We have posted a number of times discussing aspects of Artificial Intelligence (AI) in project management, but what exactly is AI?  This post looks at the components in the AI ecosystem and briefly outlines what the various terms mean.

𝗔𝗿𝘁𝗶𝗳𝗶𝗰𝗶𝗮𝗹 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲: a range of computer algorithms and functions that enable computers and machines to mimic the problem-solving and decision-making capabilities of the human mind.

• Automatic Programming: is a technology that enables computers to generate code or write programs with minimal human intervention.

• Knowledge Representation: is concerned with representing information about the real world in a way that a computer can understand, so it can utilize this knowledge and behave intelligently.

• Expert System: is a computer system emulating the decision-making ability of a human expert. A system typically includes: a knowledge base, an inference engine that applies logical rules to the knowledge base to deduce new information, an explanation facility, a knowledge acquisition facility, and a user interface.

• Planning and Scheduling: an automated process that achieves the realization of strategies or action sequences that are complex and must be discovered and optimized in multidimensional space, typically for execution by intelligent agents, autonomous robots, and unmanned vehicles.

• Speech Recognition: the ability of devices to respond to spoken commands. Speech recognition enables hands-free control of various devices, provides input to automatic translation, and creates print-ready dictation.

• Intelligent Robotics: robots that function as an intelligent machine and it can be programmed to take actions or make choices based on input from sensors.

• Visual Perception: enables machines to derive information from, and understand images and visual data in a way similar to humans

• Natural Language Processing (NLP): gives computers the ability to understand text and spoken words in much the same way human beings can.

• Problem Solving & Search Strategies: Involves the use of algorithms to find solutions to complex problems by exploring possible paths and evaluating the outcomes. A search algorithm takes a problem as input and returns a solution in the form of an action sequence.

𝗠𝗮𝗰𝗵𝗶𝗻𝗲 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴: is concerned with the development and study of statistical algorithms that allow a machine to be trained so it can learn from the training data and then generalize to unseen data, to perform tasks without explicit instructions. There are three basic machine learning paradigms, supervised learning, unsupervised learning, and reinforcement learning.

• Supervised learning: is when algorithms learn to make decisions based on past known outcomes. The data set containing past known outcomes and other related variables used in the learning process is known as training data.

• Unsupervised learning: is a type of machine learning that learns from data without human supervision. Unlike supervised learning, unsupervised machine learning models are given unlabelled data and allowed to discover patterns and insights without any explicit guidance or instruction.

• Reinforcement Learning (RL): is an interdisciplinary area of machine learning concerned with how an intelligent agent ought to take actions in a dynamic environment to maximize the cumulative reward.

• Classification: a process where AI systems are trained to categorize data into predefined classes or labels.

• K-Means Clustering: cluster analysis is an analytical technique used in data mining and machine learning to group similar objects into related clusters.

• Principal Component Analysis (PCA): is a dimensionality reduction method used to reduce the dimensionality of large data sets, by transforming a large set of variables into a smaller one that still contains most of the information in the large set.

• Automatic Reasoning: attempts to provide assurance about what a system or program will do or will never do based on mathematical proof.

• Decision Trees:  is a flow chart created by a computer algorithm to make decisions or numeric predictions based on information in a digital data set.

• Random Forest: is an algorithm that combines the output of multiple decision trees to reach a single result. It handles both classification and regression problems.

• Ensemble Methods: are techniques that aim at improving the accuracy of results in models by combining multiple models instead of using a single model. The combined models increase the accuracy of the results significantly.

• Naive Bayes: is a statistical classification technique based on Bayes Theorem. It is one of the simplest supervised learning algorithms.

• Anomaly Detection: the identification of rare events, items, or observations which are suspicious because they differ significantly from standard behaviours or patterns.

𝗡𝗲𝘂𝗿𝗮𝗹 𝗡𝗲𝘁𝘄𝗼𝗿𝗸𝘀: are machine learning (ML) models designed to mimic the function and structure of the human brain and help computers gather insights and meaning from text, data, and documents by being trained to recognising patterns and sequences.

• Large Language Model (LLM): is a type of neural network called a transformer program that can recognize and generate text, answer questions, and generate high-quality, contextually appropriate responses in natural language. LLMs are trained on huge sets of data.

