Matt Maccaux, Global Field CTO at HPE, spoke on an interesting topic. There are plenty of ways to make mistakes with artificial intelligence.
Analytics Anti-Patterns (the worst practices)
Anti-patterns are what organizations do to solve problems when it “seems” right but really isn’t correct.
Volume 1 (presented earlier)
- Silos of information create duplication of data
- Build data lakes but prevent users from ever accessing them
- Organizations put all the info in one common repository and let users what they want with it
- Deploying big data environments in commodity hardware and try to scale.
- with GPU’s need to think about things differently
- Pet projects. if you build it they will come. Doesn’t work for technology projects
Volume 2 of the patterns
AI / ML / DL hype
- the 2018 Gartner hype cycle is rife with AI / ML technologies
- They fail because execs don’t understand and can’t get past the hype
- You need to understand the core needs
Rush to the Cloud
- The cloud is appealing for AI / ML needs
- When Matt asks his customer how long would a new data scientist be productive. The answer: 6-9 months
- The cloud has lots of promise but…
- You have to thing through scale, security, data trust
- You have to think through budget, costs, communication, etc.
- Don’t rush. Be thoughtful
Story: one client gave their data scientist access to AWS. Over the weekend, they forgot to turn off a job. By the end of the weekend, the bill came to $35,000
Replicating the Monolith
When going to the cloud, you can just redeploy all your normal data scientist tools. Most data scientists love to use the tools they used in schools. But will those open source tools scale? Does it make sense to use those tools? You end up with lots of siloed deployments of these tools.
Data Scientist as Lone Explorer
Corporations give these relatively new and young data scientists to answer open ended problems. Sometimes they find issues that can’t be solved. Sometimes they find projects where it’s not worth fixing. Sometimes they can predict something but it’s hard to operationalize. If someone has a prediction, you have to be able to act upon it. All of this leads to churn with data science.
When your data scientists strike gold then you need to assemble a team to operationalize it.
Quote: 86% of organizations struggle with operationalizing insight
Give these data scientists a common and enterprise platform. This runs from back end to data governance, to the tools used to create and run models.