Tuesday, April 21, 2020

What you Can Learn from the Panel on Artificial Intelligence and Project Management

What you Can Learn from the Panel on Artificial Intelligence and Project Management

In February 2020, I was honored to be on the AI Panel discussion at the New Jersey Chapter of Project Management International. George Pace was the moderator and serving on the panel with me was Sandipan Gagopadhyay, David Dalessandro, and Mike Moran. I was selected due to my work managing AI projects in three different companies.

The discussion started out about AI in general and then focused on how Project Management relates to AI. Here are a few of the main points made by the discussion. Thanks to the panelists who have provided additional input below.

Discussion of Four Example Articles on AI The following articles were provided to the chapter members in advance of the meeting. While the panel didn't delve into any particular one in a lot of detail, they were discussed in general. We will explore some of the general intents of these articles in more detail below.

Will It Take Away Jobs?
(Mike Moran provided the following section of this article)
In a sense, AI is a personality test. Some predict that we will have much more leisure time. Others claim that we will all be unemployed. Those two statements are actually the same prediction. It just depends on whether you are an optimist or a pessimist.

What's historically true is that we humans have a difficult time predicting the effects of technology. We are worried about how all the things we are doing now might go away, but we fail to imagine all the new things that will need to be done. In 1790, 90% of the US workforce were farmers. In 1990, it was under 3%. Unemployment wasn't 87%. And today we produce oodles more food than we did in 1790. And the 1990 workforce included droves of women not in the 1790 workforce. What really happened is that technology has made it possible for fewer people to do the things that many people were once needed for, while the rest found other ways to be useful. Technology tends to do that, even if we think every new technology could be the apocalypse. Maybe someday that will be true, but it's probably not that way to bet.

Some people will certainly lose their jobs someday because AI automates away their jobs. But for most people, AI will automate tasks that they currently perform, not their entire job. The people who will lose their jobs to AI are the people who refuse to use AI. They will lose their jobs to those that do.

Take self-driving cars as an example. Despite the hype, there are many thorny issues that need to be resolved for self-driving cars to be more than just a novelty--insurance coverage, changes to roadways, and how parking is configured are just a few. But that's not the real story of AI when it comes to driving. The real story is that AI makes humans better drivers. Many new cars can parallel park themselves. They can tell you when another car is in your blind spot. Or when you are weaving out of your lane. They can apply the brakes in an emergency faster than you can.

Most AI is like that. It doesn't take away your job. It makes you better at your job.

(Thanks Mike for that insight! I agree with that perspective. AI is a work augmentation tool. If AI ever develops to the point of taking away Project Management jobs, I think we have a lot more to be concerned about than our jobs! For more information on this subject, I am providing this article titled "How AI Is Creating Jobs Not Killing Them For Low-Skilled Workers?" on Medium.  Mike's writing reminds me of the phrase “self-fulfilling prophecy.” Some people believe AI will help them in their work and so they embrace it, while others believe it will take their job away, so they ignore it. It's like the quote attributed to Henry Ford: “Whether you think you can or you think you can’t – you’re right!”)

AI needs data and AI is not Magic Pixie dust
  • Some people have the misconception that you can "sprinkle a bit of AI" on something and it automatically becomes magical -it's not that easy and not that magical
  • The AI used in Chatbots, Facial Recognition and image recognition is called "Supervised learning".  For example, if we were training AI to recognize photos of dogs Vs. Cats, it would need a human to manually "label" lots of pictures as dogs and cats. Then the AI uses that information to identify new pictures as either dogs or cats. 
  • I mentioned something during the panel discussion that astonished many in the audience. I told the audience that they are helping train AI! Here's how: you've probably seen reCaptcha (CAPTCHA stands for  Completely Automated Public Turing test to tell Computers and Humans Apart). It asks you to prove you are not a robot by reading and typing some obfuscated letters or to identify things (like street signs) in photos. When you reply, you are actually training AI! Apparently not many are aware that they are helping AI to identify words in old books and helping to train Google's AI visual recognition.
  • Yes, supervised learning AI needs lots of data
  • Note: there is another type of AI called "Reinforcement Learning" where AI does not need lots of data in the same way as Supervised Learning. In layman's terms, it is given a reward/goal that has a value and then it keeps attempting to improve itself to attain that reward. This video shows an example of Google Deepmind teaching how to run and jump. You can find more to "play" with AI on Google's page of experiments.
  • Here is a good article explaining some of the main terms of AI and another one explaining AI, Machine Learning and Deep Learning

