the International Foundation for Autonomous Agents and Multiagent Systems, which is incorporated as a 501(c)(3) non-profit organization in South Carolina, USA. == Current and previous conferences == 2024: Auckland, New Zealand (May 6-10) 2023: London, United Kingdom (May 29-June 1) 2022: Auckland, New Zealand (May 9–13) 2021: London, United Kingdom (May 3-May 7) 2020: Auckland, New Zealand (May 9–13) 2019: Montreal, Canada (May 13–17) 2018: Stockholm, Sweden (July 10–15) 2017: São Paulo, Brazil 2016: Singapore City, Singapore 2015: Istanbul, Turkey 2014: Paris, France 2013: Saint Paul, USA 2012: Valencia, Spain 2011: Taipei, Taiwan 2010: Toronto, Canada 2009: Budapest, Hungary 2008: Estoril, Portugal 2007: Honolulu, USA 2006: Hakodate, Japan 2005: Utrecht, The Netherlands 2004: New York, USA 2003: Melbourne, Australia 2002: Bologna, Italy == Activities == Besides the main program that consists of a main track, an industry and applications track, and a couple of special area tracks, AAMAS also hosts over 20 workshops (e.g., AOSE, COIN, DALT, ProMAS, to mention a few) and many tutorials. There is also a demonstration session and a doctoral symposium. Finally, each year AAMAS features a bunch of awards, most notably the IFAAMAS Influential Paper Award. It publishes proceedings which are available online.
Overwatch
Overwatch (abbreviated as OW) is a multimedia franchise centered on a series of multiplayer first-person shooter (FPS) video games developed by Blizzard Entertainment. Overwatch was released in 2016. Overwatch 2 was released in 2022 and the original game was taken offline upon its release, though Blizzard renamed it back to Overwatch in 2026. Overwatch features hero-based combat between two teams of players fighting over various objectives, along with other traditional gameplay modes. Released in 2016, Overwatch lacked a traditional story mode. Instead, Blizzard employed a transmedia storytelling strategy to disseminate lore regarding the game's characters, releasing comics and other literary media, as well as animated media that includes short films. The game enjoyed both critical and commercial success, and garnered a devoted following. The fan community around the franchise has produced a large amount of content including art, cosplay, fan fiction, anime-influenced music videos, Internet memes, and pornography. Blizzard helped launch and promote an esports scene surrounding the game, including an annual Overwatch World Cup, Overwatch League a minor league, and the Overwatch Champions Series which borrowed elements found in traditional American sports leagues. == Gameplay == Both games in the Overwatch series are team-based hero shooters. Players select a hero character from a large roster (52 as of Season 2), divided among three class types. These are: Tanks, who have higher health and generally meant to help protect their teammates from damage, but are larger and easier to hit; Damage, who act as the team's offensive leads; and Support, who heal, provide buffs for teammates, or de-buff the opposing team. Each role also features sub-roles with extra passives. These sub-roles include 'Initiator', 'Stalwart', and 'Bruiser' for Tank. 'Specialist', 'Flanker', 'Recon', and 'Sharpshooter' for Damage. 'Medic', 'Tactician', and 'Survivor' for Support. Players are generally free to change to different heroes while inside their spawn room during the course of a match in response to the current tactics employed by other players. As of the development of Overwatch 2, a standard game features one tank player, two damage players and two support players, a change from having two of each class in its predecessor. Players choose their class before the match, and can only pick characters within that class for the duration of the game. There are different styles of game modes, however, that allow players to choose characters from any class throughout the game. Each hero has a skill kit that includes a primary attack, active skills that require a cooldown period before they can be used again, passive skills that remain active at all times, and an Ultimate skill that can only be used once they fill their Ultimate meter either by damaging opponents, mitigating damage, healing teammates or by passively generating it over time. An update in 2025 saw each hero receive a total of four unique abilities known as perks. Each hero has two minor and two major perks; minor perks consist of smaller changes to a hero's kit, while major perks are intended to affect the match more significantly. At the beginning of each match, all heroes are set to level 1 for each player. As the match progresses, players can individually level up their respective heroes, minor perks are unlocked at level 2, and major perks are unlocked at the maximum level 3. When perks become available, players may only select one of each type of perk; a selected perk becomes irreversibly attached to the current hero for the remainder of the match. If a player switches to another hero mid-match, the previously selected hero retains their level and perk progress. Game types of Overwatch are split between standard matches, competitive play, custom games, and arcade modes. Standard matches have matchmaking based loosely on the player's skill level as measured by the game. Competitive mode uses more strict matchmaking based on a player's current rank on the competitive ladder, with their rank increasing or decreasing when they win or lose a game, respectively. Arcade modes do not use matchmaking and are generally more experimental modes compared to standard and competitive modes. Custom games are created via the workshop and can be utilised to make game modes that are very different from the base game. The workshop, is the software in Overwatch which creates the game using either presets and settings or rules and conditions made by code. These game modes can be published directly onto Overwatch’s custom browse tab or shared off platform using a 5 digit alphanumeric code. Standard and competitive game modes are randomly selected at the start of each match, and are objective based, requiring teams to control a fixed objective point for a duration of time, or escort a payload to a target zone before match time expires. These modes include: Assault (introduced in Overwatch): Also known as 2 Capture Points (or 2CP), Assault has the attacking team tasked with capturing two target points in sequence on the map, while the defending team must stop them. Assault-style maps were removed from main gameplay rotation after Overwatch 2 released but available in the game's arcade mode. It is still available in the game's custom game modes. Since Season 2, Assault-style maps are available in Arcade Mode daily routines. Escort (introduced in Overwatch): Also known as "Payload" by the community, The attacking team is tasked with escorting a payload to a certain delivery point before time runs out, while the defending team must stop them. The payload vehicle moves along a fixed track when any player on the attacking team is close to it, increasing in speed if multiple attackers are present, the increase capping at 3, but will stop if a defending player is nearby; should no attacker be near the vehicle, it will start to move backwards along the track. The payload will also heal any attacking players by 10 health per second while they are near the payload. Passing specific checkpoints will extend the match time and prevent the payload from moving backwards from that point. Hybrid (Assault/Escort) (introduced in Overwatch): The attacking team has to capture the payload (as if it were a target point from Assault) and escort it to its destination, while the defending team tries to hold them back. Control (introduced in Overwatch): Each team tries to capture and maintain a common control point until their capture percentage reaches 100%. This game mode is played in a best-of-three format. Control maps are laid out in a symmetric fashion so no team has an intrinsic position advantage. Push (introduced in Overwatch 2's launch): Each team attempts to secure control of a large robot that pushes one of two barriers to the opposing team's side of the map, whilst being escorted by at least one team member, stopping when enemy players are nearby, similar to the payload movement system in Escort. The team that pushes the payload fully to the other side, or furthest into the enemy territory before the time runs out, wins the match. Flashpoint (introduced in Overwatch 2 in 2023): Similar to Control, each team attempts to capture and maintain a common control point until their capture percentage reaches 100%. This game mode takes place on significantly larger maps with five separate control points, which take a shorter amount of time to capture as compared to a standard Control map. A central control point is always activated first; after it is secured by one team, the remaining four are activated in a random order. The first team to secure three control points wins. Clash (introduced in Overwatch 2 in 2024): Clash maps feature symmetrical maps with five control points. Teams initially vie for control of the central point, with the winning team progressing to the next control point, towards the opponent's base. Opponents can push back by winning control points and shifting the next point away from their base. If a team captures the point closest to the opponent's base, they win. Otherwise the match plays out until one team wins control five times. Arcade modes may include variations of the above modes with experimental rules, and can also include modes like Deathmatch and Capture the Flag. Other common arcade modes include: Elimination (introduced in Overwatch in 2016): Two teams face off in a series of rounds, attempting to wipe out the other team; once a player is killed they remain out of the game until the next round, though they can be revived by Mercy's 'Resurrect' ability. If no team has won a round by a certain time, then the winners are decided by the team that can first take a neutral control point. Players cannot change heroes until the next round. Some of these can be played in "lockout" mode, in which the heroes selected by the winning team for a round are "locked" and cannot be selected in future rounds. Total Mayhem (i
Stochastic parrot
In machine learning, the term stochastic parrot is a metaphor that frames large language models as systems that statistically mimic text without real understanding. The word "stochastic" – from the ancient Greek "στοχαστικός" (stokhastikos, 'based on guesswork') – is a term from probability theory meaning "randomly determined". The word "parrot" refers to parrots' ability to mimic human speech. The term was introduced in a 2021 paper on AI ethics titled "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜" and authored by Timnit Gebru, Emily M. Bender, Angelina McMillan-Major, and Margaret Mitchell. The paper outlined possible risks associated with large language models (LLMs). In December 2020, it was the subject of a workplace dispute between Gebru (then co-leader of Google's Ethical Artificial Intelligence Team) and Google, which had requested the retraction of the paper. The incident culminated in Gebru's controversial departure from the company. The paper