Thoughts on AI/Machine learning for Project planning and scheduling?

- Is it going to help our roles and the organisations and industries we work in? - not in the next 5 years.

- What are some of the issues likely to be faced in implementing these solutions? Lawyers will impede it, and schedulers - for example - the scheduling community will not accept "Progress Over-ride" over "Retained Logic" but don't you think "Progress Over-ride" is a baby step in AI/Machine Learning?  It is.

- How may it change the role of planners/schedulers ? will save some typing.

Thanks John,

My biggest concern is that we're throwing data anlaysis solutions at schedules, treating schedules as data. But they aren't.

The number of activities, durations and relationshisp don't determine the success of a project of not. Schedules are representative of something, of a process or methodology of delivering an end product. The schedule attempts to model this, but a schedule never truly captures that represetnation in a way that machines can analyse this.

Consider a simple example, lets say you have 1000 schedules that contain activities labelled as "Excavate Trench". You might think, that this can be used to provide some analysis on duratins, method of linking to other tasks etc. But whats missing is information such as - what trneching method is being applied? one crew?two crews? what is the soil condition? Is it being performed by machine or by hand? in wet weather? how much time for task includes setup?

The schedule doesn't capture this level of data, so simply having a thousand versions of this isn't particularly helpful. And even then, what will it predict? that the duration to excavate a tench should be 20days? not the 10days I've planned? Withouth providing a reason for why, how can I justify making it 20days?  "because the machine learning algorithm told me so?" isn't going to be a good answer.

It is going to help our roles and the organisations and industries we work in?

AI could help building schedules for simliar type of projects but they must be made project specific. I.E. building a project in the winter in Florida is very different from building the SAME project in Alask or a very northern state.. I do see this helping in prefabrication, and mass production make use look at things in a more modular way.

- What are some of the issues likely to be faced in implementing these solutions?

As I mention specific project locations which includes differnt types federal, state and local regulations. different approval processes and different production rates.  If something happens who do you go to for recourse ? The company who wrote the AI ? Who do you sue ?

- How may it change the role of planners/schedulers ?

AI could it could do the initial heavy lifting if given a set of paramaters to create the schedule.

In Spider Project it was always possible to create and to use any kinds of norms like material requirements per volume unit for different activity types, resource crews and resource productivity for different types of assignments, etc. Besides, it is possible to create to use in all projects the library of typical fragments.

So for Santosh example Spider Project users create types of excavation depending on methods and soil.

For excavation of different types resource productivity is defined in resource productivity database (we call it reference-book).

Typical crews are stored in another reference-book.

Entering volume of excavation in physical units (m3) and activity type we can automatically take from the reference book the crew that must be assigned and assigned resource productivity. Duration is also calculated automatically.

But adjusting resource productivity based on actual data cannot be automated without manual analysis of all factors that can influence it like weather conditions, people health, risk events, etc. I am sceptical about automated adjustment of planning parameters.

I think if you gave me a project to schedule and AI tool and 5 days I would produce a much better schedule than without an AI tool (after I have had the learning curve.)

But will happen is that because you have the AI tool it would be cut down to 2 days to produce the schedule, and would be worse.  

- And of course lots of things would be blamed on the "black box", just like with voting machines - but I think it could provide better answers as long as they are simple enough to explain.

"because the computer says so" is an answer that will never stick.