• Radial Basis Function Networks: are a type of neural network used for function approximation problems. They are distinguished from other neural networks due to their universal approximation and faster learning speed.

• Recurrent Neural Networks (RNN): is a type of neural network where the output from the previous step is used as input to the current step. In traditional neural networks, all the inputs and outputs are independent of each other. For example, when it is required to predict the next word of a sentence, the previous words are required and hence there is a need to remember the previous words.

• Autoencoders: is a type of neural network used to learn efficient coding of unlabelled data (unsupervised learning). An autoencoder learns two functions: an encoding function that transforms the input data into code, and a decoding function that recreates the input data from the encoded representation.

• Hopfield Networks: is a recurrent neural network having synaptic connection pattern such that there is an underlying Lyapunov function (method of stability) for the activity dynamics. Started in any initial state, the state of the system evolves to a final state that is a (local) minimum of the Lyapunov function.

• Modular Neural Networks: are characterized by a series of independent neural networks moderated by some intermediary to allow for more complex management processes.

• Adaptive Resonance Theory (ART): is a theory developed to address the stability-plasticity dilemma. The terms adaptive and resonance means that it can adapt to new learning (adaptive) without losing previous information (resonance).

Deep Learning:  is a method in artificial intelligence (AI) that teaches computers to process data in a way that is inspired by the human brain. Deep learning models can recognize complex patterns in pictures, text, sounds, and other data to produce accurate insights and predictions. The adjective deep refers to the use of multiple layers in the network.

• Transformer Model:  is a neural network that learns context and thus meaning by tracking relationships in sequential data by applying an evolving set of mathematical techniques to detect subtle ways even distant data elements in a series influence and depend on each other.

• Convolutional Neural Networks (CNN): is a regularized type of feed-forward neural network that learns feature engineering by itself via filters or kernel optimization.

• Long Short-Term Memory Networks (LSTM): is a recurrent neural network (RNN), aimed to deal with the vanishing gradient problem present in traditional RNNs.

• Deep Reinforcement Learning: is a subfield of machine learning that combines reinforcement learning (RL) and deep learning.

• Generative Adversarial Networks (GAN): is a class of machine learning frameworks for approaching generative AI. Two neural networks contest with each other in the form of a zero-sum game, where one agent’s gain is another agent’s loss.  Given a training set, this technique learns to generate new data with the same statistics as the training set. A GAN trained on photographs can generate new photographs that look at least superficially authentic.

• Deep Belief Networks (DBN): are a type of neural network that is composed of several layers of shallow neural networks (RBMs) that can be trained using unsupervised learning. The output of the RBMs is then used as input to the next layer of the network, until the final layer is reached. The final layer of the DBN is typically a classifier that is trained using supervised learning. DBNs are effective in applications, such as image recognition, speech recognition, and natural language processing.

For more discussion on the use of AI in project management see:
https://mosaicprojects.com.au/PMKI-SCH-033.php#AI-Discussion

I’m wondering what is causing the confusion appearing in so many posts lately concerning the definition of the critical path. Is it:

1. A lack of knowledge?
2. People being out of date and using superseded definitions?
3. People not understanding the difference between a characteristic and a definition?

As most people know (or should know) the definition used by the PMI Practice Standard for Scheduling (Third Edition), the International Standards Organization (ISO) and most other reputable authorities in their standards is similar to:

Critical Path: sequence of activities that determine the earliest possible completion date for the project or phase. 

For more on the development of this definition see: Defining the Critical Path.

To deal with the questions above, in reverse order:

The difference between a characteristic and a definition.

The definition of a phrase or concept (the ‘critical path’ is both) should be a short, concise, statement that is always correct. A characteristic is something that indicates the concept may be present.

Everyone of significance has always agreed the critical path is the sequence of activities determining the earliest possible completion of the project (or if the project has staged completions, a stage or phase).  This is the basis of the current valid definitions. As a direct consequence of this in a properly constructed CPM schedule, the float on the critical path is likely to be lower than on other paths but not always. Low float or zero float is a characteristic that is often seen on a critical path, but this is a consequence of its defining feature, it being longer than other paths. 

Superseded definitions.

In the 1960s and 70s, most CPM schedules were hand drawn and calculated using a day number calendar. This meant there was only one calendar and constraints were uncommon.  When there are no constraints and only a single calendar in use, the critical path has zero float! From the 1980s on, most CPM schedules have been developed using various software tool, all of which offer the user the option to impose date constraints and use multiple calendars (mainframe scheduling tools generally had these features from the 1960s on).