was later presented at the 2021 ACM Conference, and the term "stochastic parrot" has seen widespread use in academic research concerning generative AI and LLMs. The term has been interpreted negatively as an insult towards AI. == Background == Timnit Gebru is an AI ethics researcher, Emily M. Bender is a linguist specializing in computational linguistics, and Margaret Mitchell is a computer scientist specializing in algorithmic bias. Gebru had joined Google in 2018, where she co-led a team on the ethics of artificial intelligence with Mitchell. In late 2020, the paper "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜" was co-written by Gebru and five other researchers, four of whom were Google employees. The paper argues that large language models (LLMs) present significant risks such as environmental and financial costs, inscrutability leading to unknown dangerous biases, and potential for deception as LLMs do not understand the concepts underlying what they learn. The paper states that LLMs are "stitching together sequences of linguistic forms ... observed in its vast training data, according to probabilistic information about how they combine, but without any reference to meaning." Therefore, they are labeled "stochastic parrots". === Dismissal of Gebru by Google === After the paper was submitted for consideration to the 2021 ACM Conference, Google requested that Gebru either retract the paper from the conference or remove the names of Google employees from it. Gebru refused to do so without further discussion, and emailed Google Research vice president Megan Kacholia that if the company could not explain the request for retraction and address other concerns regarding similar projects, she would plan to resign after a transition period, stating that they could "work on a last date". The following day, on December 2, 2020, Gebru received an email saying that Google was "accepting her resignation". Her abrupt firing sparked protests by Google employees and negative publicity for the company. == Usage == The phrase has been used by AI skeptics to signify that LLMs lack understanding of the meaning of their outputs. Sam Altman, CEO of OpenAI, used the term shortly after the release of ChatGPT in December 2022, tweeting "i am a stochastic parrot, and so r u". The term was nominated as the 2023 AI-related Word of the Year by the American Dialect Society. == Debate == Some LLMs, such as ChatGPT, have become capable of interacting with users in convincingly human-like conversations. The development of these new systems has deepened the discussion of the extent to which LLMs understand or are simply "parroting". According to machine learning researchers Lindholm, Wahlström, Lindsten, and Schön, the term "stochastic parrot" highlights two vital limitations of LLMs: LLMs are limited by the data they are trained on and are simply stochastically repeating contents of datasets. Because they are just making up outputs based on training data, LLMs do not understand if they are saying something incorrect or inappropriate. Lindholm et al. noted that, with poor quality datasets and other limitations, a learning machine might produce results that are "dangerously wrong". === Subjective experience === In the mind of a human being, words and language correspond to things one has experienced. For LLMs, according to proponents of the theory, words correspond only to other words and patterns of usage fed into their training data. Proponents of the idea of stochastic parrots thus conclude that statements about LLMs are due to "the human tendency to attribute meaning to text", and claim this occurs despite the LLMs not actually understanding language. === Fine-tuning === Kelsey Piper argued that the claim that LLMs are stochastic parrots or mere "next-token predictors" focuses on pre-training, ignoring that modern LLMs are also fine-tuned to follow instructions and to prefer accurate answers. === Hallucinations and mistakes === The tendency of LLMs to pass off false information as fact is held as support. Called hallucinations or confabulations, LLMs will occasionally synthesize information that matches some pattern. LLMs may fail to distinguish fact and fiction, which leads to the claim that they can't connect words to a comprehension of the world, as humans do. Furthermore, LLMs may fail to decipher complex or ambiguous grammar cases that rely on understanding the meaning of language. For example: The wet newspaper that fell down off the table is my favorite newspaper. But now that my favorite newspaper fired the editor I might not like reading it anymore. Can I replace 'my favorite newspaper' by 'the wet newspaper that fell down off the table' in the second sentence? GPT-4, an LLM released in March 2023, responded yes, not understanding that the meaning of "newspaper" is different in these two contexts; it is first an object and second an institution. === Benchmarks and experiments === One argument against the hypothesis that LLMs are stochastic parrot is their results on benchmarks for reasoning, common sense and language understanding. In 2023, some LLMs have shown good results on many language understanding tests, such as the Super General Language Understanding Evaluation (SuperGLUE). GPT-4 scored in the >90th-percentile on the Uniform Bar Examination and achieved 93% accuracy on the MATH benchmark of high-school Olympiad problems, results that exceed rote pattern-matching expectations. Such tests, and the smoothness of many LLM responses, help as many as 51% of AI professionals believe they can truly understand language with enough data, according to a 2022 survey. === Expert