AI Bias and Interpretability
  • There is a whole research area of AI to determine how it arrives at its decisions. It's called Interpretability or Explainable AI (XAI). Because bias has been discovered in AI, there is a lot of research going into XAI with many methods being used. This article gives a good overview of AI and Interpretability with good examples.
  • Henry mentioned during the panel discussion that he worked on a large scale enterprise AI project where two data centers were trained with the same data but arrived at different answers. The audience audibly expressed their surprise at this.
  • Mike Moran asked why should this surprise us; people sometimes express different opinions from each other, so why not AI?
Do I need to Learn about AI?
The panel had various views on this. In my opinion, it will depend on the types of projects you want to manage.
  • If you are working on a project that is going to implement AI, then, of course, you would need to understand the tasks involved. As a PM, you can work with your project team to take the objective and decompose it as a WBS.
  • If you want to make yourself marketable, probably many software projects will use AI in the near future, so that's another reason to learn more about AI.
  • For all PMs, your knowledge of AI will need to be about how it can be applied to the work that you do. AI has a high ROI where lots of repetitive work is being done. For project managers, you can easily list the tasks you do that are repetitive. My experience has been that I have had repetitive tasks in communications (reports and notifications), scheduling, risk, and issue management.

A few sentences on AI and Project Management

(This section was contributed by Sandipan Gagopadhyay)

1. AI such as machine learning is good at synthesizing patterns from vast amounts of data. If an organization is using a project management solution (such as MS Project Server) over a few years consistently, then information can be harvested to derive a number of insights:
  • Supervised Learning examples - Prediction of accuracy in time and cost estimation in projects
  • Prediction of issues and their impact based on challenges and risks based on NLP and ML-based review of unstructured RAID logs
  • Unsupervised Learning examples - Classification of high variability when projects involve specific clusters of technology, teams, locations that in turn can shed light on underlying policy or process issues
  • Here's a Forbes article titled "AI in Project Management" https://www.forbes.com/sites/cognitiveworld/2019/07/30/ai-in-project-management/#74c18a31b4a0

2. AI/ML projects have unique requirements for Project Managers in that the techniques used require modifications to the SDLC, whether in waterfall or Agile

  • AI/ML projects involve precursors including the collection and determination of quality of data that will drive machine learning
  • Key steps to identifying the feasibility of a machine learning solution such as feature identification and selection require an iterative with a fundamentally undetermined outcome and timeframe - This requires prototyping early on before any commitments are made to business or in the delivery of value, financial or otherwise
  • The quality assurance and validation functions require an engaged and a deeper supervisory role in the overall approach to AI rather than testing results at the end
  • Please see FDA's proposed regulatory framework for AI/ML in Medical Devices - https://www.fda.gov/media/122535/download
  • Artificial Intelligence and Machine Learning in Software as a Medical Device - https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-software-medical-device
(Sandipan, thanks for that helpful input!)

The Software Development Life Cycle (SDLC) 

There was some discussion among the panelists about how the SDLC is different for AI projects. My experience has been mostly with supervised learning AI. In supervised learning, humans have to tag (or label) lots of data in order to provide training for the AI.
So, I've found that for chatbots (and lots of AI projects), there needs to be a lot of work both in the beginning and after go-live. In the beginning, people need to be involved to identify and analyze data to train the AI. I've also found that many business people believe that AI learns on its own or will teach itself. So, they think that once the AI is live, there will be little effort required. With Supervised Learning, humans have to be involved in analyzing what is happening in user interactions and training/re-training the AI. This needs to happen continually to increase the effectiveness of the AI. So, there is a large amount of effort required even after go-live.

AI and Project Management

The Project Management Institute (PMI) has made available a 4-page report on Project Management and AI. They provide results of a survey of 551 Project Managers. The paper is titled "AI Innovators: Cracking the Code on Project Performance" (2019). A review of the report was done by TechRepublic titled "6 AI technologies changing project management" mentioning 6 AI technologies that are impacting organizations. My takeaways from the article are:
  • Some companies are actively involved in AI and seeing benefits already while others are lagging behind. 
  • The article looks at the principles being used by "AI Innovators" Vs. "AI Laggards".  
  • It mentions that Accenture research found that visionary organizations apply five key principles identified by the acronym MELDS: Mindset, Experimentation, Leadership, Data, and Skills.


With my experience in Natural Language Processing and chatbots (and my knowledge of Machine Learning, Deep Learning, RPA, and predictive analytics), I can see how these AI technologies and others can be used to help Project Managers. I've reviewed several articles (see the reference section below) and agree with many of them mentioning ways that AI can help PM. However, I've not yet found a whole lot of specific information showing how these AI technologies are being applied to help Project Managers.
I do believe that AI will grow in its use, so PMs do need to learn about how to lead projects using it, how it works, what it's benefits are, and how it can be applied to help them.

I would like to hear your comments and questions about this article and your thoughts on AI and project management. Please comment below. Thanks for your time to read this.

All the Best! Have a great day!
Henry Will

Reference: Additional Information on Artificial Intelligence

No comments: