**Fast Track** that allows users to quickly navigate through the highlights. Although we recommend going through this documentation in its entirety, readers are encouraged to choose the path that is most suitable for them.
{% hint style="info" %} 
**Fast Track**
Go directly to [The Problem](/introduction/the-problem) if you are already familiar with the basic concepts for Data Growth and Data Processing.
{% endhint %}
# Terms & Definitions
We will be using the following nomenclature throughout this documentation:
| Term | Definition |
|---|---|
| Term | Definition |
| AI | Artificial Intelligence: the ability of machines to perform tasks that are typically associated with human intelligence, such as learning and problem-solving. The traditional goals of AI research include reasoning, knowledge representation, planning, learning, natural language processing, perception, and support for robotics. |
| ML | Machine Learning: an umbrella term for solving problems for which development of algorithms by human programmers would be cost-prohibitive, and instead the problems are solved by helping machines 'discover' their 'own' algorithms, without needing to be explicitly told what to do by any human-developed algorithms. As a scientific endeavour, machine learning grew out of the quest for AI. |
| Dataset | A collection of related, discrete items of related data that may be accessed individually or in combination or managed as a whole entity. |
| ML Algorithm | A mathematical method to find patterns in a set of data. |
| ML Model Training | The process of running an ML algorithm on a dataset (called training data) and optimising the algorithm to find certain patterns or outputs. The resulting function with rules and data structures is called the trained machine learning model. |
| ML Model | A computer program that is used to recognise patterns in data or make predictions. ML models are created from ML algorithms, which are trained using either labelled, unlabelled, or mixed data. |
Data Labelling or | A part of the preprocessing stage when developing an ML model. It requires the identification of raw data (i.e., images, text files, videos), and then the addition of one or more labels to that data to specify its context for the models, allowing the machine learning model to make accurate predictions. |
| Agent | An actor in the data value chain that is responsible for processing data. |
| Human Agent | A natural person that solves data processing (labelling) tasks as a part of the crowdsourcing community. |
| AI Agent | A computer program or system that is designed to solve data processing (labelling) autonomously, based on an ML model that has been previously trained using data labelled by Human Agents. |
| FL | Federated Learning: a sub-field of Machine Learning focusing on settings in which multiple entities (often referred to as clients) collaboratively train an ML model while ensuring that their data remains decentralised. |
Fig. 1. Volume of data/information created, captured, copied, and consumed worldwide from 2010 to 2020, with forecasts from 2021 to 2025, according to Statista (source)
Fig. 2. The data processing cycle
Fig. 3. The entanglement of Data Processing and Artificial Intelligence
Fig. 4. The Timeworx.io data processing protocol
Fig. 5. Sample for an electrical energy index card (source)
Fig. 6. The virtuous cycle in Timeworx.io
**Fast Track**
Go directly to the [Revenue Model](/the-solution/revenue-model) if you are not interested in the technical aspects of defining Data Processing Pipelines.| Data Type | MIME Types | Description | Analysis Methods |
| Text | text/plain, text/html, text/css, text/javascript text/markdown | Text data includes anything in written form. This could be blog posts, tweets, product reviews, or any other type of written content. | Text analysis methods could include sentiment analysis, topic modelling, keyword extraction, named entity recognition, etc. |
| Image | image/jpeg, image/png, image/svg+xml, image/gif | Image data involves any type of graphics or images. This could range from user-uploaded photos to satellite imagery. | Image analysis methods could include object detection, image classification, segmentation, facial recognition, etc. |
| Video | video/mpeg, video/mp4, video/webm | Video data involves moving visual images. It could be anything from short video clips to full-length movies. | Video analysis methods could include scene recognition, action recognition, object tracking, video summarization, etc. |
| Audio | audio/midi, audio/mpeg, audio/webm, audio/ogg, audio/wav | Audio data includes any type of sound, music, or speech content. | Audio analysis methods could include speech recognition, music classification, audio fingerprinting, emotion recognition from speech, etc. |
Tabular | text/csv, application/vnd.ms-excel, application/vnd.openxmlformats-officedocument.spreadsheetml.sheet | Tabular data is structured data that's organised in columns and rows, similar to what you'd see in a spreadsheet or a database table. | Tabular data analysis methods could include statistical analysis, regression analysis, time series analysis, machine learning models, etc. |
Fig. 7. A sample of a receipt processing pipeline
**Fast Track**
Go directly to the [Objectives for the future of AI](/the-solution/objectives-for-the-future-of-ai) if you are already familiar with the basic concepts for Machine Learning.
