3 Machine Learning Tricks To Improve Team Performance

 Hype about performance data has been high for some time now, whether collecting it, presenting it, or interpreting it. Managers at all levels have been on a quest to quantify employee performance, and therefore codify how to achieve it. But after decades of data gathering, a good formula for creating the perfect employee hasn’t yet emerged.

“I think we're really at the beginning of what I see as a correction of the overreliance on metrics and easily measured quantities that has really characterized the last century of business,” says best-selling author Brian Christian. “We’re at the beginning of putting human values and preferences at least on an equal footing with that.”

Particularly in the past decade, organizations have seen an overall shift toward more human values such as empathy and inclusion. But how does an organization track and measure something like empathy? In other words, how can a system accurately capture and incentivize intangible data? Those questions are what inspired Christian to write his latest book, The Alignment Problem: Machine Learning and Human Values.

Developed from interviews with computer scientists attempting to align an AI system with human values, these are three machine learning tricks Christian learned which can help you track and improve team performance.

1. Cultures are limited by the information put into them.

If you’ve ever prompted an AI to write a story or create an original image, you know firsthand how limited the results can be. An artificial intelligence is limited by the information that is put into it, whether that’s a collection of short stories or the entirety of Google image search results.

“The question of AI bias, AI fairness, has been one of the fastest growing areas within the field of AI in the last five years,” Christian explains. “If you say [to an AI], ‘Show me pictures of nurses’, they're going to be overwhelmingly female. If you say, ‘Show me pictures of CEOs’, they're going to be overwhelmingly western white men wearing suits. There are these stereotypes.”

The same is true of your organizational and team cultures. If you want a culture of high performance on your team, you must make sure the right information is in place to make that culture possible:

- Whose values are being implemented in this culture? Are they being implemented, or merely given lip-service?

- Are values actively or passively absorbed by this culture? How can values be more actively communicated, implemented, and rewarded?

2. Experience allows you to succeed at a higher level.

Imagine you’ve walked into a casino, knowing that some of the slot machines pay out more than others. It’s up to you to figure out which ones have the higher wins. This is the basic premise of what’s called the multi-armed bandit problem, a question that aims to find an algorithm for the most optimal solution.

“It was considered for most of the 20th century to be an unsolvable problem,” Christian says. “But shockingly, it turns out that there are solutions to this, and my favorite are a group of what are called regret minimization algorithms.”

This is how Christian explains it:

In the multi-arm bandit problem, you're really trading off between the payout itself and the information that you gain from trying different things. Your uncertainty about a given machine can be represented by a huge error bar. The more experience you gain, the more you can kind of tighten those error bars down and say, ‘No, I'm pretty sure this pays out 60% of the time. I've played it enough to know.’

The more experience you have, the more accurately you can plan a path to success, assessing your risk and failing smaller while succeeding at a higher level. Encouraging your teams to adopt this kind of thinking may result in more calculated risk-taking, which ultimately results in better outcomes.

3. Incentivize the best behavior you see, not the perfect ideal.

Sometimes as a manager you may feel you’ve communicated a certain performance expectation ad nauseam with no change. What gives? The solution here may lie in how rewards are structured.

“There's a famous computer science paper called Policy in Variance Under Reward Transformation. Basically, what that means is there are ways we can modify the reward structure to make it more easily learnable, but without changing the behavior that we get at the end,” Christian says.

For example, say your goal is to have a computer put a golf ball into a hole. You could withhold any reward until the ball is in the hole, or, you could “transform that into a reward that has to do with proximity,” according to Christian.

But remember, all systems are flawed.

Christian advises leaders to keep in mind that the point here is to adapt machine learning to human systems, not the other way around.

“If all you care about while hiring is giving yourself the best chance of getting the very best candidate, then there is this beautiful rule called The 37% Rule,” Christian explains. “Interview the first 37% of the candidates, send them all home. As you interview everyone else, the very first person who's better than the best person from that initial 37%, hire them immediately on the spot.”

That might be the optimal way to solve the problem of how to hire the best candidate. But, there’ s a pretty major flaw, Christian says.

“The 37% Rule only succeeds 37% of the time,” he laughs. “You can follow the optimal strategy and you'll still fail 63% of the time. A lot of the time when you do fail, you will end up with no candidate at all. That might not be a realistic tradeoff for a hiring manager.”

“If you get points proportional to how close the ball is to the hole, that might produce a system that's a lot more easily learned.”

Looking to hire skilled software developers? Contact TP&P Technology - Leading Software Outsourcing Company in Vietnam Today

Article resource: https://www.forbes.com/sites/forbesbusinesscouncil/2022/12/02/software-development-time-estimation-how-long-should-it-take-to-develop-a-product/?sh=4a2370e76cee