rebuttals === Some AI researchers dispute the notion that LLMs merely "parrot" their training data. Geoffrey Hinton, a pioneering figure in neural networks, counters that the metaphor misunderstands the prerequisite for accurate language prediction. He argues that "to predict the next word accurately, you have to understand the sentence", a view he presented on 60 Minutes in 2023. From this perspective, understanding is not an alternative to statistical prediction, but rather an emergent property required to perform it effectively at scale. Hinton also uses logical puzzles to demonstrate that LLMs actually understand language. A 2024 Scientific American investigation described a closed Berkeley workshop where state-of-the-art models solved novel tier-4 mathematics problems and produced coherent proofs, indicating reasoning abilities beyond memorization. The GPT-4 Technical Report showed human-level results on professional and academic exams (e.g., the Uniform Bar Exam and USMLE), challenging the "parrot" characterization. Anthropic conducted mechanistic interpretability research on Claude, using attribution graphs to identify circuits. The research showed how the LLM processes information via chains of fuzzy logical inference, and indicated an ability to plan ahead. They found that Claude 3.5 Haiku "employs remarkably general abstractions", forms "internally generated plans for its future outputs" and "works backwards from its longer-term goals". They noted that "The mechanisms of the model can apparently only be faithfully described using an overwhelmingly large causal graph." They also found that the model includes "mechanisms that could underlie a simple form of metacognition", in that it "thinks about" the level of its own knowledge before reaching its answer. === Interpretability === Another line of evidence against the 'stochastic parrot' claim comes from mechanistic interpretability, a research field dedicated to reverse-engineering LLMs to understand their internal workings. Rather than only observing the model's input-output behavior, these techniques probe the model's internal activations, which can be used to determine if they contain structured representations of the world. The goal is to investigate whether LLMs are merely manipulating surface statistics or if t
R.U.R.
R.U.R. is a 1920 science fiction play by the Czech writer Karel Čapek. "R.U.R." stands for Rossumovi Univerzální Roboti (Rossum's Universal Robots, a phrase that has been used as a subtitle in English versions). The play had its world premiere on 2 January 1921 in Hradec Králové. It introduced the word "robot" to the English language and to science fiction as a whole. R.U.R. became influential soon after its publication. By 1923, it had been translated into thirty languages. R.U.R. was successful in its time in Europe and North America. Čapek later took a different approach to the same theme in his 1936 novel War with the Newts, in which non-humans become a servant-class in human society. == Characters == Parentheses indicate names which vary according to translation. On the meaning of the names, see Ivan Klíma: Karel Čapek: Life and Work (2002). == Plot == === Synopsis === The play begins in a factory that makes artificial workers from synthetic organic matter. (As living creatures of artificial flesh and blood, that later terminology would call androids, the playwright's 'roboti' differ from later fictional and scientific concepts of inorganic constructs.) Robots may be mistaken for humans but have no original thoughts. Though most are content to work for humans, eventually a rebellion causes the extinction of the human race. === Prologue (Act I in the Selver translation) === Helena, the daughter of the president of a major industrial power, arrives at the island factory of Rossum's Universal Robots. Here, she meets Domin, the General Manager of R.U.R., who relates to her the history of the company. Rossum had come to the island in 1920 to study marine biology. In 1932, Rossum had invented a substance like organic matter, though with a different chemical composition. He argued with his nephew about their motivations for creating artificial life. While the elder wanted to create animals to prove or disprove the existence of God, his nephew only wanted to become rich. Young Rossum finally locked away his uncle in a lab to play with the monstrosities he had created and created thousands of robots. By the time the play takes place (circa the year 2000), robots are cheap and available all over the world. They have become essential for industry. After meeting the heads of R.U.R., Helena reveals that she is a representative of the League of Humanity, an organization that wishes to liberate the robots. The managers of the factory find this absurd. They see robots as appliances. Helena asks that the robots be paid, but according to R.U.R. management, the robots do not "like" anything. Eventually Helena is convinced that the League of Humanity is a waste of money, but still argues robots have a "soul". Later, Domin confesses that he loves Helena and forces her into an engagement. === Act I (Act II in Selver) === Ten years have passed. Helena and her nurse Nana discuss current events, the decline in human births in particular. Helena and Domin reminisce about the day they met and summarize the last ten years of world history, which has been shaped by the new worldwide robot-based economy. Helena meets Dr. Gall's new experiment, Radius. Dr. Gall describes his experimental robotess, also named Helena. Both are more advanced, fully-featured robots. In secret, Helena burns the formula