AI/ML Product Management by Uber Sr PM, Kai Wang
What is Al (and ML & DL)
Al
A system which is able to automatically perform tasks that usually require human intelligence.
ML
A subset of Al techniques which enables machine to learn from historical data and improve with experience.
DL
A subset of ML models which utilizes artificial neural networks to solve advanced Al problems such as CV and NLP.
What is an Al product
A product which leverages Al techniques, often ML, to perform a limited set of tasks within a certain context to improve human productivity.
Al products (PMs)
-Applied ML (~70%)
Use ML to create, improve, and enhance products to solve real world problems.
- Al Services (~20%)
ML -based software services purposely built to address a certain ML use case.
-ML Tooling (~10%)
ML tools/platforms/frameworks to facilitate ML development and improve efficiency.
A traditional version of a product is considered successful if it meets all of the defined functional requirements.
A machine-made product must meet all functional, performance, and model accuracy requirements, depending on the data used and the settings.
To develop a software product, developers usually use 1-2 or several frameworks. To develop a machine learning product, developers need to try out many different models, libraries, algorithms, and frameworks. Need data engineers to collect and process data. They have to build and train the model you know. We need engineers to implement the model in production and software engineers to build the entire infrastructure. The team needs to be big, so collaboration and communication is very important.
If it is software products it is usually possible to explain the results, but machine learning products are usually difficult to explain because the rules learned by the model are difficult to interpret or simply not available.
Al PM Skill Set
ML Technical Depth for Al PMs
Key takeaways
Build your ML foundations, and be technical enough
Remain user-centric
Have a fallback/mitigation plan
Kai Wang, Uber Sr PM