Orchestrating ai modules

Abstract

A computer-implemented method for controlling a data processing module, the method comprising: receiving a high-level prompt, at a high-level interface of the module; performing a delegation process to transform the high-level prompt into a low-level prompt; outputting the low-level prompt, at a low-level interface of the module; receiving, at the low-level interface, a low-level response based on the low-level- prompt; performing an amalgamation process to determine a high-level response based on the high-level prompt and the low-level response; and outputting the high-level response, at the high-level interface.

Description

[0001] ORCHESTRATING Al MODULES

[0002] TECHNICAL FIELD

[0003] The present disclosure relates to data processing modules that can transform prompts into responses using type-theoretic methods. The disclosure has particular, though not exclusive, relevance to the orchestration of such modules.

[0004] BACKGROUND

[0005] The volume, sophistication and complexity of information processing has increased greatly in recent decades. This trend has been particularly manifest in relation to Artificial Intelligence (Al), where the development of the state of the art has been analogous to the early development of conventional computer programming. In conventional programming, early low-level programming languages gave way to more user-friendly high-level programming languages, enabling programmers to develop increasingly complex and portable software with less explicit dependence on the specific machine architecture. However, while programmers continued to write code from scratch, the scalability of software remained limited. It was only with the development software engineering as a discipline that the scalability problem was effectively addressed: software engineers would draw on pre-existing libraries of code modules which they could combine in novel ways to solve novel programming problems. In this way, software engineering can be seen as a high-level architectural endeavour.

[0006] Scalability remains a serious limitation in the development of Al software. While Al systems such as those based on Large Language Models (LLMs) and other machine learning models provide valuable functionality when applied to specific tasks, there continues to be a lack of flexibility and adaptability in such systems, in addition to limitations in scalability with respect to the number of such systems that can effectively collaborate. It is a purpose of the present disclosure to provide a step-change improvement in relation to Al systems, analogous to the development of software engineering from elementary computer programming.

[0007] SUMMARY

[0008] 1

[0009] According to an aspect of the invention, there is provided a computer-implemented method for forming a data processing module, the method comprising configuring the module to: receive a high-level prompt, at a high-level interface of the module;perform a delegation process to transform the high-level prompt into a low-level prompt;

[0010] output the low-level prompt, at a low-level interface of the module;

[0011] receive, at the low-level interface, a low-level response based on the low-level- prompt; perform an amalgamation process to determine a high-level response based on the high-level prompt and the low-level response; and

[0012] output the high-level response, at the high-level interface;

[0013] wherein:

[0014] the high-level interface is configured to:

[0015] receive prompts having a high-level-prompt-type; and

[0016] output responses having a high-level-response-type; and

[0017] the low-level interface is configured to:

[0018] output prompts having a low-level-prompt-type; and receive responses having a low-level-response-type.

[0019] Advantageously, the method provides for a highly structured means of converting high-level-prompts into high-level responses, using principles of mathematical type theory, that can be readily combined with other modules to provide sophisticated solutions to complex problems. In this way, the disclosure provides scalable methods for orchestrating Al modules or agents, at an architectural level, to provide adaptable and co-operative Al systems that are not constrained in the manner of other contemporary Al designs.

[0020] Optionally, the method may comprise forming a composite-module by compositing the module with at least one further-module according to a set of binary composition operations based on type-theoretic matching of respective prompt and response types of the module and the at least one further-module.

[0021] Advantageously, type-theoretic matching can ensure proper compatibility of different cooperating modules such that errors and inconsistencies may be avoided.

[0022] Optionally, forming the composite-module comprises forming a hierarchical-composite-module configured to:

[0023] receive, at a further-high-level interface of the further-module, a further-high-level-prompt based on the low-level prompt and transforming the further-high-level prompt into a further-low-level prompt;

[0024] receive at the further-low-level interface a further-low-level response, based on the further-low-level prompt;determine a further-high-level response based on the further-high-level prompt and the further-low-level response and provide the further-high-level response to the further high-level interface; and

[0025] receiving the low-level response based on the further-high-level response.

[0026] Advantageously, hierarchical composition can enable the information contained in higher level prompts to be processed at a lower level in a way that can be more efficient. It will be appreciated that a hierarchical chain of modules may be formed using hierarchical composition, such that different inputs, in the form of the original high-level-type prompts, may be processed by different lower-level modules, according to whichever lower-level module possesses the most appropriate information processing capabilities.

[0027] Optionally, forming the composite-module comprises forming a parallel-composite-module configured to:

[0028] receive the high-level prompt and a further-high-level prompt at a composite-high-level interface comprising the high-level interface;

[0029] output the low-level prompt and a further-low-level prompt at a composite-low-level interface comprising the low-level interface;

[0030] receive the low-level response and a further-low-level response at the composite-low-level interface; and

[0031] output the high-level-response and a further-high-level response;

[0032] wherein:

[0033] the further-low-level prompt is based on the further-high-level prompt; the further-low-level response is based on the further-low-level prompt; and the further-high-level response is based on the further high-level prompt and the further-low-level response.

[0034] Advantageously, parallel composition can enable parallel processing of certain prompt-based information, which may increase the overall speed of processing of the information.

[0035] Optionally, each product comprises an ordered pair of the respective prompt types or response types.

[0036] Optionally, forming the composite-module comprises forming a disjunction-composite-module configured to:

[0037] receive the high-level prompt or a further-high-level prompt at a composite-high-level interface comprising the high-level interface;output the low-level prompt or a further-low-level prompt at a composite-low-level interface comprising the low-level-interface, wherein the further-low-level prompt is based on the further-high-level prompt;

[0038] receive the low-level response or a further-low-level response at the composite-low-level interface; wherein the further-low-level response is based on the further-low-level prompt;

[0039] output the high-level response or a further-high-level response at the composite-high-level interface, wherein the further-high-level response is based on the further-high-level prompt and the further-low-level response; and

[0040] wherein:

[0041] if the composite-high-level interface receives the high-level prompt then the composite-high-level interface outputs the high-level response; or

[0042] if the composite-high-level interface receives the further-high-level prompt then the composite-high-level interface outputs the further-high-level response.

[0043] Advantageously, the disjunction composition, which corresponds to a logical OR combination, enables the composite module to respond appropriately to different high-level-type prompts based on their type, such that appropriate downstream processing is used dependent on the prompt type of the original input, thereby providing for improved flexibility of the composite module.

[0044] Optionally, forming the module comprises forming a co-diagonal-module, configured to: receive a further-high-level prompt at the high-level interface;

[0045] output a further-high-level response at the high-level interface;

[0046] wherein:

[0047] if the high-level interface receives the high-level prompt, then the low-level prompt is the high-level prompt, and the high-level interface outputs the high-level response; or

[0048] if the high-level interface receives the further-high-level prompt, then:

[0049] the low-level prompt is the further-high-level prompt;

[0050] the low-level response is based on the further-high-level prompt; and the high-level interface outputs the further-high-level response, wherein the further-high-level response is the low-level response.

