- We operate in an environment of fear (data encryption, GDPR, ethical implications, etc.)
- Diversity is very personal and individual
- We are locked into characteristics we cannot change that influence how we make decisions. Our personal decisions are inherently biased and myopic. It is important that we recognize this in ourselves. The way to overcome this individual bias is through diversity of teams
- Diversity can drive value from allowing us to adhere to regulations, to improving performance, and creating social responsibility resulting in better outcomes.
- Diversity can drive better analytics
- Data analytics is a hot topic across geographies and industries, but it is hard to find enough people to staff analytics positions. Diversity helps us expand the labor pool.
- Statistics show that diversity leads to better performance and supports innovation.
- Diversity is critical in ensuring the analytics are right. Analytics entails interpreting the surrounding world/data, so the algorithms and programs need to be correspondingly diverse. Diversity brings multitude of perspectives that improve decisions and outcomes (age, ethnicity, knowledge, seniority, gender, race, etc.)
- AI already has an impact on our lives - machines interpret data and get to direct people (Autonomous car, drones, buy through Alexa, etc.). Ceding broad decisions authority to machines requires us to make sure we train machines to make high-quality decisions. In order to do that, we need to overcome out personal bias and use diverse teams to develop AI. Bias can lead AI projects to erroneous outcomes (85% of projects by 2022).
- Bias can occur on teams, in the data, and in the output.
- We need to treat algorithms the way we treat our kids - build the foundation for good decisions.
- Diversity is a leadership, not management issue. Need to take personal risks and become better on soft skills
- Very sensitive topic, which means that it involves emotions - have to understand the role of emotions. Have better grasp of the soft skills.
- Diversity poses many challenges: certain dimensions cannot be readily ascertain (but can be objectively identified)
- Recruiting
- Write job description - not the description of the type of person you want to hire
- Need to rethink qualifications. Focus on the job duties and start relying on conversations.
- Keep it human - diverse people need to have support they need
- Need to change our practices - how we seek feedback (start from most senior person or the junior level employee).
- Diverse people have diverse lifestyles and we need to accommodate them.
- Do not question your culture, but its rigidity.
- Machine learning is observational - need to behave right to teach them to make the right decisions. (teach algorithms as we would teach our children)
Questions
- This is still math - need good attributes or learning is not possible?
- Part of the experimentation is dropping in other attributes - example Amazon adding weather - change the model, experiment using expert knowledge - proposed candidate attributes
- How to use machine learning to disrupt vs. perpetuate what is happening now?
- The human thinks outside of box not the machine - go for maverick ideas, run the experiments, the failure is low cost
- Create experimentation protocol that would support new ideas/disruption. The cost of experiments has gone down, so we can afford to run more experiments and fail sometimes.
- Increase of the income gap decision- who decide what is right?
- Number analysis may be based on the basic economic theory - however we do have to raise objections. There are different ways how the truth may be quantified (example of the bus driver in India vs. U.S.)