required to create robots. The revolt of the robots reaches Rossum's island as the act ends. === Act II (Act III in Selver) === The characters sense that the very universality of the robots presents a danger. Echoing the story of the Tower of Babel, the characters discuss whether creating national robots who were unable to communicate beyond their languages would have been a good idea. As robot forces lay siege to the factory, Helena reveals she has burned the formula necessary to make new robots. The characters lament the end of humanity and defend their actions, despite the fact that their imminent deaths are a direct result of their choices. Busman is killed while attempting to negotiate a peace with the robots. The robots storm the factory and kill all the humans except for Alquist, the company's Clerk of the Works (Head of Construction). The robots spare him because they recognize that "He works with his hands like a robot. He builds houses. He can work." === Act III (Epilogue in Selver) === Years have passed. Alquist, who still lives, attempts to recreate the formula that Helena destroyed. He is a mechanical engineer, though, with insufficient knowledge of biochemistry, so he has made little progress. The robot government has searched for surviving humans to help Alquist and found none alive. Officials from the robot government beg him to complete the formula, even if it means he will have to kill and dissect other robots for it. Alquist yields. He will kill and dissect robots, thus completing the circle of violence begun in Act Two. Alquist is disgusted. Robot Primus and Helena develop human feelings and fall in love. Playing a hunch, Alquist threatens to dissect Primus and then Helena; each begs him to take him- or herself and spare the other. Alquist now realizes that Primus and Helena are the new Adam and Eve, and gives the charge of the world to them. == Čapek's conception of robots == The robots described in Čapek's play are not robots in the popularly understood sense of an automaton. They are not mechanical devices, but rather artificial biological organisms that may be mistaken for humans. A comic scene at the beginning of the play shows Helena arguing with her future husband, Harry Domin, because she cannot believe his secretary is a robotess: His robots resemble more modern conceptions of man-made life forms, such as the Replicants in Blade Runner, the "hosts" in the Westworld TV series and the humanoid Cylons in the re-imagined Battlestar Galactica, but in Čapek's time there was no conception of modern genetic engineering (DNA's role in heredity was not confirmed until 1952). There are descriptions of kneading-troughs for robot skin, great vats for liver and brains, and a factory for producing bones. Nerve fibers, arteries, and intestines are spun on factory bobbins, while the robots themselves are assembled like automobiles. Čapek's robots are living biological beings, but they are still assembled, as opposed to grown or born. One critic has described Čapek's robots as epitomizing "the traumatic transformation of modern society by the First World War and the Fordist assembly line". === Origin of the word robot === The play introduced the word robot, which displaced older words such as "automaton" or "android" in languages around the world. In an article in Lidové noviny, Karel Čapek named his brother Josef as the true inventor of the word. In Czech, robota means forced labour of the kind that serfs had to perform on their masters' lands and is derived from rab, meaning "slave". The name Rossum is an allusion to the Czech word rozum, meaning "reason", "wisdom", "intellect" or "common sense". It has been suggested that the allusion might be preserved by translating "Rossum" as "Reason" but only the Majer/Porter version translates the word as "Reason". == Production history and translations == The work was published in two differing versions in Prague by Aventinum, first in 1920, followed by a revised version in 1921. After being postponed, it premiered at the city's National Theatre on 25 January 1921, although an amateur group had by then already presented a production. By 1921, Paul Selver translated either the original 1920 edition of R.U.R. or a manuscript copy close to this version into English. He probably translated the play freelance, and sold it to St Martin's Theatre in London. Selver's translation was adapted for the British stage by Nigel Playfair in 1922, but it was not produced straight away. Later that year performance rights for the U.S. and Canada were sold to the New York Theatre Guild, perhaps during Lawrence Langner's visit to Britain. Playfair's version included several changes to Čapek's original play, such as renaming the acts (the prologue became act one, and the heavily abridged final act became the epilogue), omitting around sixty lines (including most of Alquist's final speech), adding several more lines, and removing the robot character Damon (giving his lines to Radius). The omission of some lines may have been censorship from the Lord Chamberlain's Office, or self-censorship in anticipation of this, while some other changes might have been made by Čapek himself if Selver was working from a manuscript copy. An edition of Playfair's adaptation was published by the Oxford University Press in 1923, and Selver went on to write a satiric novel One, Two, Three (1926) based on his experiences getting R.U.R. staged. The American première was produced by the Theatre Guild at the Garrick Theatre in New York City in October 1922, where it ran for 184 performances. In the first performance, Domin was portrayed by Basil Sydney,