Well,  believe it or not, not every company that uses deep learning is an AI company. 

Key characteristics of an AI company:

  • Strategic in acquiring data, 
  • Relies on a unified data warehouse (having the data at one place is crucial for the team to spot the patterns!)
  • Good at spotting automation opportunities 
  • AI roles in the team as Machine Learning Engineer, Data Scientist, etc.

Ok, now how to become a good AI company?

  1. Do pilot projects. I was a bit surprised by the advice to choose a project that can be successful rather than valuable. But it makes a lot of sense. The point is to get familiar with AI and build confidence within the team and among investors that you can execute AI projects. So look for a project that you can complete. It might not be the most valuable, but it should still be meaningful. Prove to yourself that you can work on AI projects. Understand how AI may or may not apply to your business. Then go with something more valuable.
  2. Build an in-house AI team. You can start with an outsourced team working on the pilot projects. But av
  3. Provide broad AI training.
  4. Develop an AI strategy.
  5. Develop internal and external communications
  1. Define AI strategy. The reason AI strategy is the third step is that you must understand well what is AI, how does it fit your company and what are your tech capabilities (which you will learn doing the pilot project) Then you are ready to define a realistic AI strategy.

(the steps are from AI transformation playbook: 

Division between Data Science and Machine Learning (AI): 

Data Science is good atMachine Learning is good at
Optimizing process by finding optimal conitions and such leading to errors/defects (manufacturing line)Detecting defects (manufacturing)
Optimizing Recruiting funnelAutomated resume screening
A/B testing (Marketing)Recommendation engine
Crop analytics – maintaining optimal conditions on the farmDetecting weeds

How to select an AI Project

Andrew recommends working in a cross functional team comprising AI experts and business domain experts. Ai project must be a) Tech feasible (AI experts can do tech diligence) and valuable (PM will advise here). AI and IT experts will go through tech diligence estimating:

  • Can AI system meet desired performance
  • How much data is needed
  • timelines for engineering work
Tech Diligence (AI and IT experts)Business Diligence
Can AI system meet desired performanceHow much data is neededtimelines for engineering workHow AI system will lower cost, how much will it saveIncrease revenueCan we launch new product/business

Consider build vs buy

DS are more commonly in-house due to the specificity of the collected data. ML projects are either in-house or outsourced. Advice: Buil what will be specialized to you and avoid building industry standards/.

Ethical diligence – make humanity and society a better off

Food for brainstorming

automate tasks not jobs. Not that certain jobs will be extinguished completely, just their scope would change as AI will overtake some of the tasks..

Define main drivers and main painpoints in your business

It was very encouraging to hear that you can start small, with a small dataset. It is better t start than to wait to acquire a lot of data to start up.

How to work with an AI team?

  1. Specify an acceptance criteria! It is key! For example, you may want to achieve 95% accuracy , but how will you measure accuracy? You need a test set with data to test whether the algorithm works with the desired performance.

how AI team treats data: Training set comprising of pictures with lables where AI learns the A to B mapping. Test set is a different set of images with labels aiming to evaluate AI accuracy/

Case study

How to build a Smart speaker?

  1. Detect the trigger word “Hello device” audio to trigger word – yes or no
  2. Speech recognition audio to text transcrpt
  3. Intent recognition map audio (tell me a joke) to one of the X number of commands the smart speaker understands (tells weather, time, etc.)
  4. Execute the joke