[0051] Optionally, forming the composite-module comprises forming a co-pairing module, configured to:

[0052] receive, at a co-diagonal-high-level interface:a first-high-level prompt based on the low-level prompt; or a second-high-level prompt based on the further-low-level prompt; wherein:

[0053] if the co-diagonal-high-level interface receives the first-high-level prompt then:

[0054] a co-diagonal-low-level interface outputs a co-diagonal-low-level prompt, wherein the co-diagonal-low-level prompt is the first-high-level prompt;

[0055] the co-diagonal-low-level interface receives a co-diagonal-low-level response based on the co-diagonal-low-level prompt;

[0056] the co-diagonal-high-level interface outputs a co-diagonal-high-level response, wherein the co-diagonal-high-level response is the co-diagonal-low-level response, and the low-level response is based on the co-diagonal-high-level response; or

[0057] if the co-diagonal-high-level interface receives the second-high-level prompt then:

[0058] the co-diagonal-low-level interface outputs the co-diagonal-low-level prompt, wherein the co-diagonal-low-level prompt is the second-high-level prompt;

[0059] the co-diagonal-low-level interface receives the co-diagonal-low-level response based on the co-diagonal-low-level prompt;

[0060] the co-diagonal-high-level interface outputs the co-diagonal-high-level response, wherein the co-diagonal-high-level response is the co-diagonal-low-level response and the further-low-level response is based on the co-diagonal-high-level response.

[0061] Optionally, if the composite-high-level interface receives the high-level prompt, then the composite-high-level interface outputs the high-level response based on the high-level prompt and the low-level response; or if the composite-high-level interface receives the further-high-level prompt, then the composite-high-level interface outputs the further-high-level response based on the further-high-level prompt and the further-low-level response.

[0062] Advantageously, the co-pairing composition provides an efficient means to respond to different types of high-level input prompt where the low-level-type prompt has the same typestructure irrespective of the different high-level-type prompt type-structure.

[0063] Optionally, forming the composite-module comprises forming a sequencing-composite-module, configured to:

[0064] receive, the high-level prompt and a process, at a composite-high-level interface comprising the high-level interface and a further-high-level interface;

[0065] apply the process to the high-level response to form a further-high-level prompt; receive the further-high-level prompt at the further-high-level-interface;delegate the further-high-level prompt to provide a further-low-level prompt; receive a further-low-level response based on the further-low-level prompt; amalgamate the further-low-level response and the further-high-level prompt to provide a further-high-level response; and

[0066] output a composite-high-level response at the composite-high-level interface, the composite-high-level response comprising the high-level response and the further-high-level response.

[0067] Advantageously, the sequencing composition enables the construction of chains of modules, structured differently than the hierarchical composition, that can provide for more appropriate downstream processing of information representative of the original input high-level prompt type, based on the particular type-theoretic structure of that input.

[0068] Optionally, the high-level-type prompt is based on sensor data received from one or more sensors associated with an apparatus.

[0069] Advantageously, processing sensor data using a type-structured arrangements of modules can provide for superior analysis of the sensor data and thereby an improved measurement of the behaviour, characteristics or performance of the apparatus or any of its constituent component or processes.

[0070] Optionally, the composite module is configured to perform sensor fusion, based on sensor data received from one or more sensors associated with an apparatus.

[0071] Advantageously, the sensor fusion process may be improved by using type-structured configurations of modules, which may thereby enable superior interpretation of the fused data and thus provide a more accurate representation of the apparatus, systems or environments being sensed.

[0072] Optionally, the module may be an Al module.

[0073] Optionally, the high-level-type response is configured to control one or more actuators associated with an apparatus.

[0074] Advantageously, the performance of an apparatus may be improved by modifying the settings of the actuators that control the apparatus based on type-theoretic structured generation and control of composite modules.Optionally, forming the module comprises forming a stateful-module configured to: receive, at the high-level interface, stateful-input-data having a stateful-type; output, at the high-level interface, stateful-output-data having the stateful-type; wherein:

[0075] the delegation process determines the low-level prompt based on the high-level prompt and the stateful-input-data;

[0076] the amalgamation process determines the high-level response, and the stateful-output-data, based on:

[0077] the high-level prompt;

[0078] the stateful-input-data; and

[0079] the low-level-response.

[0080] Advantageously, the stateful module may benefit from prior activity by virtue of the presence of stored memory included in the stateful-input-data.

[0081] Optionally, the composite-module comprises a delegated choice module, and wherein: the further-module is configured to receive a further-high-level prompt and provide a further-low-level prompt based on delegating the further-high-level prompt;

[0082] the composite-module is configured to receive the low-level response; and wherein:

[0083] if the low-level response is based on the high-level prompt, then the module is configured to provide the high-level response as a composite-high-level response for the delegated choice module; or

[0084] if the low-level response is based on the further-high-level prompt, then the further-module is configured to provide the further-high-level response as the composite-high-level response for the delegated choice module.

[0085] Advantageously, the delegated choice module may provide for a more flexible and / or adaptable method of responding to different inputs with different further or high-level prompt types.

[0086] Optionally, the module is an agentic Al module.

[0087] Optionally, the method may further comprise:configuring a plurality of Al modules, including the module, to provide a digital twin of an apparatus, the configuring based on sensor data derived from sensors associated with the apparatus.

[0088] Advantageously, the digital twin may provide for an improved software representation of an apparatus or process, that provides more accurate measured characteristics of the apparatus or process and the can enable more precise and efficient control over the processes occurring within any such apparatus.

[0089] Optionally, the method may further comprise:

[0090] configuring an Al control module to:

[0091] receive a first prompt structured according to a first type-theoretic structure; select the module based on matching the first prompt to the first type-theoretic structure;

[0092] provide the high-level-type prompt, based on the first prompt, to the module; receive the high-level-type response from the module;

[0093] determine a problem condition based on the high-level-type response; modify the module to form an augmented module, wherein the augmented module:

[0094] has an augmented type-theoretic structure, different to the first type-theoretic structure, the augmented type-theoretic structure based on the first type-theoretic structure, and

[0095] is configured to provide an augmented response based on the augmented type-theoretic structure and the first prompt, the augmented response satisfying a threshold condition with respect to the problem condition.

[0096] Advantageously, the method may detect an error, or problem condition, modify the error prone module to provide an augmented module that may provide a valid response of the appropriate type to address the problem to at least a threshold extent.

[0097] Optionally, the composite-module is configured to perform on or more of sensor control, sensor data fusion, decision making, navigation planning and actuator control for an autonomous vehicle.

[0098] Optionally, the composite-module is configured to perform automation of process engineering.Optionally, the composite-module is configured to perform on or more of medical image data analysis, medical image data fusion, biomarker analysis and generation of augmented reality imagery based on medical image data.

[0099] Optionally, the composite-module is configured to perform network data analysis to protect the security and / or integrity of data within the data network from a cyber-attack.

[0100] Optionally, the composite-module is configured to perform network data analysis to detect fraud.

[0101] Optionally, the composite-module is configured to perform logistical and / or supply chain efficiency improvements or optimisations.

[0102] Optionally, the composite-module is configured to conform a user interface to requirements of a user to reduce the number of human-machine interactions required to interact with a user.

[0103] Optionally, the composite-module is configured to determine one or more of: a predictive maintenance model for an apparatus, fault prediction for an apparatus, fault detection for an apparatus, quality inspection for an apparatus, and control instructions fora robotic apparatus.

[0104] According to a further aspect of the invention there is provided a program fora computer which when executed on a processor is configured to perform any method disclosed herein.

[0105] According to a further aspect of the invention there is provided a computer program product, comprising computer code which when executed on a processor is configured to perform any method disclosed herein.

