When pondering what to write about looking forward to 2018 I had a shortlist of three hot topics:
- Artificial Intelligence (AI)
So, reading about what might be coming up in 2018 I would like to share three interesting posts from consultancies about AI and data analytics in 2018. The first is Top 10 AI technology trends for 2018 from pwc - An exciting year and prospects to look forward to! To me Explainable AI is the next goal to provide a competitive advantage. Using AI to predict outcomes is great, but if the model can't explain why it has made a prediction it is difficult to know how to improve it, or the limits of how you can trust it. This is a pretty big difference to how humans learning and provide feedback.
The next article was a great overview of current landscape for analytics, AI, and automation from McKinsey, data analysis skills and under-utilised assets are definitely an area of weakness that 2018 could start improving. With all the potential that could be unlocked, it seems a no-brainer.
It’s difficult to sum up Machine learning evolution in a simple infographic, but think pwc did a good job - just bear in mind reality is a lot more nuanced. Also AI predictions, on technology used, more then 5 years out are usually wrong! It's hard to argue with the general trend of combining techniques. This is something that I have observed since I started studying AI. Where neural nets were abandoned for a decade in favour of other techniques. Now with much larger corporate focus with giants like Apple, Google, and Facebook all having substantial investment and looking for a competitive edge, I'd hope that this wouldn't be the case.
But what about the data scientist building a lot of this infrastructure? One key concern I have picked out from looking around on twitter to gauge opinion, is below
Data scientists: what's your team's approach to tracking the quality of models in production?— Caitlin Hudon👩🏼💻 (@beeonaposy) December 28, 2017
(How do you know if a model is decaying? How do you quality-check the data going into a model? Who builds and tracks these things?)
worth bearing in mind if you are introducing predictive analytics. Your model reflects the data that you had at the time and choices made during the training process. How are you going to check the performance and effectiveness going forward? Especially if you don't have an Explainable AI.
The second is how important the foundation is, Sean Taylor highlights that a lot of the value of data scientists work is how it is used downstream
This has pretty interesting implications:— Sean J. Taylor (@seanjtaylor) December 28, 2017
- As a manager, you should be maximizing how much people are learning from each other, not just how much they can accomplish.
- As a data scientist, you should see at least half the value of your work in how much people learn downstream.
So think about the benefits that a good data analytic culture can bring. Not all of the value will come from where this work is done. If it's done right then the learning will be occurring in other roles.
In the past year I have been looking at some of the business aspects of AI. So this year my resolution has been to brush up on the more technical aspects. Expect to see my write about the DataCamp Data Science courses and others that I take in 2018.
- Time Series Forecasting with Recurrent Neural Networks
- Creating a Data Visualization GraphQL Server with a Loosely Coupled Schema
- DataCamp courses
- Data Can Lie–Here’s A Guide To Calling Out B.S.
- The Most Important Skill In The Age Of Artificial Intelligence (AI)
- The Step-By-Step PM Guide to Building Machine Learning Based Products
- Roundup of 2018 AI Predictions
- The Case Against Deep-Learning Hype