Using more than one calendar can cause different float values to occur within a single chain of activities, this is discussed in Calendars and the Critical Path.  

Date constraints can create positive or negative float (usually negative) depending on the imposed date compared to the calculated date and the type of constraint, this is discussed in Negative Float and the Critical Path.

Consequently for at least the last 40 years, the definition of a critical path cannot be based on float – float changes depending on other factors.

Knowledge?

One of the problems with frequently repeated fallacies is when people do a reference search, they find a viable answer, and then use that information assuming the information is correct. This is the way we learn, and is common across all disciplines.

Academic papers are built based on references, and despite peer review process, can reference false information and continue to spread the falsehood. One classic example of this is the number of books and papers that still claim Henry Gantt developed the bar chart despite the fact bar charts were in use 100 year before Gantt published his books (which make no claim to him having invented the concept), for more on this see: https://mosaicprojects.com.au/PMKI-ZSY-020.php#Barchart. Another common falsehood is Henry Gantt ‘invented project management’ – his work was focused on improving factory production processes: https://mosaicprojects.com.au/PMKI-ZSY-025.php#Overview

Academics are trained researchers, and still make mistakes; the rest of us have a bigger challenge! The spread of un-reviewed publications via the internet in the last 20+ years started the problem. Now Generative AI (Gen AI) and large language models (LLM) are exacerbating the problem. For most of us it is getting harder and harder to understand where the information being presented a person, or in an article originated. Gen AI is built to translate data into language, it has no ability to determine if the data it has found is from a credible source or not. And as more and more text is produced by the various Gen AI tools the more often wrong information will be repeated making it more likely the wrong information will be found and repeated again, and again.   

I’m not sure of the solution to this challenge Gen AI is clearly not skilled in project management practice (even the PMI AI tool), for more discussion on this important topic see: https://mosaicprojects.com.au/PMKI-SCH-033.php#AI-Discussion  

Reference

One reference that is reliable is Mosaic’s Easy CPM.  It incorporates most of what we know, focused on  developing and using an effective schedule in any software tool. The book is designed to provide practical guidance to people involved in developing, or using, schedules based on the Critical Path Method (CPM), and act as a reference and practice guide to enhance the effectiveness of their scheduling practice.

For more see: https://mosaicprojects.com.au/shop-easy-cpm.php 

The best known LMM (Large Language Model) is ChatGPT. The developers acknowledge ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers. Fixing this issue is challenging, as during the LLM training there is no single source of truth (the app is biased towards what it has been taught). The ‘training source’ for ChatGPT is the internet which has 1000s of incorrect posts on various aspects of project management – GIGO applies.  

To overcome this problem a number of developments based on groomed training materials are in development, some proprietary in-house, and some to be publicly available. Two of these are:

Marvin, a collaboration between Projecting Success and Praxis, Mosaic is supporting this development – watch this space in 2024.

PMI’s Project Infinity, based on the vast library of publications and papers held by PMI.  This tool is still in Beta, but as a PMI member I’ve had access to test the model.

Project Infinity Test (1)

The ‘intelligence’ in LLM applications is the ability to understand and reply to language in a coherent way.  You ask a question, the app understands what you have asked, searches its learning, and formulates an answer you can understand. What seems to be missing is the ability to qualify the information in its systems, particularly when specific facts are required.

My test for Project Infinity was based on asking “What is the critical path?”.  The correct answer to this question based on the PMI Practice Standard for Scheduling (Third Edition) and ISO Standards is something similar to: 

Critical Path: sequence of activities that determine the earliest possible completion date for the project or phase.  See: Defining the Critical Path 

As anyone with even a limited knowledge of CPM scheduling knows, ‘Zero Float[1]’, or float in general cannot be used to define the critical path for the following reasons:

1. Date constraints can cause changes in the float value, usually to create negative float, see: Negative Float and the Critical Path
   
2. Multiple calendars can change float values along the critical path, see: Calendars and the Critical

A summary of my ‘chat’ with Project Infinity is:

Q1 – what is the critical path?

       ‘The critical path is the longest sequence of activities in a project plan which must be completed on time in order for the project to be completed on schedule. It is determined by identifying the activities that have zero slack or float time,….’