[0106] According to a further aspect of the invention there is provided an apparatus comprising a data processing module, the module comprising:

[0107] a high-level interface configured to receive a high-level-type prompt and output a high-level-type response; and

[0108] a low-level interface configured to output a low-level-type prompt and receive a low-level-type response;

[0109] wherein the module is configured to:

[0110] perform delegation by transforming the high-level-type prompt into the low-level-type prompt; andperform amalgamation by determining the high-level-type response based on the high-level-type prompt and the low-level-type response, wherein the low-level-type response is based on the low-level-type prompt.

[0111] Further features and advantages of the invention will become apparent from the following description of preferred embodiments of the invention, given by way of example only, which is made with reference to the accompanying drawings.

[0112] BRIEF DESCRIPTION OF THE FIGURES

[0113] Figure 1 shows a schematic diagram of a delegation-amalgamation module;

[0114] Figure 2 shows a schematic diagram of a pair of modules combined in a hierarchical composition;

[0115] Figure 3 shows a schematic diagram of a pair of modules combined in a parallel composition;

[0116] Figure 4 shows a schematic diagram of a pair of modules combined in a disjunction composition;

[0117] Figure 5 shows a schematic diagram of a pair of modules combined in a co-diagonal composition;

[0118] Figure 6 shows a schematic diagram of modules combined in a co-pairing composition; Figure 7 shows a schematic diagram of a pair of modules combined in a sequencing composition;

[0119] Figure 8 shows a schematic diagram of a stateful module;

[0120] Figure 9 shows a schematic diagram of a delegated choice module;

[0121] Figure 10 shows a flow diagram of a method for generating an augmented module; and

[0122] Figure 11 shows a computer program product containing computer code configured to, when executed, perform any method disclosed herein.

[0123] DETAILED DESCRIPTION

[0124] Details of systems and methods according to examples will become apparent from the following description with reference to the figures. In this description, for the purposes of explanation, numerous specific details of certain examples are set forth. Reference in the specification to ‘an example’ or similar language means that a feature, structure, or characteristic described in connection with the example is included in at least that one examplebut not necessarily in other examples. It should be further noted that certain examples are described schematically with certain features omitted and / or necessarily simplified for the ease of explanation and understanding of the concepts underlying the examples

[0125] Throughout the present disclosure reference is made to ‘modules’. It will be appreciated that these modules may be implemented either in hardware or software or a combination of both hardware and software. In examples, modules may comprise Al components, such as LLMs, neural networks, or any other Al model or system. When provided with autonomous Al functionality, such modules may, for example, be referred to as ‘agents’ or ‘autonomous agents’ or ‘Al agents’ or ‘agentic modules’ or ‘agentic components’. Herein, modules may be called data processing modules and may also be called delegation-amalgamation modules. In the present disclosure, ‘Al’ may refer to any machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Al systems may be arranged to perform tasks with varying levels of autonomy. Some Al systems may incorporate machine learning models, which may include LLMs or neural networks, for example. Such models may be trained prior to deployment of the Al system and / or the Al system may be arranged to carry out ongoing learning or adaptiveness after deployment.

[0126] Every module has two interfaces called the high-level interface and the low-level interface. The high-level interface can alternatively be called a first interface and the low-level interface can alternatively be called a second interface.

[0127] Reference is further made to “prompts” and “responses”. In this context, a prompt may be any data item suitable to be provided as input to an interface, e.g. as a question to be answered, or a problem to be solved, or a task to be accomplished. A response is any data item suitable as output of an interface which evidences that a prompt has been appropriately dealt with, e.g. an answer to a question, a solution to a problem, or evidence that the task has been accomplished. Prompts and responses may comprise data of any suitable modality, such as a text string, code snippet, image, video file, audio file, or any combination of such modalities. In accordance with the present disclosure, every interface has a “type” describing those prompts it will expect and, for every prompt, another type describing what a response to that prompt will comprise. Herein, the word “type” refers to structures or systems defined in the branches of mathematics and theoretical computer science known as type theory. The prompt of the high level-interface of a module can be referred to as the module’s high level prompt (or high-level-type prompt) and similarly for the response of the high-level interface ofa module (a high level response or high-level-type response), the prompt of the low-level interface of a module (a low level prompt or low-level-type prompt) and the response of the low-level interface of the module (a low level response or low-level-type response).

[0128] It will be appreciated that a prompt may be defined by, or generated in response to, input from a human user, or from another module, or from a software system, or from any other suitable information processing apparatus. The use of types to describe the nature of what might be a prompt or response is a key distinguishing feature of the present disclosure compared to previous approaches.

[0129] Figure 1 shows an example of a delegation-amalgamation module 100. This module exemplifies the fundamental structure of a module according to the present disclosure. The module has two interfaces via which it can communicate information to other modules or information processing apparatus of any appropriate form. These interfaces are the high-level interface 102 and a low-level interface 104. These interfaces have a type-theoretic character. The types that define the structure of the high-level interface 102 and its prompts and responses will not, generally, be the same as the types that define corresponding structures in relation to the low-level interface 104, although in special cases they may be the same. The high-level interface 102 is structured to ingest an input in the form of a high-level prompt 106, where the type of the prompt matches the type the interface is expecting for prompts. The module 100 can then operate on the inputted high-level- prompt 106 to transform it into a low-level prompt 108. This transformation process can be called delegation. The delegation process may include providing the high-level prompt 106 to an LLM or any other suitable Al model or system which forms part of the module 100. The Al model or system may in turn have functionality to access software tools and / or data sources which may be internal or external to the module 100, for example using Model Context Protocol (MCP) or any other suitable framework. The low-level prompt 108 is required to have as its type one that matches the type of prompts expected by the low-level interface 104.

[0130] The low-level prompt 108 can then be output to any other information processing apparatus to convert it into a suitable response, in the form of a low-level response 110, which is structured to have a type that matches the type specified for responses to the low-level prompt 108 by the low-level interface 104. The other information processing apparatus could be another module of the type discussed herein, or a different kind of Al model or system, or any other suitable component implemented in software and / or hardware, and may be fully or partially autonomous or may operate with human input. The low-level interface 104 can therefore receive the low-level response 110. The module 100 can then process the low-levelresponse 110, together with the original high-level prompt 106, to generate a high-level response 112. This generation of the high-level response 112 can be called amalgamation. The amalgamation process may use the same or different Al model(s) or system(s) as the delegation process. This high-level response 112 has the appropriate type specified by the high-level interface 102 for the high-level prompt 106. In this way, the high-level interface 102 can provide, or emit, the high-level response 112 as an output that is responsive to the original input in the form of the high-level prompt 106.

[0131] It is noticeable that the module 100 has two distinct phases. In the first phase, the module 100 transforms the high-level prompt 106 of the high-level interface 102 into the low-level prompt 108 of the low-level interface 104. In the second phase, the module 100 transforms the low-level response 110 of the low-level interface 104 into a high-level response 112 of the high-level interface 102. The behaviour of the module 100 is thus bidirectional going in the first phase from the high-level interface 102 to the low-level interface 104 and in the second phase from the low-level interface 104 to the high-level interface 102. That is a fundamental difference between the present disclosure and the prior approaches that have a unidirectional character. At a high level, a unidirectional flow is sufficient to model the flow of data, however, the present disclosure enables modelling the flow of control, which needs to be bidirectional.

