- Think before you act. Don’t say sorry afterward, but think about the potential impact and risks before you start building solutions.
- Take back control of the algorithm. Inspirational lecture by Mark Surman of the Mozilla Foundation. Making trustworthy and transparent AI is hard, but people in AI need to try harder.
- While you are busy with innovating, also be busy with figuring out how to do it better. It should be a habit.
- Topics such as ethics, trust, privacy, transparency and bias are still popular. At the same time there is still much work to be done. It's good that people talk about it, but there are also too many examples where these topics are not applied at all. Doing something cool is not enough, you should do it in a responsible way.
- There is a helicopter flying on Mars. Without AI this was not possible! The helicopter works together with a self-driving robot that collects materials. With AI the robot could do in a month what without AI would have taken a year.
- If you work in data science, it’s important to know how to search. The internet is a big ocean of knowledge. You can learn to swim, but only experience can teach you where to swim.
- At LEGO they listen to machines to predict if something is wrong. You can only come up with these kinds of methods by talking to the people who do the work today. Be curious and be in touch with other people.
- Communication is becoming more important. Within data science teams, people sometimes forget to communicate with other departments. Now that data science is a key technology, it’s important that we communicate a lot more. Also with people who have less knowledge about how it works.
- Peter Norvig: "Language is the best technology we have ever invented."
- Know your principles and implement ethics from the start. Don’t do it once a year.
- As an organization you have to take flexibility into account. There are many new initiatives and tools in the AI domain. Also, the AI ACT is coming. Protect yourself from getting locked in so you don’t have the flexibility to move to future winners.
- Investors love KPIs, but do we focus on the right KPIs? A good example is dating platforms that focus on how long you swipe on their platform. It’s easy to measure, but shouldn’t the KPI be how many successful dates their users have?
- With federated learning you don't have to share your data to collaborate. There are more and more possibilities to collaborate on sensitive data in a secure way. Earlier this year we also wrote how you can use BranchKey and AskAnna for this.
- Many examples of machines becoming more intelligent. This allows us to automate more. It even has the potential to end modern slavery.
- New models become available fast. But in the latest NLP models you already see that they are trained with less growth in features. Are we already satisfied? Or is it temporally and are we distracted with the new cool kid: autogenerated videos (yes, videos...not images)
- Nowadays “young” people just start and accept they will not have the knowledge of all the underlying tools, logic and math. Accept that you don’t know everything...that also means you have to be able to trust each other. The past has shown that interests can be different. To move forward, we need to find a solution for this.
- MLOps is too much about pushing data science work into a DevOps kind of framework. Shouldn’t it be about going from Research to Production?