Q2 – Why are you claiming the critical path has zero float when it may have negative float?

        ‘Negative float can occur when an activity’s late finish date is earlier than its early finish date, which means that the activity is taking longer than planned and is delaying the project. However, the critical path always has zero float because any delay in the critical path activities will cause a delay in the overall project timeline. …….  In summary, while negative float can occur in non-critical path activities, the critical path always has zero float because any delay in these activities will cause a delay in the overall project timeline.’

Q3 – so what happens when the finish milestone and the project overall have a constrained finish?

        ‘When the finish milestone and the project overall have a constrained finish, total float may be used to provide an early indication of potential project completion slippage.’
[Note: I’m assuming ‘Total Float’ can be negative PI did not say so.]

Q4 – Why does your third answer contradict the first answer?

        ‘My third answer does not contradict the first answer. ……’

The full exchange can be seen at Project Infinity Test 1 – Critical Path. 

Conclusions

This simple test shows at the current level of development Project Infinity has a lot or learning to do:

• It ignored key PMI documents, in particular the PMI Practice Standard for Scheduling (Third Edition)
   
• It failed to recognize a direct contradiction in its answers.
  

Therefore, while LLM tools can help save time bringing together information from diverse sources, their training to date is limited to absorbing information from documents, the next stage of development involving qualifying and grading the data may be a way off. So if you do not know the right answer to a question, you cannot rely on an AI tool using LLM to provide you with a way out.  

To make matters worse, accountability in AI is a complex issue. We know AI systems can misstep in numerous ways, which raises questions about who is responsible? This is a complex legal issue and in the absence of someone else who is demonstrably at fault, you are likely to carry the can!

For more on AI in project management see:  https://mosaicprojects.com.au/PMKI-SCH-033.php#AI

[1] The concept of the critical path having zero total float arose in the 1960s when computer programs were relatively simple and most schedules were manually drawn and calculated. With a single ‘Day Number’ calendar and no constraints the longest path in a network had zero float. The introduction of computer programs in the 1980s that allowed multiple calendars and constraints invalidated this definition.

In our previous blog, AI is coming to a project near you!, we identified a large number of project management software applications using Artificial Intelligence (AI) and the rapid spread of the capability. But what exactly is AI? This post offers a brief overview of the concept.   

AI is not as new as some people imagine. Some of the mathematics underpinning AI can be traced back to the 18^th century and many of the fundamental concepts were developed in the 20^th, but there was very limited use of AI. The ability to make widespread practical use of AI required the development of computers with sufficient processing capabilities to process large amounts of data quickly.  Each of the developments outlined below were enabled by better processors and increased data storage capabilities.

Types of AI

The modern concept of intelligent processing is more than 50 years old, but the way a computer application works depends on the design of the application.  Very broadly:

Decision tables have been in software since the 1960s, the decision table applies a cascading set of decisions to a limited set of data to arrive at a result.  The ‘table’ is hard-wired into the code and does not change. Many resource levelling algorithms are based on decision tables to decide what resources get allocated to which activities on each day.

Expert systems, known as rule-based systems in the 1960s, use explicitly crafted rules to generate responses or data sets based on an initial question. These systems were the basis of many automated Chatbots and help systems. The system’s rules are ‘hard-wired’ and do not change without external intervention.

Data mining was developed in the 1990s. The application uses generalized instructions to look at large volumes of data to discover previously unknown properties. Generalized processes means the data being examined does not need to be labelled or predefined. The application works out a suitable structure, and then draws classes information from within the data set. These system can be interactive, but are not self-learning. Extracting knowledge from ‘big data’ supports Business Intelligence and other business and marketing improvement initiatives.

Machine learning (ML). Basic ML is similar to data mining. It concerned with the development and study of statistical algorithms that can effectively generalize to perform tasks without explicit instructions. ML focuses on prediction and recognition, based on properties the application has learned from training data. The basic functions of ML were defined in the early 2000s and the concept continues to evolve and develop. The basic ML approach can be seen in a range of project management tools where the application recommends durations, lists likely risks, or performs other predictive assessments on your current project, based on data from previous projects.

Algorithmic Decision Making, a subset of ‘expert systems’ focused on using conventional machine learning and statistical techniques such as ordinary least squares, logistic regression, and decision trees to automate traditional human-based decision-making processes. Research suggests, well designed algorithms are less biased and more accurate than the humans they are replacing, improving predictions and decisions, but care is needed; they can also perpetuate blind spots, biases and be built to be fundamentally unfair.

Generative AI (Gen AI) extends the capability of ML. Gen AI uses generative artificial neural networks and ‘deep learning’ to deliver enhanced performance. Gen AI has been applied to large language models (LLM), computer vision, robotics, speech recognition, email filtering, agriculture, medicine, and many other fields. Each branch of development takes the basic principles of Gen AI and adapts them to the specific needs of the researchers. Latest trends are linking different strands of Gen AI to create new things such as generating pictures from verbal descriptions, and linking Gen AI to the IoT (Internet of Things) and additive manufacturing functions to produce computer designed ‘stuff’.

Large language models (LLM) are the branch of generative AI with most direct relevance to project management. LLM uses deep learning algorithms that can perform a variety of natural language processing (NLP) tasks. They are trained using massive datasets which enables them to recognize, translate, predict, or generate text or other content. LLM applications must be pre-trained and then fine-tuned so that they can solve text classifications, answer questions, summarize documents, and generate text, sound, or images. The challenge with LLM is in the training materials, the system only knows what it has been taught. This branch of Gen AI burst into prominence with the development of ChatGPT. Its developer, OpenAI (a research company), launched ChatGPT on November 30, 2022 – a year later and ChatGPT has world-wide attention.

LLM underpins most of todays advanced AI applications and can generate content across multiple types of media, including text, graphics, and video. While early implementations have had issues with accuracy and bias the tools are improving rapidly.

Progress to date indicates that the inherent capabilities of Gen AI will fundamentally change enterprise technology, and how businesses operate. Rather than simply requiring technical skills, employees will also need critical thinking, people skills, and a good understanding of ethics and governance processes.

In the four months from January to April 2023 the percentage of employees in Australia using Gen AI grew from 10% to 35%[1].  This rapid growth in use raises concerns around safety, privacy and security but businesses that do not explore the use of Gen AI in their organisations or industry risk being left behind.

The technological world is becoming very closely integrated:

Source: ACS Australia’s Digital Pulse | A new approach to building Australia’s technology skills – click to expand.

For more on the application of AI to project management software see: https://mosaicprojects.com.au/PMKI-SCH-033.php#AI

Our next post will look at the use of LLM in project management.

[1] Australian Computer Society research 2023. Australia’s Digital Pulse | A new approach to building Australia’s technology skills

Like it or not, Artificial Intelligence (AI) is coming to a project near to you. As well as adaptations of the generic tools such as ChatGPT many new tools are being released and established tools upgraded to embed AI in various ways. But care is needed – “Garbage in, Garbage out” can still diminish the value of AI.

Generative AI tool that create content based on their ‘large language model’ need to be trained on quality resources rather than the mass of misinformation floating around the internet. This is not hard to do and can generate impressive results (but be careful of copyright).  Most project management tools with embedded AI are built around machine learning (ML) and learn from the data you generate; the more advanced tools then apply AI to create insights.

These enhanced tools bring almost limitless processing capabilities to give meaning to data and help make crucial decisions to achieve project strategies. They can take over various technical tasks allowing project managers to deal with more crucial tasks as well as improving various estimating and risk assessment processes by provide insights from previous projects to enable managers to do a better job. These systems can also assist by keeping track of the project and checking key metrics like budgets, milestones, and other resources.

AI is still improving, and so are AI project management tools. All of this creates numerous benefits for the project management process. However, here are some core benefits you can expect right away. 

• Better Project Estimations 
• Improved Scheduling and Planning
• More reliable Roadmaps and Budgets
• More Predictability

There are a surprisingly large number of tools embedding ML and/or AI, too many to list here!  What we have done is augment the Mosaic PM Software and Tools listing to highlight tools with some ML or AI capability – look for the blue – AI – in the categorized listings at:
https://mosaicprojects.com.au/PMKI-SCH-030.php 

If you know of any additional tools missing from the list (or tools in the list that should be flagged ‘AI’) let me know and I will update the list.

For more on the application of AI to project management software see: https://mosaicprojects.com.au/PMKI-SCH-033.php#AI