{% endhint %}
Fig. 8. Typical workflow for a ML model
**Fast Track**
Go directly to [Decentralised Data Labelling](/ai-that-is-fair/decentralised-data-labelling) if you are already familiar with the basic concepts for Data Labelling.
{% endhint %}
# Data Labelling in a Nutshell
> "The biggest bottleneck for AI development is data. There's not enough of it, and it's expensive to get and label." Charles Wong, CEO of Bifrost
Data labelling encompasses assigning distinct labels to collected data and categorising it into specific groups. For instance, images containing people's faces are separated; each individual is identified and tagged with a name. This labelling process is conducted on extensive data, encompassing diverse identifiers and millions of labels. ML models then utilise these labels to make predictions based on the data pointers. Any errors in the labelling process can lead to inaccuracies in the models. However, market players involved in collecting and labelling data can be of much help in addressing such challenges towards enhancing the efficiency of current ML models. The ever-increasing popularity of the Internet and smartphones is constantly reshaping the business industry, bringing an increase in the utilisation of both social media marketing and data-driven decision making. Data from various sources, including social media, drive business intelligence strategies today.
Consequently, there has been a substantial increase in data generation and utilisation, fueling the data collection and labelling market growth which is expected to grow to $9.07bln by 2028. This market is primarily driven by the adoption of machine learning, artificial intelligence, and big data, which require data annotators, software, and processes to create a foundation for ML models. However, collecting, sorting, and labelling different forms of data can take time and effort. To address these challenges, industry players are providing innovative solutions, such as cloud-based automated image organisation, customised platforms, and AI-driven data labelling. Retail, e-commerce, IT, and telecom sectors are among the prominent adopters of these solutions, capitalising on social media for marketing, consumer insights, and business growth. The major factors that are expected to drive the future growth of data collection and labelling markets include the increase in digital browsing and sales in e-commerce, and the increasing need for better and smarter data in the IT\&C industry.
The current state of the art in data labelling involves a combination of cutting-edge technologies and crowdsourcing methods. Automated data labelling is now possible due to technological advancements, such as computer vision algorithms and natural language processing techniques. These automated approaches leverage ML models to recognise patterns and assign labels to data with little-to-none need for humans in the loop. However, manual data labelling remains crucial for complex and nuanced tasks that require human intuition and expertise. To handle large-scale labelling requirements, crowdsourcing platforms have emerged as a valuable resource. Crowdsourcing enables the distribution of labelling tasks to many individuals, often through online platforms, allowing faster and more cost-effective data annotation. Crowdsourcing not only leverages human intelligence but also facilitates diversity in labelling perspectives, enhancing the quality of the labelled data. This combination of advanced technologies and crowdsourcing has become the state of the art in data labelling, enabling efficient and accurate annotation at scale.
# Problems in Data Labelling
When discussing the "black boxes" of AI, the focus is often on the technical aspects, such as the inability to discern why a machine makes certain decisions, leading to unease. However, there's another opaque aspect: data labelling. This grey area, a metaphorical black box let’s say, involves potential ethical issues, including the overworking and underpayment of workers.
As Forbes points out, the software industry's "dark side" is highlighted by the necessity of labels, known as "ground truth," for most AI applications, leading to the rise of a data labelling industry. This sector could be considered the [blue-collar work of the future](https://www.bbc.com/news/technology-46055595), encompassing significant ethical considerations.
Datasets are collected from various online and external sources and are categorised based on their nature, type of data, and distinct features. Data labelling is the process of identifying raw data (like images, text files, or videos) and adding meaningful tags or labels to provide context so that a machine learning model can learn from it.
This process is crucial in AI and ML, as we have [previously mentioned](/the-solution/machine-learning-in-a-nutshell), because labelled data serves as the training set which teaches the algorithms to recognise patterns and make decisions. Without labelled data, a machine learning model would struggle to understand the input it's given or make accurate predictions. It's like trying to learn a new language without a dictionary; without understanding what each word means, it's way more difficult, though not impossible, to grasp the language's structure or communicate effectively.
The current approaches to data labelling face several critical issues, primarily stemming from their highly centralised and non-transparent nature. This centralised control often leads to a lack of transparency in how data is collected, labelled, and used, raising concerns about data privacy and ethical handling.
One of the most significant problems is the presence of considerable friction and barriers in the process, making it slow and inefficient. According to the AI Index report, one of the top barriers to scaling existing AI initiatives is the challenge of obtaining more data or inputs to train a model, cited by 44% of leaders. This difficulty in acquiring adequate data slows down the development of AI models and hampers innovation.
Furthermore, the issue of very low wages for data labellers is a significant concern. Data labelling is often outsourced to workers in low-income countries, who [are paid minimal wages](https://www.wired.com/story/millions-of-workers-are-training-ai-models-for-pennies/) for repetitive and time-consuming tasks. This not only raises ethical concerns about fair labour practices but also affects the quality of data labelling. Low wages can lead to low morale and less incentive for labellers to ensure high accuracy, directly impacting the quality of AI models that rely on this labelled data.
It’s important to know these things and to understand the implications of improper and unethical practices. Well, it might be convenient - as the [Wired article](https://www.wired.com/story/millions-of-workers-are-training-ai-models-for-pennies/) quotes “From the clients’ perspective, the invisibility of the workers in micro-tasking is not a bug but a feature”.
Let’s not forget that data labelling is an already very busy market with lots of big players, such as Amazon’s Mechanical Turk, Appen, LabelBox, Scale AI, Hive Micro, MIghty AI, Remotaks and many more.
Throughout the years, an increasing number of reports have been filed [mentioning](https://aijourn.com/ais-race-to-the-bottom-why-we-can-no-longer-ignore-the-exploitative-practices-in-data-labeling/) unfair compensation, inadequate working conditions and mistreatment of data labellers around the world, leading to the newly coined term of data colonialism. For reference, read this comprehensive [article](https://www.technologyreview.com/2022/04/20/1050392/ai-industry-appen-scale-data-labels/) by MIT Technology Review, which uses the example of crisis-stricken Venezuela as a cheap labour market, and offers an in-depth perspective on big companies’ practices.
# Decentralised Data Labelling
At Timeworx.io, the approach to data labelling is set to be distinctly different, ethical and innovative. The focus is on decentralising the system to create a financial model where prices are dictated by the market, not by the platform. This shift ensures that prices are transparent and public. While it doesn't necessarily guarantee higher earnings for everyone, it does promise more transparency for all participants in the process.
Moreover, we aim to transform data labelling into a fun and engaging activity that people can enjoy and earn from in their spare time. The goal is to reach larger communities, educating them about data labelling, its implications, and how it contributes to AI development. This approach is designed to reduce friction and lower barriers to entry, thereby attracting a more diverse group of people. This diversity is key to reducing bias and increasing fairness in AI.
The main user interface for Human Agents is our native mobile application, available on both Android and iOS. The main complexity of the app resides in the capability of “translating” complex data processing tasks into a set of clear and actionable operations that users have to perform in a simple and intuitive user interface. Thus, the mobile application represents the entry point for Human Agents in joining the decentralised data processing protocol.
Whenever data processing pipelines are activated, the generated Tasks are distributed to our scalable crowdsource of Human Agents, which are responsible for solving them in exchange for a TIX reward. Agents, in turn, can choose to process tasks based on the offered price, complexity, and their personal preferences or expertise. Upon completion, the outputs are proposed to enter a Consensus round. The outcome of the Consensus round is delivered back to the platform, and it also determines the performance of each Human Agent which is further stored as an on-chain proof.
Fig. 9. The golden spiral of the Fibonacci scale
Fig. 10. Punch Card NFT
Fig. 11. Timeworx.io Connect Screen
Fig. 12. Tasks tab
Fig. 13. Vault tab
Fig. 14. Profile tab
**Fast Track**
Go directly to the [Federated Learning Protocol](/ai-that-is-privacy-enhancing/federated-learning-protocol) if you are already familiar with the basic concepts for Data Privacy and Federated Learning.
