3 Machine Learning Tricks To Improve Team Performance

The hype surrounding performance data is high, whether collected, presented, or interpreted. Managers of all levels are on a mission to measure employee performance and codify the methods to get there. After decades of data collection, there has yet to be a single formula to create the perfect employee.

Brian Christian, best-selling author, and the speaker, says that we are at the beginning of what he sees as a correction to the overreliance upon metrics and easily measured quantities which has characterised the past century of business. “We are beginning to place human values and preferences on at least an equal footing with this.”

3 Machine Learning Tricks To Improve Team Performance
3 Machine Learning Tricks To Improve Team Performance

Organizations have shifted towards more human values like empathy and inclusion in the last decade. How can an organization measure empathy? So, how does a system capture intangible reward data accurately? Christian’s latest book, the Alignment Problem: Machine Learning, Human Values, was inspired by these questions.

These are three machine-learning tricks Christian discovered from interviews with computer scientists trying to align an AI system with human value. They can be used to track and improve your team’s performance.

1. Information is what limits cultures.

You’ve probably asked an AI to create a story or an original image. This is because it can produce minimal results. Artificial intelligence is limited by what information it receives, whether it’s a collection or all of the Google image search results.

Christian says that AI bias or fairness has been one of the fastest-growing areas in the field of AI over the past five years. If you ask an AI, “Show me pictures about nurses,” they will be overwhelmingly male. If you ask for pictures of CEOs, the majority of them will be white western men in suits. These stereotypes exist.”

This is also true for your team and organizational cultures. You must ensure that you have the correct information to create a high-performance culture in your group.

  • Which values are being implemented within this culture? Do they exist, or are they merely spoken of?
  • Is this culture actively or passively taking in values? What can be done to make deals more actively communicated, applied, and rewarded

2. You can succeed at a higher level with experience

Imagine walking into a casino and realizing that specific slot machines pay more than others. You have to decide which ones offer the highest wins. This is the basic idea of the multi-armed bandit question, which aims to find the best algorithm for solving the problem.

Christian says that it was considered an intractable problem for the majority of the 20th century. Surprisingly, there are solutions. My favorite is a set of algorithms called regret minimization algorithms.

Christian explains it this way:

Multi-arm bandit is a situation where you are trading between the payout and the information you get from trying other things. A vast error bar can represent your uncertainty about a machine. You can tighten these error bars and say “No, I’m sure this pays out 60% of the time.” The more experience you have, the better. I know .’.

Your experience will help you plan your path to success. You can assess your risk better and make mistakes less often while succeeding at a higher level. This type of thinking can lead to more calculated risk-taking and better outcomes.

3. Encourage the best behavior, not the ideal.

You may sometimes feel like you have communicated a performance expectation ad nauseam to your manager with no changes. What’s the solution? What structure is the best way to reward?

“Policy in Variance under Reward Transformation is a well-known computer science paper. Christian explains that there are many ways to modify the reward structure to make learning easier but keep the behavior the same at the end.

Let’s say your goal is for a computer to put a golf club into a hole. According to Christian, you can either withhold any reward until it is in a spot or “transform” that reward into one that has to do something with proximity.

“If the distance the ball is from the hole is proportional to your points, you might be able to learn a much more simple system.”

However, remember that all systems can be hampered.

Christian suggests that leaders keep in mind that machine learning is applied to human systems, not vice versa.

Christian says, “If you only care about hiring the best candidates, then this rule is called The 37% Rule.” Interview the top 37% of candidates and send them home. Interview everyone, and the person you find the most qualified to replace the 37% of the candidates, then hire the person immediately.

This might be the ideal solution to the problem of finding the suitable candidate to hire. Christian points out that there is a significant flaw to this approach.

He laughs, “The 37% Rule only works 37% of the time.” You can still follow the optimal strategy, but you will fail 63% of the time. You will often end up with no candidate if you fail. There might be a better tradeoff for a hiring manager.

- Advertisement -
Avatar photo
Samatha Vale
Samatha a senior writer for HC's entertainment team. She is an entreprenuer, mother and an excellent writer. She's also an avid reader, music enthusiast and all around inquisitive person - which is just a nice way of saying she's nosy.

Latest articles

Related articles