[0132] An advantage of modules with the above-described structure, of type-theoretically structured high- and low-level interfaces, is that they can be combined to form composite modules. By appropriately matching the type-structure of prompts and / or responses, such combinations can work together in a consistent and interoperable manner, whereas, without such typebased matching, errors and inconsistencies may occur. It will be appreciated that composite modules may be comprised of very large numbers of modules connected in intricate structures. However, the basic building blocks of any such multi-module composite will be combinations, or compositions, of pairs of modules. A pair of modules can include a delegation-amalgamation module (which can be referred to as a first module) and a further-delegation-amalgamation module (which can be referred to as a second module). Since only two modules are involved in the most elementary compositions, such compositions may be referred to as binary compositions.

[0133] Figure 2 shows a pair of modules 200 composited together in a hierarchical composition. The first module 201 is as described above in relation to figure 1. Similar features of the first module 201 have been given the same reference numerals (modulo 100) to those of figure 1 and may not necessarily be described further here. The second module 220 has the same fundamental structure as the module described in figure 1 , with a high-level interface 222 anda low-level interface 224. The high-level interface 222 is the same as low-level interface 204. The overall hierarchical composition 200 has as high level-interface 202 and as low-level interface 224. Given a high-level prompt to 206 to the high-level interface 202 of the hierarchically composed module 200, the first module 201 constructs a low-level prompt 208. The prompt 208 is then supplied as the high-level prompt 226 to the second module 220. The second module 220 can then transform that inputted high-level prompt 226 into the second module’s 220 low-level prompt 228.

[0134] The second module’s 220 low-level prompt 228 can then be provided to any suitable downstream information processing apparatus (not shown) which could include being provided to further modules composited together according to any type-theoretically appropriate scheme. This downstream processing then provides a low-level response 230, of the appropriate type-theoretic structure, to the low-level interface 224 of the second module 220. The second module 220 can then process the low-level response 230 and the high-level prompt 226 to generate the required high-level response 232 provided at the second module’s 220 high-level interface 222.

[0135] The low-level response 210 of the first module 201 is then taken to be the high-level response 232 of the second module 220. The first module 201 can then generate its high-level response 212 as described above in relation to figure 1. In this way, the composite module 200 can use the more specialized modules 201 , 220 to assist in processing the original high-level prompt 206.

[0136] Figure 3 shows a module 300 composited together in a parallel composition from a first module 308 and a second module 312. The composite module 300 is designed to receive two high-level prompts: a first high-level prompt 306 to be processed by the first module 308 and a second high-level prompt 310 to be processed by the second module 312. These high-level prompts 306, 310 can be considered a product, in that they make up an ordered pair, with each prompt of the pair having its own (generally different) type-theoretic structure and information content. The first module 308 generates a first low-level prompt 314 and the second module 312 generates a second low-level prompt 316, which together with the first low-level prompt 314 make up and ordered pair, which is the composite low-level prompt 314, 316. The composite low-level prompt 314, 316 can then be processed downstream in any appropriate manner to generate a composite low-level response 318, 320 made up of an order pair of a first low-level response 318 and a second low-level response 320, which are provided to the low-level interface 304 of the composite module 300. The first low-level response 318 is processed, along with the first high-level- prompt 306, by the first module 308, to generatea first high-level response 322. Similarly, the second low-level response 320 is processed, with the second high-level-prompt 310, by the second module 312, to provide the second high-level response 324. Together, as an ordered pair, the first 322 and second 324 high-level responses are provided at the high-level interface 302 of the composite module 300. At each stage, as prompts and responses are generated and communicated, they retain the correct type-theoretic structure required by the high- 302 or low- 304 level interfaces, as appropriate. It will be appreciated that while only a first 308 and second module 312 are disclosed in this specific example, a parallel composition of any number of modules can be constructed, with a high-level prompt comprising an ordered tuple of any number of elementary hight-level-type prompts. In this way, parallel processing can be achieved without errors or inconsistencies.

[0137] Figure 4 shows a pair of modules 400 composited together in a disjunction composition. This composite module 400 is designed to accept either of two (generally differently type structured) high-level prompts: a first high-level prompt 406 (which can be called a high-level prompt) and a second high-level prompt 408 (which can be called a further-high-level prompt). The first high-level prompt 406, when received, is processed by a first module 410 of the composite module 400 while the second high-level prompt 408, when received, is processed by a second module 412 of the composite module 400. The composite module 400 thereby produces either a first low-level prompt 414 (via the first module 410) or a second low-level prompt 416 (via the second module 412), at the composite modules low-level interface 404. The first-low-level prompt 414 may be called a low-level prompt, while the second-low-level prompt 416 may be called a further-low-level prompt. The low-level prompt 414, and the second-low-level prompt 416 can then be processed downstream in any appropriate manner to generate a low-level response 418, 420. The low-level response 418, or the further-low-level response 420 is then processed by either the first 410 or second 412 module according to whichever module produced the low-level prompt 414, or the further-low-level prompt 416. The combination of the relevant low-level response 418, or further low-level response 420 and the relevant high-level prompt 406, or further high-level-prompt 408 then provides an appropriately type-structured high-level response 422, or further high-level response 424 at the composite module’s composite high-level interface 402. This composition of modules enables the composite module 400 to react appropriately to different structures of incoming high-level prompts. In this example, the composite module 400 has only two sub-modules: the first 410 and second 412 modules. However, it will be appreciated that this structure of disjunction composition can be extended to a disjunction of any number of sub-modules, which would give the composite module the ability to process a wide range of incoming high-level prompts.Figure 5 shows co-diagonal composite module 500. Given a low-level interface 504, the module 500 has as high-level interface 502 one of whose prompts are the disjunction of a high-level prompt 506 and a further-high-level prompt 508 expected by the low-level interface 504. Responses at the high-level interface 502 are the disjunction of the responses of the low-level interface 504. Delegation for the module 500 therefore just takes either a high-level prompt 506 or a further-high-level prompt 508 and returns the low-level prompt 510. This low-level prompt 510 can be processed downstream in any appropriate manner to provide a low-level response 512. The high-level response of module 500 is then either i) returned as a first response disjunct 514 (or high-level response) based on the high-level prompt 506; or ii) returned as the second response disjunct 516 (which may be called a further-high-level response) if the further-high-level prompt 508 was originally input. This structure of composite module 500 is useful in constructing other, more complex, composite modules, such as the co-pairing module described below.

[0138] Figure 6 shows two modules 606, 607 composited together to provide a co-pairing module 600. The co-paring module 600 requires the low-level interfaces 604 of a first module 606 and a second module 607 to be the same, in terms of type-structure. The composite module 600 is then built by i) creating a disjunction composition of the first module 606 and the second module 607, as described above in relation to figure 4; and then ii) hierarchically composing this disjunction composite 606, 607 with the appropriate co-diagonal module 616, as described above in relation to figure 5.

[0139] The co-diagonal module 616 receives, at its co-diagonal high-level interface 622 either a first-high-level prompt 618 based on the low-level prompt 612 provided by the first module 606 or a second-high-level prompt 620 based on the further-low-level prompt 614 provided by the second module 607. If the first high-level prompt 618 is received, then the co-diagonal module 616 provides a co-diagonal-low-level prompt 626, which is a copy of the first high-level prompt 618, at the co-diagonal low-level interface 624. Similarly, if the second high-level prompt 620 is received then the co-diagonal module 616 provides a co-diagonal-low-level prompt 626 which is a copy of the second high-level prompt 620. The co-diagonal low-level interface receives a co-diagonal low-level response 628, based on the co-diagonal low-level prompt. The co-diagonal high-level interface then provides a co-diagonal high-level response 630 based on the co-diagonal low-level response 628 and either the first 618 or second 620 high-level response, according to whichever was received. The low-level response 632 received by the first 606 and second 607 modules that make up the disjunction composite module are based on the co-diagonal high-level response 630 and are processed, as described above inrelation to figure 4 to provide the high-level response 634 at the high-level composite interface 602.

[0140] Figure 7 shows two modules, a first module 701 and a second module 722, composited together in a sequencing composition module 700. The first step is to composite together the two high level interfaces 723, 702 to create a new composite high-level interface740. That is, the composite high-level interface 740 takes as a prompt a pair comprising a high-level prompt P 706 for the first module 701 and a process C 714 forturning a high-level response 712 for the high-level prompt P 706 into a second high-level prompt 726 (which can also be called a further high-level prompt) for the second module 722. The first module outputs a low-level prompt 708 at its low-level interface 704, which is part of a composite low-level interface 742. The low-level interface 704 receives a low-level response 710 based on the low-level prompt 708. The high-level interface provides a high-level response 712 based on the low-level response 710 and the high-level prompt 706. The high-level response 712 is provided as part of the output of the sequencing composition module 700 at its composite high-level interface 740. The high-level response 712 is also subject to the process 714 which generates the further high-level prompt 726. The further high-level prompt 726 is processed by the second module 722 to provide a further high-level response 732. Collectively, the high-level response 712 and the further high-level response 732 are provided to the composite high-level interface 740 as the output of the sequencing composition module 700.

[0141] The low-level interfaces 704 724 are combined to form a new composite low level interface 742 of the sequential composition module 700. Delegation in the sequential composition module 700 takes a high level prompt P 706 and process C 714 and creates a composite low level prompt based on a first component being a first low level prompt 708 and a second component being a second low-level prompt 728 based on transforming a first low level response 710 into the second low-level prompt 728 for the sequential composition module 700. The first component 708 is obtained by applying delegation in the first module 701 to the high level prompt P 706. The second component 728 is generated by the process of first performing amalgamation in the first module 701 , then applying process C 714 to the high-level response 712 to generate the second high-level prompt 726, and then performing delegation in the second module 722 to generate the second low-level-prompt 728.

[0142] Amalgamation in the sequential composition module 700 takes a composite low level response 710, 730 for the sequential composition module 700 in the form of a low level response 710 for the first module 701 and a second low level response 730 for the second module 722 and applies amalgamation for both modules 701 , 722 respectively, to create therequired high level response 712 for the first module 701 and a second high level response 732 for the second module 722. This data constitutes a composite high-level response 712, 732 for the sequential composition module 700. It will be appreciated that this concept can be extended to provide a sequential composition module having any number of submodules.

[0143] Figure 8 shows a stateful module 800, that is, a module with a memory or ‘state’ based on previous activity. Modules, especially those that include Al systems, can be particularly useful when they have access to the products of prior activity, such as context including previous prompts, responses, and / or other data previously processed by the module 800 (or another module), which can be referred to as predetermined states. The stateful module 800 receives at its high-level interface 802, stateful-input-data 808 (that determines the state of the stateful module 800) and a high-level prompt 806. The stateful-input-data 808 has a specific type-theoretic structure that can be called its stateful-type. The stateful module 800 generates a high-level response 814, which has the same type-theoretic structure as the high-level prompt 806. A low-level interface 804 can provide a low-level prompt 810 based on both the high-level prompt 806 and the stateful-input-data 808. The low-level interface 804 receives a low-level response 812 based on the low-level prompt 810. The high-level response 814 is based on the low-level response 812, the stateful-input-data 808 and the high-level prompt 806. The high-level interface 802 also provides stateful-output-data 816 that is representative of the state of the stateful module 800 that may have changed as a result of processing the stateful-input-data 808 together with the high-level prompt 806. The stateful-output-data 816 has the same type-theoretic structure (the stateful-type) as the stateful-input-data 808 and can subsequently be used as an input for any stateful module.

[0144] Figure 9 shows a delegated choice module 900. The structure of this composite module 900 is similar to the parallel composition of figure 3, in that a composite high-level prompt 906 can be received at the high-level interface 902, where this high-level prompt 906 is an ordered pair of high-level prompts structured for a first 908 and a second 910 submodule respectively. As in relation to figure 3, each submodule 908, 910 can process its high-level prompt into respective low-level prompts 912. However, the low-level interface 904 expects a response to one - but not both - of the low-level-prompts 912. Whichever component of the low-level-prompts 912 is answered (or responded to), in the form of a low-level-response 914, 916, is fed into the respective component module 908, 910 to compute a high-level response 918, 920 from that component module and that response is outputted as the high-level response 918, 920 for the delegated choice module 900. This delegation in the choice of which submodule 908, 910 responds can improve efficiency and precision by enabling the autonomous selection of the more appropriate sub-module to perform the processing. It will be appreciatedthat this concept can be extended to a delegated choice module having any number of submodules.

[0145] High-level prompts can be generated in many different ways. The following discussion provides some specific examples but should not be taken as limiting as it will be appreciated that many other examples are also possible.

[0146] High-level-type prompts can be generated from data produced by sensors. The data can include sensor data that is representative of the sensor’s physical environment, such as image data produced by a camera or lidar system or could be sensor data representative of the internal state of some apparatus that the sensor is designed to monitor. Alternatively, or additionally, the data can include meta-data that defines settings being used to control the sensor, such as frequency ranges or details of filters used to limit frequency ranges. Modules of the present disclosure can be used to control sensors, such as by processing high-level-prompts that include both sensor data and meta-data to provide high-level responses that reconfigure the sensor’s settings (as encoded in meta-data) to improve the sensor’s functionality, such as in response to changing environmental conditions.

[0147] In other examples, an apparatus may comprise a large number of sensors, some of which may employ similar modalities, such as a number of cameras pointing in different directions, while others may employ different modalities, such as cameras, lidar systems, radar systems, accelerometers, pressure sensors, temperature sensors, or others. This variety of sensors may be present in the sensor suite of an autonomous vehicle, for example. In such cases, it may be necessary to perform sensor fusion, to combine the diverse sensor data to provide for a more accurate representation of the vehicle and its environment than would be possible using individual sensors in isolation from one another, or by using a single sensor modality. Composite modules of the present disclosure can advantageously be employed to perform this function. Different combinations of modules may perform better under different environmental conditions, so the ability to reconfigure the composition of modules 'on-the-fly’ can be especially valuable. For example, limiting the frequency range of one sensor may reduce its performance and usefulness when considered in isolation, but where environmental conditions, such as the presence of fog or smoke, hinder the performance of another sensor, suitably filtered sensor input may optimally complement the performance of the other sensor. Such changes may be counterintuitive, but, with an appropriate composition of modules according to the present disclosure, may become possible in real time. This may be particularly valuable for any apparatus that is exposed to environmental conditions that vary rapidly and in an unpredictable way.The output from composite modules of the present disclosure, the high-level responses, can be used in many different ways. Where the high-level prompts are based on data from sensors, the high-level responses can readily be used to control not only those sensors, but other actuators associated with the apparatus that monitored by those sensor systems. For example, where the apparatus is an autonomous vehicle, it will have a wide variety of actuators designed to control its motion, including steering, braking, accelerating, modifying suspension characteristics and so on. Controlling these appropriately can improve the safety of the vehicle, especially where conditions vary rapidly, where different modules may optimally be included or removed from a composite module to provide for improved analysis of the conditions and control of the vehicle.

[0148] In some examples, modules of the present disclosure may be agentic Al modules. Compositing such modules together, based on the type-theoretical approaches disclosed herein, can enable the orchestration of large assemblies of agents in a scalable manner. This approach allows complex compositions to be formed in a way that maintains error-free and consistent interactions between the agents because of type-theoretical matching between the agents that interact with each other directly.

[0149] Where a number of agentic Al modules are composited together, with inputs in the form of high-level prompts, it is possible to build a digital twin of the apparatus and, potentially, its environment. A digital twin is a software-based representation of the system of interest. Sensor data relevant to the system can be fused and analysed to provide for a more accurate representation of the internal state of the system than would be possible when considering the raw sensor data in isolation. Correspondingly, high-level responses can be formed to provide control signalling to the system concerned to enable superior control and performance of the system than would otherwise be possible

[0150] An example of a digital twin for which composite Al modules of the present disclosure can be particularly relevant is a digital twin of an industrial manufacturing or processing apparatus. For example, in the petrochemicals industry, it is well known that the input, crude oil, can be quite variable in its composition, even for crude of a given grade. The crude may be lighter or heavier (having a greater proportion of shorter- or longer-chain hydrocarbons, respectively) or may be sweeter or more sour (having less sulphur or more sulphur, respectively). Since oil refining can involve a very large number of steps, having a scalable composite agentic Al module-based system can provide for superior monitoring and control of the industrial process concerned.Another example of a digital twin is a system designed to monitor and control an industrial plant processing agricultural produce, such as grain. Again, the inputs can be quite variable, with differing levels of moisture content, protein content and enzymatic activity. Since the number of processing steps can be large, with unpredictable variations in the inputs, a scalable composite Al module-based system can provide for improved performance.

[0151] Composite Al module-based systems can be used to address variations that naturally arise in business beyond those caused by variable inputs, as discussed above. For example, changes in business policy may necessitate changes at an operational level, which may be simulated or modelled using a digital twin, to optimize performance within the constraints imposed by new policies.

[0152] An example of changes in policy that can arise for any business operating industrial plant or machinery are changes in environmental policy. These may be caused by changes in official regulations or on a purely internal basis. In either case, reductions in pollutants released in effluent or other emissions can require expenditure of additional energy, expenditure that can itself increase emissions. This can lead to a highly complex adaptive optimization problem, which must also account forthe need to maintain an efficient output of the desired end product. Scalable agentic Al based solutions can be constructed using the type-theoretic compositional methods disclosed herein, where established approaches adopt a less efficient and less effective approach, since the established approaches lack the scalability that comes from compositing individual Al modules using type-based matching. It will be appreciated that other policy changes can similarly affect physical apparatus and systems at an operational level and that the present disclosure is not limited to changes based on environmental policy alone.

[0153] The present application is intended to encompass a broad range of applications, as illustrated in the examples above, wherein composite Al module-based systems may include broader applications to provide enhanced performance, adaptability, and scalability for complex organisational and adaptability challenges. The following, provided by way of non-limiting examples, are domains and scenarios that fall within the scope of the inventive concepts described herein.

[0154] Autonomous Driving. Composite Al module-based systems may integrate multiple artificial intelligence subsystems — including, but not limited to, perception engines (for processing visual, lidar, radar, and other environmental sensor data, together with data from other autonomous vehicles, typically communicated via a wireless network such as a 5G network,that provide data about the position, velocity and intended changes to the vehicles disposition before they are implemented), decision-making modules (for path planning and hazard avoidance), and control modules (for actuating steering, braking, acceleration, and suspension systems). By orchestrating the interactions of such subsystems, vehicles can safely and efficiently navigate dynamic environments, continuously optimising responses to changing road and weather conditions, traffic density, and unforeseen obstacles such as road closures or traffic gridlock where route modifications may be desirable.

[0155] Enterprise Workflow Automation. The present disclosure contemplates the use of composite Al modules to coordinate a plurality of Al-driven components for the automation of enterprise processes. Such processes may include, without limitation, document classification and processing, workflow routing, automated approval systems, compliance verification, and intelligent exception handling, thereby streamlining enterprise operations, increasing throughput, and reducing manual intervention.

[0156] Healthcare Diagnostics. Composite Al module-based systems may be employed to combine the outputs of diverse machine learning models — including deep neural networks for medical image analysis, natural language processing modules for electronic health records, and predictive analytics engines for patient risk stratification. Other modules may provide analysis of biomarkers including those based on genetic testing of a patient and time series analysis where variations overtime of some biomarkers may provide more valuable insights than can be obtained from a single ‘snap-shot’ test result. The aggregation and orchestration of these modules enables more accurate, timely, and reproducible medical diagnoses, clinical decision support, and personalised care recommendations. The orchestration of such modules can also be advantageously deployed in real-time, such as where a surgical procedure is being performed, to provide a surgeon with a real-time augmented reality view of a patient, based on fusing multiple inputs from diverse medical imagining sources. Furthermore, where a surgical procedure encounters a sudden and unexpected, potentially life-threatening, complication, the ability to augment Al modules, to enable a diagnosis and appropriate response to such dynamic changes may be extremely useful.

[0157] Financial Services. Within the financial sector, composite Al modules can be leveraged to coordinate fraud detection engines, risk assessment models, regulatory compliance systems, and customer service automation tools. Such orchestration permits the real-time identification and mitigation of fraudulent activities, dynamic adjustment of risk models in response to market fluctuations, and efficient, responsive handling of client requests and interactions. The detection of fraud may be based on the use of a diverse array of Al modules designed todetect and respond to a correspondingly diverse array of cyber-attacks. Such attacks can be intended to corrupt the security and / or integrity of data within a computer network. It will be appreciated that different compositions of Al modules may be needed to detect and / or respond to different forms of cyber attack to safeguard the security and integrity of data. This application applies not only to networks and systems concerned with financial data, but more broadly to computer networks and systems that use data for any purpose that could be disrupted by cyber-attacks, including online retail, national or global telecommunications, and the control of critical infrastructure such as electrical power grids or water treatment facilities.

[0158] Logistics and Supply Chain Optimisation. Composite Al module-based systems may orchestrate advanced route planning modules, inventory management algorithms, demand forecasting engines, and real-time tracking systems. This enables dynamic optimisation of shipment routes, inventory levels, and scheduling, taking into account variables such as weather, traffic, supplier disruptions, and demand variability — all of which contribute to enhanced delivery performance, reduced costs, and improved customer satisfaction. It will be appreciated that, in this way, composite Al modules, which may be augmented in response to dynamically changing conditions, can reduce energy consumption and thereby reduce pollution and improve environmental performance relative to conventional control systems, while still achieving the required logistical outcomes. It will further be appreciated that, in this context, optimisation can mean providing an improvement to performance that need not be maximal in any absolute sense.

[0159] Customer Service Automation. The integration of chatbots, recommendation engines, sentiment analysis modules, and context-aware dialogue systems can be accomplished via composite Al modules as described herein. This approach enables organisations to deliver seamless, effective, and highly personalised user interactions across multiple platforms, improving customer engagement, satisfaction, and retention. In some examples, the customer interaction may be improved by using augmented Al modules to modify a user interface, to conform to the requirements of a specific user, such as users with a particular impairment or other need, so that fewer human-machine interactions are required to achieve the specific user’s desired outcome.

[0160] Smart Manufacturing. In industrial production environments, composite Al module-based systems can synchronise predictive maintenance models for early fault detection and prevention. For example, such systems may be used to monitor the performance of aircraft engines during flight to determine appropriate maintenance schedules, but can also monitor aircraft performance more generally to determine if an emergency is likely to occur, such asdue to unexpected environmental conditions, where novel compositions of Al modules may be required to perform the necessary analysis in real time. Quality inspection systems (including computer vision for defect detection), and robotic control modules (for adaptive automation) can also be supported or controlled by composite Al modules of the present disclosure. These orchestrated systems work together to optimise production processes, minimise downtime, ensure product quality, and adapt to changing inputs or operational conditions and / or constraints.

[0161] It will be appreciated that the above examples are illustrative and not exhaustive. The methods and systems disclosed herein may be extended or adapted to other domains or operational scenarios, all of which fall within the scope of the present inventive concepts. The compositional approach described ensures the scalability, reliability, and adaptability necessary for addressing a wide variety of organisational and industrial challenges.

[0162] Figure 10 shows a flow chart 1000 of the steps of a method for constructing an augmented Al module designed to address, or overcome, a problem detected in a delegation-amalgamation module. This method can be orchestrated by an Al control module according to the following steps.

[0163] At step 1 , the Al control module receives a first prompt with a given type-theoretic structure. This prompt will typically comprise an instruction to perform some function. At step 2, the Al control module selects the delegation-amalgamation module since its high-level interface matches the structure of the first prompt. At step 3, the delegation-amalgamation module is provided with a high-level prompt. It will be appreciated that, while the type-theoretic structure of the high-level prompt matches that of the first prompt, the first prompt may have been modified to produce the high-level prompt, by, for example, removing some of the information content or adding some additional information content. At step 4, the Al control module receives the high-level response from the delegation-amalgamation module. In many instances, the high-level response may be an appropriate and useful response. However, in other cases an error may have occurred. At step 5, the high-level response is analysed and if an error has occurred and is detected then a problem condition is established. In some examples, the high-level response may simply be an error message, while in other cases some more analysis by the Al control module may be necessary. At step 6 the Al control module modifies the delegation-amalgamation module to generate an augmented module. This modification may involve compositing the delegation-amalgamation module with one or more other modules according to any suitable combination of binary compositions, to overcome or address the problem condition.The augmented module will therefore have a different type-theoretic structure than the delegation-amalgamation module, typically because the augmented module will include one or more other modules with their own type-theoretic structures. The augmented module can then provide an augmented high-level response, where the nature of the augmentation is selected such that the problem condition is addressed constructively. That is, the problem may or may not be solved completely, but at least a threshold level of improvement should be achieved by the augmentation process.

[0164] In one example, a delegation-amalgamation module may ‘crash’ when presented with certain information content in the high-level prompt or may produce an obvious error. Where the Al control module generates an augmented module, the additional modules composited with the augmented module may be able to respond appropriately to the otherwise problematic information content. In this way, it may be possible to avoid having to re-program the delegation-amalgamation module and yet still overcome the problem. The alternative, of reprogramming the delegation-amalgamation module would risk introducing other, unexpected errors or bugs into the code, whereas augmenting the module enables the augmented module to maintain the sound functionality of the original delegation-amalgamation module, without introducing such risks, while still addressing the problem.

[0165] The methods described herein may be performed using any suitable computing apparatus. An exemplary computing system may include one or more processors, memory, a power supply, and one or more input and output devices. The computing system may be a single device or may include multiple devices connected over a network. The one or more processors may include, for example, one or more of each of a central processing unit (CPU), a graphics processing unit (GPU), and / or a neural processing unit (NPU). Any of these processors may have multiple cores, and various steps of the methods described herein with reference to Figs. 1-10 may be parallelized between cores and / or between processors to the extent possible. The memory (which in the present disclosure may refer to working memory and / or storage) may store program code for implementing any of the functional components or modules described herein. The program code may be written in any suitable programming language and may make use of any software development framework such as PyTorch and / or Tensorflow. Certain subroutines, and / or the modules themselves, may further make use of lower layer task-specific and / or hardware-specific frameworks, such as CUDA by Nvidia (RTM) or Triton by OpenAI (RTM) for model training and / or inference. The modules may further be connected to one another and / or to software tools and / or data sources (with read access or write access) using MCP or any other suitable framework. The power supply mayinclude a mains supply and / or a battery. The input and output devices may enable a user to interact with a user interface, for example to provide prompts or to inspect or modify the operation or composition of modules as described herein with reference to Figs. 1-10.

[0166] At least some aspects of the examples described herein with reference to Figs. 1-10 comprise computer processes or methods performed in one or more processing systems and / or processors. However, in some examples, the disclosure also extends to computer programs, particularly computer programs on or in an apparatus, adapted for putting the disclosure into practice. The program may be in the form of non-transitory source code, object code, a code intermediate source and object code such as in partially compiled form, or in any other non-transitory form suitable for use in the implementation of processes according to the disclosure. The apparatus may be any entity or device capable of carrying the program. For example, the apparatus may comprise a storage medium, such as a solid-state drive (SSD) or other semiconductor-based RAM; a ROM, for example, a CD ROM or a semiconductor ROM; a magnetic recording medium, for example a hard disk; optical memory devices in general; etc. As an example, Figure 11 shows a computer program product 1100. This product 1100 may take the form of any type of non-volatile computer accessible memory or non-transitory computer-readable medium, such as a memory stick as shown here. The product 1100 stores computer code that, when executed on suitable information processing apparatus, can perform any method disclosed herein.

[0167] It is to be understood that any feature described in relation to any one embodiment may be used alone, or in combination with other features described, and may also be used in combination with one or more features of any other of the embodiments, or any combination of any other of the embodiments. Furthermore, equivalents and modifications not described above may also be employed without departing from the scope of the invention, which is defined in the accompanying claims.

Claims

CLAIMS1 . A computer-implemented method for forming a data processing module, the method comprising configuring the module to:receive a high-level prompt, at a high-level interface of the module;perform a delegation process to transform the high-level prompt into a low-level prompt;output the low-level prompt, at a low-level interface of the module;receive, at the low-level interface, a low-level response based on the low-level- prompt; perform an amalgamation process to determine a high-level response based on the high-level prompt and the low-level response; andoutput the high-level response, at the high-level interface;wherein:the high-level interface is configured to:receive prompts having a high-level-prompt-type; andoutput responses having a high-level-response-type; andthe low-level interface is configured to:output prompts having a low-level-prompt-type; and receive responses having a low-level-response-type.

2. The method of claim 1 , further comprising forming a composite-module by compositing the module with at least one further-module according to a set of binary composition operations based on type-theoretic matching of respective prompt and response types of the module and the at least one further-module.

3. The method of claim 2, wherein forming the composite-module comprises forming a hierarchical-composite-module configured to:receive, at a further-high-level interface of the further-module, a further-high-level-prompt based on the low-level prompt and transforming the further-high-level prompt into a further-low-level prompt;receive at the further-low-level interface a further-low-level response, based on the further-low-level prompt;determine a further-high-level response based on the further-high-level prompt and the further-low-level response and provide the further-high-level response to the further high-level interface; andreceiving the low-level response based on the further-high-level response.

4. The method of claim 2, wherein forming the composite-module comprises forming a parallel-composite-module configured to:receive the high-level prompt and a further-high-level prompt at a composite-high-level interface comprising the high-level interface;output the low-level prompt and a further-low-level prompt at a composite-low-level interface comprising the low-level interface;receive the low-level response and a further-low-level response at the composite-low-level interface; andoutput the high-level-response and a further-high-level response;wherein:the further-low-level prompt is based on the further-high-level prompt; the further-low-level response is based on the further-low-level prompt; and the further-high-level response is based on the further high-level prompt and the further-low-level response.

5. The method of claim 2, wherein forming the composite-module comprises forming a disjunction-composite-module configured to:receive the high-level prompt or a further-high-level prompt at a composite-high-level interface comprising the high-level interface;output the low-level prompt or a further-low-level prompt at a composite-low-level interface comprising the low-level-interface, wherein the further-low-level prompt is based on the further-high-level prompt;receive the low-level response or a further-low-level response at the composite-low-level interface; wherein the further-low-level response is based on the further-low-level prompt;output the high-level response or a further-high-level response at the composite-high-level interface, wherein the further-high-level response is based on the further-high-level prompt and the further-low-level response; andwherein:if the composite-high-level interface receives the high-level prompt then the composite-high-level interface outputs the high-level response; orif the composite-high-level interface receives the further-high-level prompt then the composite-high-level interface outputs the further-high-level response.

6. The method of claim 1 , wherein forming the module comprises forming a co-diagonal-module, configured to:receive a further-high-level prompt at the high-level interface;output a further-high-level response at the high-level interface;wherein:if the high-level interface receives the high-level prompt, then the low-level prompt is the high-level prompt, and the high-level interface outputs the high-level response; orif the high-level interface receives the further-high-level prompt, then:the low-level prompt is the further-high-level prompt;the low-level response is based on the further-high-level prompt; and the high-level interface outputs the further-high-level response, wherein the further-high-level response is the low-level response.

7. The method of claim 5, wherein forming the composite-module comprises forming a co-pairing module, configured to:receive, at a co-diagonal-high-level interface:a first-high-level prompt based on the low-level prompt; ora second-high-level prompt based on the further-low-level prompt; wherein:if the co-diagonal-high-level interface receives the first-high-level prompt then:a co-diagonal-low-level interface outputs a co-diagonal-low-level prompt, wherein the co-diagonal-low-level prompt is the first-high-level prompt;the co-diagonal-low-level interface receives a co-diagonal-low-level response based on the co-diagonal-low-level prompt;the co-diagonal-high-level interface outputs a co-diagonal-high-level response, wherein the co-diagonal-high-level response is the co-diagonal-low-level response, and the low-level response is based on the co-diagonal-high-level response; orif the co-diagonal-high-level interface receives the second-high-level prompt then:the co-diagonal-low-level interface outputs the co-diagonal-low-level prompt, wherein the co-diagonal-low-level prompt is the second-high-level prompt;the co-diagonal-low-level interface receives the co-diagonal-low-level response based on the co-diagonal-low-level prompt;the co-diagonal-high-level interface outputs the co-diagonal-high-level response, wherein the co-diagonal-high-level response is the co-diagonal-low-level response and the further-low-level response is based on the co-diagonal-high-level response.

8. The method of claim 7, wherein:if the composite-high-level interface receives the high-level prompt, then the composite-high-level interface outputs the high-level response based on the high-level prompt and the low-level response; orif the composite-high-level interface receives the further-high-level prompt, then the composite-high-level interface outputs the further-high-level response based on the further-high-level prompt and the further-low-level response.

9. The method of claim 2, wherein forming the composite-module comprises forming a sequencing-composite-module, configured to:receive, the high-level prompt and a process, at a composite-high-level interface comprising the high-level interface and a further-high-level interface;apply the process to the high-level response to form a further-high-level prompt; receive the further-high-level prompt at the further-high-level-interface;delegate the further-high-level prompt to provide a further-low-level prompt; receive a further-low-level response based on the further-low-level prompt; amalgamate the further-low-level response and the further-high-level prompt to provide a further-high-level response; andoutput a composite-high-level response at the composite-high-level interface, the composite-high-level response comprising the high-level response and the further-high-level response.

10. The method of any preceding claim, wherein forming the module comprises forming a stateful-module configured to:receive, at the high-level interface, stateful-input-data having a stateful-type; output, at the high-level interface, stateful-output-data having the stateful-type; wherein:the delegation process determines the low-level prompt based on the high-level prompt and the stateful-input-data;the amalgamation process determines the high-level response, and the stateful-output-data, based on:the high-level prompt;the stateful-input-data; andthe low-level-response.

11. The method of claim 2, wherein the composite-module comprises a delegated choice module, and wherein:the further-module is configured to receive a further-high-level prompt and provide a further-low-level prompt based on delegating the further-high-level prompt;the composite-module is configured to receive the low-level response; and wherein:if the low-level response is based on the high-level prompt, then the module is configured to provide the high-level response as a composite-high-level response for the delegated choice module; orif the low-level response is based on the further-high-level prompt, then the further-module is configured to provide the further-high-level response as the composite-high-level response for the delegated choice module.

12. The method of any preceding claim, wherein the high-level prompt is based on sensor data received from one or more sensors associated with an apparatus.

13. The method of any preceding claim, wherein the high-level response is configured to control one or more actuators associated with an apparatus.

14. The method of any of claims 2 to 13, wherein the composite module is configured to perform sensor fusion, based on sensor data received from one or more sensors associated with an apparatus.

15. The method of any preceding claim, wherein the module is an Al module.

16. The method of claim 15, wherein the module is an agentic Al module.

17. The method of any preceding claim, further comprising:configuring a plurality of Al modules, including the module, to provide a digital twin of an apparatus, the configuring based on sensor data derived from sensors associated with the apparatus.

18. The method of claim 1, comprising:configuring an Al control module to:receive a first prompt structured according to a first type-theoretic structure; select the module based on matching the first prompt to the first type-theoretic structure;provide the high-level-type prompt, based on the first prompt, to the module; receive the high-level-type response from the module;determine a problem condition based on the high-level-type response; modify the module to form an augmented module, wherein the augmented module:has an augmented type-theoretic structure, different to the first type-theoretic structure, the augmented type-theoretic structure based on the first type-theoretic structure, andis configured to provide an augmented response based on the augmented type-theoretic structure and the first prompt, the augmented response satisfying a threshold condition with respect to the problem condition.

19. A program for a computer which when executed on a processor is configured to perform the method of any preceding claim.

20. An apparatus comprising a data processing module, the module comprising:a high-level interface configured to receive a high-level prompt and output a high-level response; anda low-level interface configured to output a low-level prompt and receive a low-level response;wherein the module is configured to:perform a delegation process by transforming the high-level prompt into the low-level prompt; andperform an amalgamation process by determining the high-level response based on the high-level prompt and the low-level response.

Images