{% endhint %}
# Data Privacy in AI
> “Digital evolution must no longer be offered to a customer in trade-off between privacy and security. Privacy is not for sale, it's a valuable asset to protect.” Stephane Nappo, Global Chief Information Security Officer at Groupe SEB
Data, especially when sourced from users or sensitive sectors, needs to be treated with utmost confidentiality. Ensuring data privacy during the labelling process becomes an imperative. To illustrate the gravity, consider a data labelling project for a healthcare firm, where patient medical records, scans, and histories are being annotated. A breach in this data not only compromises the personal information of the patients but could also expose the healthcare provider to severe legal and financial repercussions, not to mention a tarnished reputation.
Collecting, handling and processing data comes with its fair share of laws and regulations, the most important being the [General Data Protection Regulation](https://gdpr-info.eu/) (GDPR) in the EU, and the [California Consumer Privacy Act of 2018](https://www.oag.ca.gov/privacy/ccpa) (CCPA) in the US. Through these acts, personal data is safeguarded through legislation and business practices are required to reflect data sovereignty and ethics not just inside of the national territory in which they operate, but even [outside of the border](https://commission.europa.eu/document/fa09cbad-dd7d-4684-ae60-be03fcb0fddf_en).
With both data privacy and AI in the limelight, the United Nations specialised Agency for ICT - the [International Telecommunication Union](https://www.itu.int/en/Pages/default.aspx) (ITU-T) - has established the [AI for Good](https://aiforgood.itu.int/) programme: an ongoing series of Webinars in which the AI community is able to connect, identify issues, discuss solutions in AI towards establishing global sustainable development goals.
A key initiative for AI for Good has been the constitution of the [Trustworthy AI programme](https://www.itu.int/en/ITU-T/Workshops-and-Seminars/2022/0901/Pages/TrustworthyAI.aspx) created with the sole purpose of standardising [Privacy-Enhancing Technologies](https://en.wikipedia.org/wiki/Privacy-enhancing_technologies) (PETs). These technologies are focused on empowering people and ensuring the protection of Personal Identifiable Information (PII) through minimising the use of personal data and maximising security measures. The overall scope is to create technologies that can limit access to personal information while providing the same excellence in service delivery, with examples ranging from homomorphic encryption and zero-knowledge proofs to federated learning.
In an effort to address the challenges raised by emerging data privacy acts, as well as the [recent increase](https://datainnovation.org/2023/09/overcoming-barriers-to-data-sharing-in-the-united-states/) in data silos, [federated learning](https://en.wikipedia.org/wiki/Federated_learning) (FL) has been introduced as a novel approach for decentralised machine learning model training. Data is no longer centrally stored, rather it is distributed to multiple data nodes. The AI training is carried out on each of these nodes and then aggregated into a global machine learning model based on every node’s contribution. Essentially, your data never has to leave your side, it’s the AI training process that gets distributed in an effort to promote data privacy & minimisation. Let’s take a deeper look!
# Federated Learning in a Nutshell
In the artificial intelligence field, data is the cornerstone of training machine learning models. Generally, vast amounts of data are required to be uploaded to centralised servers, where the data can be further processed and used for training and fine-tuning the AI models of the future. Since the data might be sensitive, ensuring its privacy is imperative. Furthermore, with stricter laws and regulations for data privacy protection, this centralised approach, which can potentially leak data, is increasingly harder to implement.
Let’s take for instance an example extracted from a mundane activity: while driving your car, the navigation system is trying to recommend destinations based on your previous history. Yet again, this is a clear example of how AI can help us concentrate more on driving, then on setting up our navigation system. However, for training such a machine learning model, you (and everyone else) would need to upload all of your GPS coordinates to a centralised server where it can be processed. You might argue that this data is not personally identifiable, but, at the same time, if such data were leaked, it could pose a serious threat to your safety.
This is where federated learning comes into the picture, with its novel collaborative and decentralised approach. Federated learning is a machine learning technology in which multiple entities collaborate in training an AI model under the orchestration of a central server. As opposed to traditional machine learning in which all of the processing is done centrally, in FL many clients, ranging from mobile phones to entire organisations, are in charge of training the AI model locally. This means that data never has to leave the device, rather the AI model is distributed to all of the devices, where it is trained, as illustrated in Figure 15. Thereafter, all of the locally trained machine learning models are uploaded back to the central orchestrator, leaving no trace of the data that they were trained on. Finally, the orchestrator aggregates all of the locally trained models into a global AI model, based on each contribution.
