Positive machine learning

How to integrate machine learning successfully into the business.

.

The steep hill of machine learning

Interest in machine learning has exploded in the last years. Companies want that want to use it and employees are changing careers.

Companies and individuals are chasing the latest developments in machine learning; studying LLM’s and piloting with agents. Although these are great technologies to experiment with, jumping right into the technology is often not the best start.

Here are the top 3 limitations of this approach:

1. You limit your scope to a particular technology; you leave big opportunities on the table
2. You get carried away with technology; often there exists alternative paths of least resistance
3. Your organization cannot adapt to new ways of working; because you disconnected from the organization, the project will never go live

Procedure trap

How to integrate machine learning successfully into the business? About 10 years ago, the organization I worked for had an answer to this questions: process.

We need processes so that we can replicate successful machine learning projects.

At the time fake Scrum was the go to methodology for software engineering. Since machine learning involved some programming, every project was assigned a couple of process experts with fancy titles. The problem was that, although projects rarely succeeded, no-one was questioning the process. The answer was more managers, more meetings and taking fewer risks.

The procedure revisited

There is nothing wrong with procedures; on the contrary. For machine learning, the right approach was just not commonly known back then.

If we can choose or invent a methodology for conducting machine learning projects in organization, what characteristics should it have?

These are the aspects that I think are most important:

• Explore business opportunities and challenges
• Find solutions with and without machine learning
• Let people of multiple disciplines work together
• Break problems and solutions down in small projects
• Allow projects to fail
• Set ambitious long term goals
• Settle for very modest short term goals
• Prepare the organization for change

Ambition level

The ambition level is the most interesting part of the requirement list. On the long term we need to be bold, while on the short term we need to stay modest.

This may sounds contradicting, but compare it to this: One day you deicide to sign up for a marathon and you decide to run it faster than all your friends.
To make that happen, you join all the trainings of the national athletics track team. What happens? I bet you will be injured after one week and force to quick.

This should drive down the point of modest goals on the short term. But how do modest goals in machine learning look like?
Here are some examples:

1. Let one or two colleges use an excel sheet to improve decision making
2. Use pen and paper to work out a better sales approach
3. Track the parts that you manufacture in a notebook and conduct a manual analysis

This is a fast track to machine learning success. Why? By taking modest steps like these, you are likely to skip one or two machine learning or technology projects that would take months while adding nothing but a disconnect to the organization.

Positive psychology

Here we arrive at positive psychology.

Positive psychology arose in the late 90’s as the science of human well-being and flourishing.
Instead of focusing on what can go wrong with the human psyche, positive psychology uses what goes right to improve well-being and performance. Typically this is achieved with small behavioural changes.

Main stream psychology tended to focus on what can go wrong. For an organization that starts with machine learning, the tendencies are similar:

• Focus on machine learning methodology, because you don’t understand advanced math, best practices, etc
• Build heavy technology because you believe that is your weakness and you believe you need to fix it
• Invest heavily in project management because you want to avoid any failure at all cost
• Obsess over the latest technology because you are afraid to be laughed at if you settle for less
• Set up intense pilot projects, staffed with externals, to force quick success and avoid failure

Behind those tendencies is fear, a disconnect with the organization and a lack of good examples to follow. Positive psychology helps you to steer away form fear, negative thoughts, self doubt and other negative emotions. It let’s you use your strengths to build a better future.

With positive psychology you

• Practice happiness with small wins (this fit’s with modest goals)
• Focus on what you can do (guards you against the technology pitfall)
• Invite people from different disciplines because you focus on a positive experience, and not on their (totally irrelevant) weaknesses like math skills or missing credentials.
  
• Are not afraid to fail, because you know that you are on the right track

This is why I think positive psychology is such a good fit for machine learning projects.

Positive machine learning

Positive machine learning applies the principles of positive psychology to machine learning projects in organizations.

The principles are:

• Formulate bold goals with a diverse set of people to create shared positive frame for the long run
• Start with modest goals, and actively seek positive team experiences to reinforce positivity
• Be protected against small failures due by the positive long term frame and positive short term experiences