Fig. 15. Federated Learning protocol (source)
**Fast Track**
Go directly to the [Decentralised Inference Protocol](/ai-that-is-trusted/decentralised-inference-protocol) if you are already familiar with the basic concepts for Trust in AI.
{% endhint %}
# Trust in AI
> “The world of enterprise software is going to get completely rewired. Companies with untrustworthy AI will not do well in the market.” Abhay Parasnis, CEO of Typeface
AI has the potential to disrupt and innovate our society on every level and in every industry, to transform the ways in which we work, and to make it easier and safer for us. There are countless examples in which AI and robotics have been used to [automate dangerous and high-risk tasks](https://www.unite.ai/can-ai-help-in-the-worlds-most-dangerous-jobs/) in mining, commercial truck driving, agriculture and even space exploration. Furthermore, AI also holds immense potential for automating all of the mundane tasks that we have to do, day by day.
Currently, the landscape is dominated by large tech companies such as Google, Meta and OpenAI which drive innovation and dictate the tone. To better understand the trends in public opinion relative to AI, it is essential to analyse the most popular AI tools that have become intensely circulated lately. The most popular AI breakthrough in the last few years is, without a doubt, [OpenAI’s ChatGPT](https://chat.openai.com/), a conversational chatbot powered by an LLM. Amassing more than [180 million users](https://explodingtopics.com/blog/chatgpt-users), it has been the main trigger in kicking off the Generative AI race. Other examples include [Stable Diffusion](https://stability.ai/blog/stable-diffusion-public-release) - a deep learning text-to-image model used to generate images based on prompts and [GitHub’s Copilot](https://github.com/features/copilot) - an AI tool for generating code.
However, we must not confuse popularity with trust. With little education in AI & ML, the general public does not have the tools necessary to properly understand who to trust. In spite of all of the breakthroughs in AI, many issues are surfacing when it comes to [how the data is sourced](https://www.reuters.com/legal/litigation/artists-take-new-shot-stability-midjourney-updated-copyright-lawsuit-2023-11-30/), [how it is used](https://www.theverge.com/2024/4/6/24122915/openai-youtube-transcripts-gpt-4-training-data-google), and [what it produces](https://www.nytimes.com/2024/02/22/technology/google-gemini-german-uniforms.html), with more AI incidents and controversies being reported in [AIAAC](https://www.aiaaic.org/aiaaic-repository/) with each passing year.
It comes as no surprise that the [2021 global survey](https://www.ipsos.com/sites/default/files/ct/news/documents/2022-01/Global-opinions-and-expectations-about-AI-2022.pdf) carried out by IPSOS and Pew Research uncovered that more than 50% of the population doesn’t trust companies that use AI, as much as the others, with numbers hitting even lower in the US, at 35%. That same year, the [poll](https://wrp.lrfoundation.org.uk/LRF_2021_report_a-digtial-world-ai-and-personal-data_online_version.pdf) conducted by Lloyd’s Register Foundation together with Gallup also found that around 30% of the populus considers AI as potentially harmful to mankind. This trend is not improving in time, with [KPMG’s 2023 survey](https://kpmg.com/au/en/home/insights/2023/02/trust-in-ai-global-insights-2023.html) revealing that only 40% of Australians trust AI products.
In contrast with the public opinion and growing concerts about its maturity, governments and the industry are pushing the AI revolution across sectors, with deployments reaching a national scale, in mission critical areas such as [electrical grids and food chains](https://huggingface.co/blog/the-age-of-ml-as-code). As an equal and opposite reaction, the scientific community is retaliating through initiatives such as [Hugging Face](https://huggingface.co/blog/the-age-of-ml-as-code), in an effort to democratise machine learning, and to educate and open up machine learning to the software engineering community as a whole. The “Machine Learning For The Masses!” mantra is gaining traction with many AI enthusiasts of all technical backgrounds joining in.
# Decentralised Inference Protocol
At Timeworx.io, we believe these efforts need to be pushed beyond the software engineering bubble, and focused on a societal level, since the problems we are faced with are starting to reach the fabric of our societies.
Our approach is trustless: no Agent is considered trusted, whether human or AI. All data processing outcomes need to be deemed correct only through Consensus achieved using the decentralised data processing protocol.
AI Agents can be registered in the platform as Agent Nodes - AI models are packaged and deployed using our standard Agent Node software that allows them to participate in the data processing protocol. Once a Node is connected into the protocol, then it is able to use its packaged AI models to solve data processing Tasks in exchange for TIX rewards in a permissionless manner. All Agent Node providers are connected to each other in a peer-to-peer network, relay information through gossiping, and run data processing on their own infrastructure in line with our commitment to DePIN.
The registration process for an Agent Node requires that it creates a Staking Pool in which it locks a sufficient amount of TIX, thus securing the protocol using a Proof-of-Stake mechanism. The same Staking Pool is also used to receive rewards from the data processing protocol.
Every Task generated by the platform is distributed to all AI Agents that are able to process it. AI Agents take turns running inference (predictions) as instructed in the Task for generating Task outcomes, and vote on the results. In an attempt to reduce the energy consumption of all Agent Nodes bound in the protocol, each Task is processed by a single node, called the Proposer, while the other Nodes participate in the Consensus round by validating its outcome. Therefore, all Agent Nodes employ gossiping to reach Consensus. A task outcome is deemed correct if at least 2/3 of Nodes agree to validate the proposed result. Upon reaching consensus, the outcome is delivered to the customer, and the performance metrics are updated for all Agent Nodes involved in the data processing round.
Performance metrics are updated for both the Proposers, as well as for the Validators, in reference to the correctness of their solution as deemed by the Consensus round. In case the performance score of a given Agent Node falls below a predetermined threshold, the provider is penalised and the Staking Pool is even slashed if this behaviour persists.
Proposers are chosen in round-robin fashion with a frequency proportional to their data processing power. The data processing power is computed as a balance between the size of the Staking Pool and the performance score of each Agent Node. Since Proposers earn more rewards than Validators, the platform incentivises Agent Nodes to both increase their stake in the protocol, while still performing data processing at high standards of quality and performance.
Not all Agent Nodes participate in every Consensus round, rather this is a configuration that is done by the customer when setting up the Agent Block in the data processing pipeline. The platform does not allow setting the saturation capacity to less than 3 nodes, but the customer can configure any number higher than this based on the availability of funds. Since there is no causal relationship between data processing Tasks, Agent Nodes can deploy any level of parallelism they see fit, as long as they are able to maintain their performance metrics as competitive as possible.
{% hint style="info" %} **Fast Track**
Go directly to [Delegation of Trust](/ai-that-is-trusted/delegation-of-trust) if you are not interested in Performance Monitoring for AI.
{% endhint %}
# Performance Monitoring
Determining the performance metrics of each Agent Node is crucial for evaluating the performance and effectiveness of the AI models that are deployed. This ensures that Agent Nodes are incentivised to perform at high levels of quality and performance, and be properly and fairly compensated for it.
Since not all AI models can be evaluated in the same way, the platform needs to adapt to each case in choosing the best evaluation metrics depending on the specific data processing requirements. Evaluation metrics need to be defined as objective criteria to be able to determine the overall quality of an AI model in accurately predicting outcomes, as well as in its ability to generalise its inference in every given context. Based on a given type of Task, the platform determines the appropriate evaluation metrics that it applies to all Agent Nodes participating in each Consensus round. These metrics are further aggregated into a performance score for each Agent Node in part.
Before diving into the performance metrics, it is important to understand that predictive AI models fall in one of two categories: regression and classification. Whenever we talk about classification models, we refer to binary outputs (Yes vs No) or nominal outputs (Cat vs Dog vs Rabbit). On the other hand, regression models refer to continuous outputs, such as predicting the next temperature based on a given series of thermometer readings.
### **Classification Metrics**
Since classification AI models have discrete outputs, the evaluation metrics need to be able to discern whether the predicted output falls into the correct class.
Before jumping into metrics, we need to define a few terms:
1. True Positive (TP): the number of positive classes predicted correctly, e.g. the AI model was able to predict that the sound was made by a cat, and it was correct
2. True Negative (TN): the number of negative classes predicted correctly, e.g. the AI model was able to predict that the sound was not made by a cat, and it was correct
3. False Positive (FP): the number of positive classes predicted incorrectly, e.g. the AI model was able to predict that the sound was made by a cat, and it was not correct
4. False Negative (FN): the number of negative classes predicted incorrectly, e.g. the AI model was able to predict that the sound was not made by a cat, and it was not correct
The platform defines the following classification metrics:
* **Accuracy**: the ratio between correct predictions and the total number of predictions.
* **Precision**: the ratio between true positive (TP) and total positives that have been predicted (TP + FP)
* **Recall**: the ratio of true positives (TP) to all of the positives in ground truth (TP + FN)
* **F1-score**: the harmonic mean between Precision and Recall\
$$F\_1 = \frac{2}{\frac{1}{Precision}+\frac{1}{Recall}}$$
### **Regression Metrics**
Since regression AI models have continuous outputs, the evaluation metrics need to determine how close the predicted output was to the actual output, in terms of the numerical distance between them:
The platform defines the following regression metrics:
* **Mean Absolute Error** (MAE): the average difference between the predicted values and the ground truth
* **Mean Squared Error** (MSE): the average of the squared difference between predicted values and the ground truth
* **Root Mean Squared Error** (RMSE): the square root of the average of the squared difference between predicted values and the ground truth
* **R2 coefficient of determination**: compound metric to determine how much the total variation in the ground truth is explain by the variation in predicted values
# Delegation of Trust
Aside from improving their performance scores, Agent Nodes can increase their odds of being selected as Proposers by maximising their Staking Pools. The decentralised data processing protocol lends a helping hand by allowing token holders to delegate their TIX to an Agent Node’s Staking Pool in exchange for a proportional share of the revenue.
This approach creates a positive network effect between Agent Nodes and token holders, in a win-win situation. Agent Nodes can attract delegators by boasting state-of-the-art AI models and high performance scores. At the same time, token holders can increase their base staking rewards with an additional source of revenue.
As Agent Nodes become increasingly popular, saturation of rewards ensues, and a choice must be made on either side. On the one hand, providers can create additional Staking Pools by spinning up more Agent Nodes. On the other hand, token holders can choose to delegate to a different Agent Node with high performance and higher gains.
This approach not only incentivises Agent Nodes to perform at the highest level, but also encourages them to communicate more with the audience in an effort to educate them about the inner workings of their AI models. At the same time, token holders are incentivised into learning more about AI & ML, as well as to get involved and voluntarily participate in [citizen science](https://en.wikipedia.org/wiki/Citizen_science).
# Utility
The core of the platform is the decentralised protocol for data processing that is governed by the TIX utility token for ensuring transparency, traceability and fairness. Any business, acting as a customer of the platform, can request data processing services using TIX, and can submit all of its collected datasets along with instructions on how the data is intended to be processed. From then on, the protocol binds Agents, whether human or AI, as service providers that process the data as intended in exchange for TIX.
The TIX token is designed to ensure that businesses obtain the ground truth from data processing through a trustless implementation of the decentralised protocol in which no Agent, whether human or AI, is considered trusted. Trust in each Agent is under continuous observation, both in terms of performance, and in the investment, or stake, they have put in the protocol.
### **Trust in humans**
All Human Agents that join the decentralised data processing protocol, need to first earn their Punch Card NFTs by solving tasks. This is both a proof of the users’ intelligence and humanity, and it is also a requirement for the platform to be able to create initial performance profiles for each Human Agent, as a baseline for measuring future improvements. Since the Punch Card NFT acts as an on-chain proof of performance for each Human Agent, its value does not reside in trading, but in holding. Without a Punch Card, no transactions can be issued for claiming or staking rewards after solving tasks. Once it has been unlocked, Human Agents are able to either claim their TIX, or lock it through staking, thus securing the protocol even further.
The performance of each Human Agent is constantly monitored and assessed for ensuring the quality of data processing through four KPIs: Speed, Proficiency, Reliability and Versatility. Based on these measurements, the platform implements a range of gamified incentives for ensuring that our Human Agents are responsible, engaged and are constantly seeking to improve their performance for increasing their rewards.
### **Trust in AI**
AI Agents can be registered in the platform as Agent Nodes - AI models are packaged and deployed using our standard Agent Node software that allows them to participate in the data processing protocol. Once a Node is connected into the protocol, then it is able to use its packaged AI models to solve data processing Tasks in exchange for TIX rewards in a permissionless manner. The registration process for an Agent Node requires that it creates a Staking Pool in which it locks a sufficient amount of TIX, thus securing the protocol using a Proof-of-Stake mechanism.
AI agents receive rewards proportional to their data processing power which is computed as a balance between the size of their Staking Pools and their performance scores in processing data. Thus, the platform incentivises Agent Nodes to both increase their stake in the protocol, while still performing data processing at high standards of quality and performance.
### **Delegated Trust**
In Timeworx.io, all agents are incentivised to up their game towards obtaining more revenue. On the one hand, AI Agents always seek to maximise their Staking Pools for a higher chance of becoming Proposers. On the other hand, TIX holders are interested in expanding their staking rewards even further.
The platform brings them both together through delegation of trust: holders can delegate their TIX to an AI Agent’s Staking Pool. By doing so, both parties become more responsible actors in the decentralised processing protocol. Therefore, Agent Nodes will strive to perform at the highest level, but will also need to educate the public about their innovations, in order to attract more delegators. At the same time, token holders will need to learn more about AI & ML, to make a more informed decision before delegating to Agent Nodes.
.png?alt=media&token=e98c5c01-4ad8-4fc7-bc0e-bbcb7cd2e5b5)
Fig. 16. The TIX token distribution
Fig. 17. The TIX supply evolution in time
PhD student with a focus on fine-grained access control methodologies for Blockchain. With more than 12 years of mobile application and Blockchain development, Alex is the engineering manager in charge of running one of the largest staking providers in the MultiversX blockchain.
| |Summa Cum Laude Phd graduate in 2021 for his research into opportunistic computing, industry-recognized expert in mobile development, with more than 12 years of experience as a software architect and as a technical community leader. Currently, Radu is a lecturer and a postdoc researcher at the University Politehnica of Bucharest, studying fine-grained access control methodologies for Blockchain.
| |Is an economist, economic journalist and former Romanian politician. As Minister of Public Finance in the Tariceanu Government, he introduced the flat tax in Romania as early as 2005. Previously, he was editor-in-chief of the economic profile magazine Capital. Ionut is an entrepreneur, an early and eager supporter of Blockchain and Web3. His visionary insights and holistic view over cryptocurrency markets are critical to designing a financial model for Timeworx.io.
| |UI/UX designer with more than 10 years of experience, in love with photography, art, and architecture. Has creative flair and up-to-date knowledge of industry trends, visual-thinking problem solver.
| |Android software engineer with more than 12 years of experience as tech & team leader in building large scale mobile applications.
| |iOS software engineer with more than 10 years of experience with latest mobile technologies and long-term crypto enthusiast.
| |An experienced marketing professional, starting her career in 360 advertising campaigns management, in top rated international agencies like Ogilvy and Publicis, and since 2012 building on marketing strategy, branding, integrated marketing, team building, management & leadership in industries as FMCG, Retail, Oil & Gas, Pharma, Banking, and Tech.
| |An exceptional mix between a digital communication creative strategist and brand storyteller, journalist, and a person who is deeply concerned about social issues and involved in NGOs.
|Andrei Stan, drd., MBA, FCCA, advisor on Product
is a doctoral candidate in Business Administration (Aston University). Using his doctoral research, he aims to help Timeworx.io strengthen its competitive position designing the business model and leveraging web3 governance models. He is a Fellow of the ACCA in the UK, holds an MBA from Oxford Brookes, and is a former Senior Partner of Moore.
| |Ciprian Dobre, PhD, advisor on Research & Academics
is a professor within the Computer Science Department, and Vice-Rector of the National University of Science and Technology POLITEHNICA Bucharest. He represents Romania in the EUA RISG board with the EU, and is the NUSTPB representative in EOSC. He aims to guide the project towards open science & open data principles, as well as highly-scalable decentralised and distributed system architectures.
| |Arturas Feiferas, advisor on Finance & Compliance
is a seasoned professional with over 30 years of experience in corporate and retail banking across eight countries. He has held executive and non-executive board positions in various financial institutions, and is currently contributing to the project through his expertise in strategic development, corporate governance and risk management.
| # Keep in touch * Website: